Claude Masterclass FULL COURSE 4 HOUR (Build & Automate Anything) https://t.co/zEoKZX8cW7

Transcript

Welcome to the most comprehensive Cloud Code Masterclass

0:03

where I take you from a real beginner

0:04

in Cloud Code

0:05

to someone who's actually able to build agentic

0:07

workflows and full systems for yourself or for other businesses

0:11

Now

0:11

I use Cloud Code pretty much every single

0:13

day to run my six-figure AI business

0:15

and I'm also teaching it to over 10,000

0:17

people in my community. And so

0:19

I've put together this Masterclass in the exact

0:21

order that I wish I had when I

0:23

had started out. And as always

0:24

no fluff, no jumping around

0:26

just a clear practical roadmap that takes you

0:28

from zero to building real systems

0:30

Now

0:30

I've divided this Masterclass into five different modules

0:33

So here's exactly what you're getting

0:34

Module one is foundations. This is where we install Cloud Code

0:37

go through how it actually works

0:38

permissions, setting up your cloud.md

0:40

your cloud directory

0:41

so Cloud Code understands your projects before you

0:43

even type a word. Module two is customization

0:46

This is where we build custom skills

0:47

slash commands, hooks that automate actions for you

0:51

and MCP

0:51

which connects Cloud Code to every external tool

0:54

in your stack. Module three is prompt engineering

0:56

And here I'm going to be breaking down

0:58

how to talk to Cloud Code so you

0:59

get better Outputs every single time

1:01

Module four is agentic workflows

1:03

This is where we talk about sub-agents

1:04

agent teams

1:05

and eight hacks I've discovered after building hundreds

1:08

of workflows and at the end deployment

1:10

So everything runs 24-7 without you

1:12

And module five is the build

1:14

No more theory. You're going to watch me

1:15

build real projects from scratch

1:17

using everything that we covered

1:19

So you can see exactly how it all

1:20

comes together. Now I'm also going to be

1:21

leaving timestamps right below in this video

1:24

so you can stop the video and then

1:25

you can come back to it

1:26

and start wherever you left off

1:27

And as always

1:28

if you want all the resources and templates of this video

1:31

then you can check them out in the

1:32

second link down below

1:33

In my free private school community

1:35

you can go to the classroom section and

1:36

you can go to the templates vault

1:37

Now with that said

1:38

let's dive in

1:43

All right, here we have Cloud Code

1:45

First thing is pricing. How much does this actually cost?

1:48

Now

1:49

bear in mind that we

1:50

are not going to pay Cloud Code

1:52

We're going to pay Cloud

1:53

and through playing Cloud

1:54

we get access to Cloud Code

1:56

We get access to Cloud Codework

1:57

and we get access to Cloud itself

1:59

okay?

1:59

Now, Cloud has three different tiers

2:01

It has a free tier

2:02

and by the way

2:02

if you have the free tier

2:03

you do not have access to Cloud Code

2:05

So Cloud Code

2:06

you have to pay for it

2:07

Then we have the pro tier

2:08

which is good for everyday productivity

2:10

which is 17 bucks a month

2:11

and it gives you access to 45 messages

2:13

for five hours. That is the limit on

2:15

usage that you can use Cloud Code

2:17

And this is the best tier that I

2:18

can give beginners starting out

2:19

unless you're creating some really complex applications

2:22

or you have a lot of automations running

2:24

in your business. And then we have the

2:25

max plan. So let me put myself here

2:28

And the max plan gives you 5x

2:30

of the usage of the pro plan

2:32

which would be about 225 messages for five hours

2:35

and you will pay around 100 bucks a

2:36

month. And now I personally have the max plan

2:39

so I'm personally paying $100 per month

2:42

On Cloud Code

2:42

just because I use it every single day

2:44

for my business and I automate a lot

2:45

of stuff that I have day to day

2:47

Now you also have the option to pay 20x

2:50

right? So instead of 5x

2:51

it will be 20x of what the pro plan gives you

2:53

and you will also be paying double the money

2:55

which means that from 100 bucks a month

2:57

you'll pay 200 bucks a month

2:58

So as a beginner

2:59

I definitely recommend sticking to the cheapest plan

3:01

which is a pro plan for now

3:03

And if you see yourself

3:04

kind of exceeding the limit or exceeding the

3:06

token cost or usage

3:08

then you might want to upgrade to the

3:09

max plan. All right

3:10

so the next thing is hosting

3:11

What does that look like?

3:12

How do we set it up?

3:13

And which applications do we use to actually

3:15

use Cloud Code? So first thing is using

3:17

VS Code. VS Code is what we call

3:19

an

3:19

IDE. Very fancy way

3:21

of saying a place

3:22

where you can look at all your folders

3:24

that you have with the Cloud Code is using

3:26

and you can also install Cloud Code

3:28

and you also have access to the AI

3:30

of VS Code itself

3:31

and it's completely free to

3:33

to download, right? Now similarly, we have Cursor

3:35

So Cursor is a very popular one for coders

3:38

just because Cursor itself has their own AI

3:41

Their own AI is very

3:42

very good when it comes to coding and

3:43

doing all the things that

3:45

like building applications and so on

3:46

So if you want the power of cursor plus cloud code

3:48

meaning if you're a bit more on the

3:50

techy side and you're a bit more of a developer

3:52

then I recommend using cursor

3:53

The only limitation of cursor is the limit

3:56

So it's going to get a bit expensive

3:57

especially when you have to pay for cursor's AI

4:00

but also for the cloud AI

4:01

And I can tell you from experience

4:03

that most times

4:04

we just use cloud code inside these platforms

4:07

We never use the AI in these platforms

4:09

at all. But I usually go

4:10

VS Code because I don't use the AI VS Code anyways

4:13

so to me it doesn't really matter

4:14

Then we have anti-gravity

4:15

which is the exact same as VS Code

4:17

Anti-gravity is one of the popular tools that came out recently

4:20

which is part of Google's suite

4:22

And anti-gravity is the exact same as VS Code

4:24

and it also has its own AI that

4:26

we can use. And I honestly see no

4:27

difference between anti-gravity and VS Code

4:29

unless you want to use

4:31

the Gemini models inside of anti-gravity to build

4:33

your own interfaces and make them look pretty

4:36

right? So either one works fine

4:37

Then we have the terminal

4:38

which is where we

4:40

are going to be using cloud code

4:41

Machine inside of our computer without having to

4:43

go to Antigavity or VS Code

4:44

The only limitation with the terminal

4:46

is that it doesn't look too pretty

4:48

It doesn't look very visual

4:49

It's sort of like codey and it looks

4:50

a bit complex when you're starting out

4:52

So as a beginner

4:52

I do not recommend this

4:53

And then finally, we have Cloud

4:55

So Cloud Desktop also has the ability to

4:57

use Cloud Code

4:58

The only limitation here is that

5:00

you will not be able to see your

5:01

files and it won't be as structured

5:05

as Antigavity or VS Code

5:07

Now for today's video

5:08

we're going to focus on VS Code and

5:09

Antigavity

5:10

because most of the videos in this course

5:12

are either using VS Code

5:13

or either using Antigavity. But they're very

5:15

very similar

5:15

so I'll show you exactly how to

5:17

download

5:18

Cloud Code in each one of these platforms

5:19

So let's start with VS Code

5:21

All you have to do is go to

5:22

code.visualstudio.com and you will get to this page right here

5:25

the open source AI code editor

5:27

You can press download from macOS if you're macOS

5:30

or you can press download here

5:32

and you'll be able to go on this

5:33

page and you can choose

5:34

your device right here. I'm just going to

5:35

press Mac here

5:37

As you can see

5:37

now it's downloading the application itself

5:39

We have this little pop-up come up

5:41

Says happy coding. We can drag this across

5:44

Now it's going to download the actual application

5:46

in our computer. Now once you have it opened up

5:48

you will get on a page that looks

5:49

like this. So this is Visual Studio Code

5:51

VS Code in the name

5:52

You'll have different buttons here

5:53

and you'll have the main interface right here

5:55

So this is going to be the main

5:56

place

5:57

where we're going to be using Cloud Code

5:58

and actually building different things

5:59

whether it's automations or applications

6:01

Now in order to download Cloud Code

6:03

all we have to do is go down to the extensions

6:06

and I'll walk you through what each one means here

6:07

so don't worry. Extensions

6:09

you want to look for Cloud Code

6:11

and then you'll see the one from Anthropic

6:14

Make sure it's the one from Anthropic and

6:16

not this one here

6:17

or not this one's here

6:18

This is the verified one

6:19

So once you press the extension

6:20

you will get on a page like this

6:22

You can see Cloud Code for VS Code

6:23

and you can simply press install

6:25

At the time of this recording

6:26

it has 10,263,533 people

6:29

which is crazy, and it's the version 2.1.107

6:32

You know, it was published 10 months ago

6:34

So very, very new

6:36

Now I can press this button right here

6:37

which is for us to be able to open Cloud Code

6:39

and when we get on this page

6:41

I can simply say hello

6:43

Now

6:44

bear in mind that if you don't see

6:45

this page right here to say hello and so on

6:47

it's because it's going to ask you to log in

6:49

Now, as you saw from me

6:50

I can say hello

6:51

how are you? Because I already logged in to Cloud

6:53

inside here. But if you haven't logged in

6:55

to Cloud Code yet

6:56

then you most likely see this page that looks like this

6:58

which says that Cloud Code can be used

7:00

with your subscription or used with your API usage

7:03

right?

7:04

So you can either pay it through the

7:05

Cloud subscription or you can use the API usage

7:07

Now, bear in mind that if you use the API usage

7:10

which is the Anthropic console

7:11

you will end up paying a lot more

7:13

in the long run just because

7:14

the usage is going to increase by a lot

7:17

especially when you use Cloud Code

7:18

and you won't even realize it

7:19

That's the thing

7:20

So you can just press cloud.ai subscription

7:22

You can press open

7:24

It'll take you to this page here

7:25

You can press authorize

7:26

It says build something great

7:27

That's all good. Now I can go back

7:29

and you'll be able to see that now

7:30

I can actually chat to Cloud Code itself

7:33

I can press X

7:34

and now we have the full interface

7:35

So this right here will be the page

7:37

where we actually use Cloud Code

7:38

and we can run up different

7:39

conversations with Cloud Code. So on the left-hand side

7:42

we have folders

7:43

Now bear in mind that

7:44

anything that you build in Cloud Code

7:45

and I mean anything

7:46

will be stored inside a folder

7:48

So when I'm building a website

7:49

when I'm building an application

7:51

so where is the HTML stored?

7:53

Where is the document stored?

7:54

Where are all the images stored

7:56

right?

7:56

They're all stored inside a folder inside your

7:58

computer. And this is great for us because

8:00

you always have a place to store your exact files

8:03

right? And so on the left-hand side

8:05

we can press open folder

8:06

Let's say I add a new folder on the desktop

8:08

I can press new folder

8:10

I can name it example

8:12

and I can press create

8:14

And I can press open

8:15

You can press yes

8:15

I trust the authors. Now you won't be

8:17

able to see Cloud Code anymore

8:18

So to see it again

8:19

you can press this button right here

8:20

Cloud Code, and you can press X again

8:22

And now you're back in this interface

8:24

So now we have the folder set up

8:26

We have everything here. We can start adding

8:27

files

8:28

So file

8:30

We can start adding folders

8:31

So folder

8:32

And that's how you start to add

8:34

what we call the cloud directory

8:36

So the directory of cloud

8:37

So where are all the files stored?

8:38

I can press delete

8:40

And we're back to zero

8:41

Okay

8:41

so everything that we build

8:42

for this specific project that we're working on

8:44

will be stored under the example folder right

8:46

here. Now to show you how that looks

8:48

like on your computer

8:49

if I add a file here and I say file.md

8:52

say hello, okay?

8:54

And then I press command S

8:56

or you can go to file save

8:59

Inside this folder called example

9:00

we have the file.md

9:02

And so now I can go to file

9:03

I can close the folder

9:05

and I can go back

9:06

I can open the folder again

9:07

go to desktop, I can open the folder again

9:09

press open

9:10

and now we're back exactly where we started

9:12

And so this way you can have different

9:13

folders for different projects

9:14

and you can always come back to ClockCode

9:16

and start from zero. Now there are a

9:17

lot of different things in the interface itself

9:19

so let's go through them step by step

9:21

So the first thing is the conversation history

9:23

So if you want to

9:24

switch between conversations the same way that you

9:26

would do instead of Claude

9:27

you can go here to the clock itself

9:29

and you can see all the past conversations

9:31

that you've had. Or you can go to the web

9:33

you have it right here

9:34

And so now I can go to

9:34

let's say, write a LinkedIn post

9:36

which is a conversation that I had before

9:38

and we can always review the conversation that

9:39

we had in case we missed anything

9:41

or in case we want to keep going

9:42

with the conversation itself. Then we have this right here

9:44

which is just the same thing as just pressing this button

9:47

right? It spins up a new Claude instance

9:49

which means a new conversation

9:50

And then down here is where everything's at

9:52

So on the bottom side

9:54

we can see plus

9:54

In case you want to add folders or

9:56

contacts or browse the web

9:57

you can simply But here

9:59

we typically could do this if you want

10:00

to add more context to Cloud Code

10:01

Then we have the filter actions

10:03

So

10:03

there are a lot of different things that

10:04

you can do inside of Cloud Code

10:06

One of them is using slash commands

10:08

So you can put slash something

10:10

So slash

10:10

there's a lot of different things that you

10:11

can do. We can put slash clear conversation

10:14

We can put slash compact

10:15

We can do slash everything

10:17

And we'll go through exactly what a lot

10:18

of these slash commands are later in the video

10:20

so don't worry. And then one of the

10:22

most important things here is actually the permission

10:24

mode. So if I press here to edit automatically

10:27

this is the permission mode

10:28

that we're going to be using Cloud Code in

10:30

right? Now the permission mode might look different in six months

10:33

and it looked different six months ago

10:35

right? But essentially what it is

10:37

it's the way that you're telling Cloud Code to act

10:39

right? So if I put ask before edits

10:42

then anything that it does

10:43

it will ask for approval before making each

10:45

edit. If I put edit automatically

10:46

and now Cloud will edit your text

10:48

or the whole file automatically

10:50

If I put here plan mode as a permission

10:52

then what it will do

10:52

is that it will create a plan

10:54

for the actual thing. Then we have bypass permissions

10:56

which is where Cloud will not ask for

10:58

Before running the potential dangerous commands

11:01

right? And then right here

11:01

we also have the new feature

11:03

which is called effort. So if you want

11:04

a very high effort for the task itself

11:07

then you can simply tell Cloud Code that

11:08

you want higher effort. If it's lower

11:10

then you can put low

11:11

Now, I typically have this on effort medium

11:13

just because I want to keep it balanced

11:14

Now, the default permission here is actually edit automatically

11:17

and so the main ones are really ask before edits

11:19

plan mode, and bypass permissions

11:21

So let me show you an example

11:22

So let's say I tell Cloud Code to

11:24

make a new file inside our folder or project

11:27

and name it LinkedIn post

11:29

and then you can add a LinkedIn post there about life

11:32

with a happy tone

11:33

targeting people who are business owners

11:35

in the US. It's a very strange task

11:37

but essentially it has to edit our folder

11:39

it has to edit something

11:40

and it has to include

11:41

the actual text itself

11:43

So from edit automatically

11:45

I can actually go to plan mode

11:47

because I want it to plan before it does anything

11:49

and here I can just press go

11:50

So now what happens is that Cloud Code

11:51

is given a task

11:52

it will then think about what it's doing

11:55

it will create a plan of action

11:56

As you can see here

11:57

it says, I'll create a quick plan

11:58

For this, and then we can execute it

12:00

And then the plan is something that we

12:01

can review right here. Context

12:03

steps, verification

12:04

So the plan is to create a LinkedIn

12:06

post file. The user wants a new file

12:07

in their project folder containing a LinkedIn post about life

12:10

written in a happy tone

12:11

and targeting US business owners

12:12

It's a very strange request

12:14

but we can press yes

12:15

auto-accept. And now what happens

12:17

is that it creates a file

12:19

on our folder right here called LinkedIn post

12:21

that I can open up

12:22

and here we have the LinkedIn post

12:24

Now obviously this is just an example to

12:25

show you how plan mode works

12:27

right? Now let's say I wanted to use

12:29

bypass permissions

12:30

Bear in mind

12:31

this is not something that you see as default

12:33

so you will not have access to this

12:34

as soon as you get to VS Code

12:36

What you have to do here is go

12:37

to the settings here

12:38

and you can look for allow dangerously

12:42

There we go

12:43

You can turn this on right here

12:45

which means that you're now allowing bypass permissions mode

12:47

recommended only for sandboxes with no internet access

12:50

So that makes no sense

12:51

but just make sure to turn this on

12:52

And all this mode does

12:54

is that it will not ask you for

12:55

approval for anything that it does

12:57

That's why it's called dangerously skip permissions

13:00

because it's dangerous if you don't know what

13:01

you're doing. It's dangerous if you give it

13:03

access to different softwares and files

13:06

and you're not there to actually see what it does

13:08

right? So let's say I wanted to use bypass permissions

13:11

and let's say I give it the exact

13:12

same prompt right here

13:13

and let's say I said Instagram post

13:15

right?

13:17

What you'll see now is that it will

13:18

not give us a plan

13:19

It will just straight away make the file

13:21

itself without us having to do anything

13:23

As you can see

13:23

it made the file here called Instagram post

13:25

life as a business owner that we can use straight away

13:28

right? So on the one hand

13:29

this made a plan

13:29

it asks us for approval

13:31

On the other hand

13:31

it didn't ask us for anything

13:33

and it made it itself

13:34

Now

13:34

the way that I see it is that

13:35

I usually go to plan mode whenever I

13:36

want to create a plan

13:38

for a specific project that I'm doing

13:40

And then once plan mode is done

13:42

if it's something that's a bit more complex

13:44

that needs my input

13:45

I put it to ask before edits

13:46

because it will ask me before it does

13:48

anything. But if it's just a very easy task

13:50

then we can put bypass permissions

13:52

because I don't want to be there

13:53

approving everything all the time

13:54

right? Just understand what it is you're building

13:56

and what Cloud Code has access to

13:58

Now, like we mentioned

13:59

the other way is using

14:00

a software called Google Anti-Gravity

14:02

Google Anti-Gravity is known for making websites

14:04

for making nice interfaces. You might have seen it on YouTube

14:07

right? But it's also a way

14:08

for us to use Cloud Code

14:10

And so I can just go here to

14:11

antigravity.google. You can press download

14:14

and right here, you'll be able to see your own

14:15

computer

14:16

Can press download for Apple Silicon

14:18

You'll then have this page right here

14:20

similar to the VS Code

14:21

We can drag this across

14:22

Now it's downloading Anti-Gravity to our applications

14:25

And once you get to Anti-Gravity

14:26

you will get on a page that looks

14:27

like this. So very similar to the one

14:29

that we had at VS Code

14:30

On the right hand side

14:31

you have the IDE's AI

14:33

so Anti-Gravity's AI, which is the agent here

14:35

which is similar to the one that we

14:36

had in VS Code. And on the left hand side

14:38

you have the folders

14:39

so the same thing as VS Code right

14:41

here. And then we have the extensions

14:42

We can go to Cloud Code

14:44

You can simply install. If you haven't installed it

14:46

I already did. And once you have it installed

14:48

you can press this button right here

14:50

and you can start working on it

14:51

And the same way that we're using the

14:52

example folder in VS Code

14:54

we can go here

14:56

Desktop example

14:57

and we can open it up

14:59

inside of Antigravity. We open it up

15:00

we can press yes

15:01

I trust the authors

15:02

and you'll be able to use

15:03

the exact same folder that we're using in VS Code

15:06

we'll use it in Antigravity as well

15:08

So that's why I say they're pretty much

15:09

the same thing. And one more thing when

15:10

it comes to setting it up

15:11

you can also use your terminal

15:12

So if I go to terminal

15:14

new terminal here, you can simply press cloud

15:17

and you can say yes

15:18

I trust this folder

15:19

and you'll be able to use cloud in

15:20

terminal as well. So if you say hello

15:23

You can see that it says

15:23

hello

15:24

how are you? Now here we can see

15:25

the model that we're using

15:26

how many context tokens we have

15:28

what plan you're on

15:29

the email that we're using

15:31

and then the organization

15:32

and then the actual folder example as well

15:34

Now as a complete beginner

15:35

I don't recommend using the terminal for now

15:37

Just use the actual interface here

15:39

It'll be much easier

15:40

especially if you're just starting out

15:42

with the platform itself

15:43

because you don't want to overwhelm yourself too much

15:45

All right

15:45

so now you have Cloud Code installed inside

15:47

of AntiGravity and inside of VS Code

15:49

If you're just using Cloud Code

15:50

I

15:50

honestly wouldn't see a benefit of using one over the other

15:53

so just choose one and stick to it

15:55

and keep it very simple for yourself

15:56

But if you're someone who's building something a

15:58

bit more sophisticated and you want AntiGravity's

16:01

Gemini model to help you with this

16:02

then I recommend just choosing AntiGravity

16:04

because you have the upper hand itself that

16:06

you can use later on

16:07

All right

16:07

so last thing to cover here is built-in

16:09

slash commands

16:10

So we mentioned this before that we have

16:12

the little box right here

16:13

which gives us the ability to do actions

16:16

without us having to give it a prompt

16:17

So here if I put slash

16:19

I can see that I can call

16:20

any different action if I wanted to

16:22

And you can see We have context

16:23

we have model, we have customize

16:26

and we have slash commands

16:27

So these slash commands are things

16:29

they're actions

16:30

that we can do automatically just by calling

16:32

them specifically. As you can see

16:34

we have batch, cloud API, compact, context, cost, debug, extra usage

16:39

head dump

16:39

and we have a lot more here

16:41

that we can use specifically

16:42

I'm going to go through a few of

16:43

them

16:43

just so you get the idea of how

16:45

it works

16:45

So let's say I had a conversation with Cloud Code

16:47

and I said, hello, and I said

16:49

how are you?

16:50

Now there are cases where the conversation is very

16:53

very long

16:53

and you want to start from scratch

16:55

without

16:55

having to go to a new conversation here

16:57

So what you can do here is you

16:58

can press slash clear conversation

17:01

So slash clear conversation

17:02

you can go here

17:03

and now you start from zero again

17:04

So now we have a new conversation

17:06

and the next slash command is called context

17:08

So if I put slash context

17:10

And I press this one here

17:11

What Cloud Code will give me is

17:13

a

17:13

breakdown of all the contacts and tokens that

17:15

I'm using inside my account

17:17

So we have a million

17:18

of tokens that we're able to use

17:20

So far we use 17,900 tokens

17:23

which is about 2%. And this is just by saying

17:26

hello, how are you, right?

17:27

So think of every conversation as a bucket

17:30

right? The more words you add to the conversation

17:32

the more water goes inside the bucket

17:34

And it comes to a point

17:35

where it can't take any more water

17:37

and so you have to go to a new conversation

17:39

right?

17:39

And this is what the percentage is

17:41

How much water is inside the actual bucket?

17:43

How many tokens are we using for the

17:44

conversation itself? Because the token is just words

17:47

And also it tells us the

17:48

category of usage by type

17:50

So we have the system prompt

17:52

which is a prompt that we can't really see here

17:54

but it's taking up 6.2 thousand tokens

17:57

which is about 0.6%

17:58

And then we have other things like system tools

18:00

we have skills, we have messages, free space

18:03

and buffer as well

18:04

right? It's fine if you don't understand these terms right now

18:07

because we're going to cover them

18:08

later down in the course

18:09

Then we also Have MCP tools

18:11

So I downloaded MCPs inside my clock code

18:14

which is one of the modules that we'll

18:15

go through later. And each MCP

18:17

takes about X amount of tokens

18:19

So 170 tokens

18:20

is used by the Gmail authenticate

18:22

Authenticating the Google calendar takes 178

18:25

And so the more MCPs you add

18:26

the more tokens. And then we have skills

18:28

The skills are essentially instructions that clock code

18:31

uses to do a specific task

18:33

And each skill requires different tokens

18:35

right? And so here you can see

18:37

a breakdown of all the tokens that you're using

18:40

all the usage. Now

18:40

apart from clear conversation and context

18:44

we also have ones like compact

18:47

which is where you want to compact the

18:49

conversation

18:50

so you can keep going in the same

18:51

kind of context without having to go to

18:53

a new conversation

18:54

We do that because every conversation has

18:56

X amount of usage. So let's say right

18:58

now we have a conversation that takes

19:00

90% of its usage. So if you add a slash compact

19:04

it will take that 90%

19:05

It will break it down

19:06

It will summarize the conversation

19:08

so that

19:08

we can still have maybe another Sixty or seventy percent left

19:11

right? So we can keep working under the

19:13

same thread without having to have a new

19:15

conversation with clock code. Then we have cost

19:17

as well to see how much it's costing you

19:19

debug to debug, and then we have others as well

19:21

which we will cover later down the line

19:26

All right, so by this point

19:27

you understand how Cloud Code works

19:28

and how permissions are the things that keep

19:30

you in control. But right now

19:32

every single time that you start a new

19:33

session with Cloud Code

19:35

it starts from zero

19:35

because it doesn't remember anything about your project

19:38

Which is why in the next lesson

19:39

I'm going to be breaking down arguably one

19:41

of the most important files in Cloud Code

19:43

which is the cloud.md file

19:45

And I'm also going to break down

19:47

the cloud directory

19:48

So every single time that you start a new session

19:50

Cloud Code understands your project completely

19:53

without you having to type a single word

19:55

Let's dive in. All right

19:56

so what is the cloud.md file?

19:58

I'm sure you've probably heard of it

19:59

you've seen it somewhere

20:00

but essentially what it is

20:01

it's the product instruction manual for Cloud Code

20:03

So essentially, Cloud Code, whenever we're working with a workspace

20:07

it needs to have some sort of memory

20:08

And the memory is where all the files are

20:10

what the product workspace actually is

20:12

what is our objective

20:13

who we are, and all that stuff

20:15

right?

20:15

And so that's exactly what the cloud.md file actually is

20:18

Now, the cloud.md file loads automatically

20:21

which means that when we work with a new workspace

20:23

it loads automatically in a way that

20:25

Before it starts doing anything

20:26

it always looks at the cloud.md file to

20:28

get an understanding of the whole layout of

20:30

the workspace before it dives in

20:32

and does any single task

20:33

The same way that you would have an

20:34

employee coming into a company

20:36

you give them memory

20:37

and every single day that he comes back

20:38

to the company to work

20:40

he already has the memory pre-trained

20:41

So in this case

20:42

whenever we have a new conversation with Cloud Code

20:45

we are saying, hey, before you do anything

20:48

just look at the cloud.md file

20:50

understand our workspace

20:51

and then

20:52

execute the task that I want you to do

20:54

Now

20:54

the actual cloud.md file

20:56

is divided into four different kind of steps or pillars

20:59

right? The first one is product description

21:02

So this is what this app does

21:03

It's not necessarily just an app

21:05

It's just saying, this is what this workspace is

21:07

right? For example

21:08

I have a YouTube workspace where I manage

21:10

all my YouTube stuff

21:11

So this workspace is specifically designed for YouTube ideation

21:15

content creation, XYZ, right?

21:18

Then we have the tech stack

21:19

This is looking at what tools and softwares

21:20

we use

21:21

Then we have the rules and guardrails

21:23

which is what cloud Must and must not do

21:26

and rules and guardrails can be anywhere from

21:28

the text they should use

21:29

the colors they should use

21:30

the fonts

21:30

the step-by-step process they should go through to then create something

21:33

right? And these are all the different rules

21:35

It's kind of general

21:36

but you can have like any rules that

21:37

you can think of

21:38

you can put it there

21:39

And then finally, we have folder structure

21:41

where everything lives. Now

21:41

this is one of the most important ones

21:43

because as you will see when we make the actual file

21:45

there is a

21:46

structure of folders that we add inside our

21:48

workspace

21:49

The reason that is

21:50

is because we want to make things organized

21:52

and Cloud knows exactly where to store all

21:54

the different files. So we'll have a file for the scripts

21:57

we have a file for the output

21:59

which is the things that it gives us

22:01

We have a file for the ideation

22:02

Maybe we have a file for all the passwords and keys

22:05

API keys that we use

22:06

right? And so in having things organized

22:08

it's much easier for Cloud Code to do things

22:11

because it's not scattered everywhere

22:13

That's what it is. Now

22:13

what Cloud Code gains by having this file?

22:15

Well, the first thing is that it understands your project better

22:17

and every single time that you speak to Cloud Code

22:19

you don't have to repeat all the things

22:21

right? It already knows

22:22

The second thing is that it follows your

22:23

conventions. This can Be naming conventions

22:26

it can be patterns

22:26

it can be a structure that you use

22:28

Then like I mentioned

22:29

it looks at the rules

22:29

So every single time this happens

22:31

do X, Y, Z, right?

22:33

And it won't break what actually matters to

22:34

you

22:35

And then finally, it's consistent

22:37

That is probably the most important thing

22:38

because it's much better to have something that's

22:40

average but consistent than something that's really

22:43

really high quality but inconsistent

22:45

Because with AI specifically

22:47

it's very good at doing things

22:49

right? Like it's nuts

22:50

it's crazy. But you also want to have strict guardrails

22:54

right? Strict prompts that make it so that

22:56

it's not inconsistent every single time

22:58

It actually gives you the output that you

23:00

want every single time

23:01

when you are doing that specific thing

23:03

So there's two different types of ways to

23:04

be considered up. The first way is

23:06

using a three-layer architectural structure

23:09

Sounds complex, it's very easy actually

23:12

And the second way is

23:13

running the slash init. Now the two different

23:15

ways represent two different scenarios

23:17

The first scenario is method one

23:18

right? Where we write the cloudMD first

23:21

and then we tell

23:22

cloud to initialize the project

23:24

And this is only when we have a new project

23:27

when we're starting from zero

23:28

Then the second method is when you already

23:30

have an existing project. So let's say you're

23:31

building apps or maybe a landing page or maybe some agents

23:35

right?

23:36

You haven't done the cloud.md yet

23:38

So now what we do is we take all the folders

23:40

all the files

23:41

cloud will understand your whole workspace

23:43

And from there

23:44

we tell it create a cloud.md file

23:46

So one is creating cloud.md file from scratch

23:49

and then creating the folders

23:50

And the other one is you already have the folders

23:52

now we create the cloud.md

23:53

All right

23:53

method one is writing cloud.md from scratch

23:56

So here we have the cloud code project

23:58

I made a new project called cloud

24:00

And this is where we will make a file right here

24:03

just pressing this button

24:04

and name it cloud.md

24:06

And inside here, and by the way

24:08

the reason why it's md

24:09

is because it's marked down

24:10

Marked down is just a way that we format text

24:12

right? Because this is title one

24:15

this is title two

24:16

two hashtags

24:18

Then we have boldings

24:19

which is hello, right, and so on

24:21

So with that said

24:22

we have the cloud.md file that we made

24:23

Now we have to actually

24:25

write the instructions and then initialize the project

24:28

okay? Now

24:29

one thing I want to show you as

24:30

well is if I go to a new chat

24:32

I'm looking here

24:33

because it does mention make a cloud.md file for instructions

24:37

Cloud will read this every single time

24:38

So even Cloud Code incentivizes you

24:40

it tells you, hey

24:41

make this file because it reads this every single time

24:43

Okay

24:44

so here we have the prompt

24:46

that I'm going to give

24:47

Cloud Code in the cloud.md file

24:49

I'm going to copy this

24:51

I'll bring it back

24:52

I'll go here and then paste it

24:54

So what does this prompt tell ClockCode?

24:57

So you operate within a three-layer architecture

24:59

that separates responsibilities to maximize reliability

25:02

That's fancy names

25:04

LLMs are probabilistic

25:05

while most business logic is deterministic

25:08

and requires consistency. This system solves that problem

25:10

This means that LLMs are inconsistent

25:13

but business logic is consistent

25:15

So we want to solve for that inconsistency

25:17

with this exact file

25:19

right? And then we give it the three

25:20

different layers. So layer one is directives

25:22

We're basically saying

25:24

anything that has to deal with

25:25

what to do goes in the directives folder

25:28

Anything that deals with decisions

25:30

goes and follows this framework right here

25:33

And then everything to do with doing the work

25:35

so the execution, is this folder right here

25:38

and we're also making a

25:39

.env file where we store API tokens

25:42

passwords, and so on, okay?

25:44

And then we have other principles as well

25:46

I can save it first of all

25:47

so it doesn't run

25:49

And I can say

25:50

can you initialize this Direct based on the cloud.md file

25:54

cloud.md. I'm going to press go

25:57

I'm going to make this bypass permissions

25:59

It's reading cloud.md

26:00

Now it's looking at all the different things

26:01

that are inside. And as you can see

26:03

it's starting to make the folders

26:04

It's making the folders which are.env

26:06

executions, directives,.tmp

26:09

which is what we call the

26:10

cloud code directory. The cloud code directory is

26:13

essentially where all the different files are stored

26:15

And the cloud code directory usually looks like this

26:18

right? It's with these lines

26:19

So we have clouds folder

26:20

which is this main folder

26:22

And inside the cloud code folder

26:23

we have cloud.md, we have.env

26:26

which is this one here

26:27

So add your API keys

26:28

OpenAI, Google API, passwords that we're going to use

26:31

execution, we have directives

26:34

And we have dot TMP

26:36

which is empty. So let's say I go

26:37

to a new tab

26:38

and I can write this down

26:39

I can say, build me a landing page for Jam Solutions

26:42

an AI automation agency

26:43

that helps businesses to automate their workflows using AI agents

26:46

which have a hero section

26:47

what we do, testimonial section, and footer

26:49

I'm going to press go

26:51

I'm going to make this plan mode

26:53

so it plans the landing page before it

26:55

does anything. As you can see

26:57

it's looking at the cloud.md file

26:59

It didn't find any brand guidelines

27:00

which is fine. Now it's creating a plan

27:02

All right, it just made the plan

27:03

We have the context

27:04

the tax stack, which is essentially what we're going to use

27:07

the structure, and then we have decision design

27:10

hero section. I can just press yes

27:11

auto accept

27:13

I can make this bypass permissions

27:15

because that means that it doesn't ask me

27:17

to approve things all the time

27:18

And the whole point of showing you this

27:19

landing page is so that you see

27:21

how the different files

27:23

of the landing page are stored

27:25

in the different folders here

27:26

All right, it just made the landing page

27:27

I can open it up

27:31

As expected, it's pretty decent

27:33

to be honest, very simple

27:35

And you can see here that it made

27:36

another folder called frontend

27:37

which has all the different components of the

27:39

landing page. But it also made a

27:41

MD file for the directives

27:42

which is what to do

27:44

right?

27:44

Directive, build landing page, objective, tech stack, page structure, service section, testimonials, footer, file map, and so on, right?

27:51

And so this is showing you that all the different things

27:54

like we saw here

27:55

of what to do

27:56

right? The SOPs, the objectives, the inputs, scripts, outputs, edge cases

28:01

are all going to be stored under the directives

28:04

All right

28:04

so now you got to see exactly how

28:06

we made the folders from scratch

28:08

by adding the cloud.md file first

28:10

But it's also method two

28:11

which is using slash init

28:13

So let's say I have a workspace called YouTube

28:15

and YouTube has a lot of folders

28:16

it has context, it has executions, outputs

28:19

all that stuff. It has.env

28:21

which are all the different passwords and so

28:22

on. But the one thing that is missing

28:24

is the cloud.md file. And so in this case

28:27

we wouldn't use the three-layer architecture

28:29

because The folders are already made

28:32

We already have the folders

28:32

right? Which could have been done manually by

28:34

me or someone else. And so in this case

28:36

to create the cloud.md file

28:38

all I have to do is go

28:39

to slash init

28:42

And now what it's doing is that it's

28:43

going to look at the whole organization

28:45

It's looking at all the different files that

28:46

it has

28:47

And it says

28:48

now I have a good understanding of the

28:49

workspace. Let me create the cloud.md file

28:51

So now what we've done is that we

28:53

have the cloud.md file

28:55

which has the memory

28:56

it has the context of the whole workspace

28:58

So whenever we're making an automation

28:59

whenever we're making something else

29:01

when we start cloud from scratch like this

29:03

it basically has the understanding

29:05

And to show you another example

29:06

if let's say I go back to the

29:07

one that I had before

29:08

let me delete the cloud.md

29:10

Let's say you had this done manually

29:12

I would go here

29:13

I would save this

29:14

first of all

29:16

And as you can see here

29:16

the first thing it's telling me is to

29:18

tell Cloud Code to remember where you've told

29:19

it using Cloud.md

29:21

Well, in this case, I can press slash init again

29:25

bypass permissions

29:26

What it will do

29:26

is that it will look at all the

29:27

different files that are here

29:28

which in this case is to build the

29:30

landing page. As you can see now

29:31

it made the Cloud.md file

29:33

which is the product overview

29:35

which is landing page for JM Solutions

29:36

an AI automation agency

29:38

It provides guidance to Cloud Code

29:39

So the architecture will be actually used to build things

29:42

the front-end structure, the commands

29:44

the design system, and the important parts

29:46

right?

29:46

So that's a snippet of memory that Cloud

29:48

Code will now use whenever it makes something new

29:50

Now

29:50

the last thing I wanted to show you

29:52

is a more advanced version of a Cloud.md file

29:55

because under my YouTube workspace

29:56

I have commands, I have skills, right?

29:59

All these things

30:00

which allow me to do much more with

30:01

Cloud Code

30:02

than you would just by having default features

30:04

And so in this case

30:05

we can see that the Cloud.md file

30:07

same thing here, provides guidance

30:09

what this is, is a Cloud workspace template

30:11

a structure environment for working with Cloud Code

30:13

as an agent assistant across sessions

30:15

Slash prime at each start of the session

30:17

to load essential context. So if you're running from scratch

30:20

all you have to do

30:21

is go here and put slash prime

30:24

And what this will do is that it

30:25

will look at the whole structure

30:26

it will look at the Cloud.md file

30:28

it will look at everything to have an understanding

30:30

the context that it needs

30:32

to be able to tell me who I am

30:34

the workspace structure

30:35

and everything that it needs to know about

30:37

me specifically. We have the workspace structure

30:38

again, this is the Cloud Code directory

30:40

When you hear Cloud Code directly

30:41

that's all this means. So we have Cloud.md

30:43

we have shellalias.md, we have.env here

30:46

we have ncp.json, which is this file right here

30:49

the cloud commands and skills

30:50

So cloud commands and skills

30:52

Then we have references

30:53

execution outputs, references, execution output

30:56

Then we have commands

30:57

so it's telling me all the commands that

30:58

I have. Then we have the YouTube content pipeline

31:00

which are all the different skills

31:02

the thumbnails, analytics and monitoring, skill system, additional skills, critical instructions

31:07

and notes at the end as well

31:08

And that's the exact Cloud.md file that

31:10

I use every single time

31:11

and the one that I always have sitting

31:13

there

31:13

as a way for Cloud Code to understand

31:15

And who I am

31:16

and to have context

31:17

before I ask it anything

31:18

Quick one before we get back to the video

31:20

if you're looking to work with me

31:21

want one to be able to start and

31:23

scale your own AI automation agency

31:25

in a 90-day one-on-one mentorship program

31:27

then check out the first thing down below

31:28

With that being said

31:29

let's get into it

31:34

Hey, congrats on finishing module one

31:36

You've now got the foundations locked in

31:39

You have Cloud Code running

31:40

you understand how it works

31:42

and most importantly

31:43

it knows your project inside and out

31:45

But as always

31:46

you've pretty much tapped into 5%

31:48

of what Cloud Code can actually do

31:49

which is why in the next module

31:51

we're going to be taking Cloud Code

31:52

from a smart assistant all the way

31:55

to building a fully customized system

31:57

that's built around the way that you work

31:59

And it all starts with something that most people overlook

32:02

which is skills and slash commands

32:04

Let's dive in. All right

32:05

so we have Cloud Code's brain

32:07

we have skills, which makes super cloud

32:10

because cloud

32:11

on skills is like a cloud on steroids

32:13

So it can do so much more

32:14

than just the normal interface

32:15

What are skills?

32:16

So skills are instruction manuals that cloud

32:19

reads before doing a task

32:21

So the same way they use a recipe card

32:23

which are instructions to make something

32:25

in this case a chocolate cake

32:26

or it can be a burger

32:27

it can be pizza

32:27

whatever it is

32:28

which is a list of instructions that show

32:29

you how you can do that thing

32:31

Cloud has Cloud Skills

32:32

which are instructions to handle PDFs

32:35

to create websites, to create presentations

32:37

and to do all those things

32:38

So

32:39

that when we have a task that you

32:40

give to Cloud Code

32:41

let's say you say

32:41

let's make a website

32:42

and Cloud, which is the agent

32:44

goes to the skills

32:45

right, and says, okay

32:46

which skill do I have to use?

32:47

In this case, it will use the website skill

32:49

right?

32:50

And it will use the website skill to

32:51

give us a perfect result

32:53

Now, why is this important?

32:54

Because without skills, Cloud guesses

32:56

and the result is inconsistent

32:58

The reason why the result is inconsistent is

33:00

because

33:00

it doesn't have a proper way of doing

33:01

things if it doesn't have skills

33:03

right? Because skills are step-by-step instructions

33:05

And so if it doesn't have step-by-step instructions

33:08

it does its own thing

33:09

however it wants to do it

33:10

in its own way

33:11

And the way that AI does things can change over time

33:13

right? And so what it does now

33:14

might be different than what it does in 30 minutes

33:16

right? Now with skills

33:17

Cloud follows instructions, step-by-step instructions

33:20

And so with that said

33:21

because it has a clear prompt

33:23

that it has to go through

33:24

that it has to look at

33:25

the margin of error will decrease massive

33:27

which means that it will give us a

33:28

much better output. Now that you know exactly

33:29

what Cloud Code skills are on a general level

33:32

and Why they're important

33:33

where do you actually store them?

33:34

So I'm here in my folder called the YouTube workspace

33:36

which is where I manage most of my YouTube stuff

33:38

Strategy, thumbnails, and ideation, titles, hooks, bodies, all that stuff

33:42

No, this is not scripted

33:43

so don't worry

33:44

But essentially, if you see here

33:46

inside this folder

33:48

there are a series of folders

33:49

So doc cloud, brand assets, context, execution, outputs, references, scripts

33:52

By the way

33:53

if this is the first time that you've

33:54

ever been introduced to cloud code

33:55

check out this video up here

33:56

because it will give you the overview

33:58

of what most of these are

33:59

and why they're, you know, that way

34:01

But if you already know that

34:01

and if you already have the fundamentals

34:02

then this will make more sense

34:04

We will have a folder called the doc

34:05

cloud folder. Now the doc cloud folder

34:07

you'll see a dot behind it

34:08

because you will actually not be able to

34:10

see it in your folders

34:11

I know that doesn't make sense yet

34:12

but I'll show you. But essentially

34:13

we have this folder called doc cloud

34:14

and inside the folder of doc cloud

34:16

we have another folder called skills

34:18

So that's where we store skills

34:20

So now, if I go to a new page

34:21

and I name this general

34:23

I can have the folder

34:24

I can press this right here

34:25

and I can put doc cloud

34:27

Cool. And now inside the doc cloud folder

34:29

I can then go here and put new folder

34:31

input skills

34:33

Okay?

34:34

So now when we start adding skills

34:36

we add it to the folder itself

34:37

Now why do we do this?

34:38

We do this because we want to be organized

34:40

right? And so

34:40

Claude knows that every single time

34:43

that it needs to look at a skill

34:44

it always goes to the skill itself

34:46

and it goes to all of them step

34:47

by step. Now if you notice

34:48

the doc cloud folder also has commands

34:50

and so you can also go here

34:53

and you can put new folder

34:55

input commands

34:57

So as you can see now

34:57

we have inside the doc cloud folder

34:59

we have commands and we have skills

35:00

Obviously nothing inside, but we'll add them later

35:03

For the sake of this video

35:03

I'm only going to be talking about mainly skills

35:05

but that's essentially where we store plot code skills

35:07

All right, so right here

35:08

if you've ever seen something that looks like this

35:10

which we call the folder tree instead of plot code

35:13

this is how the folders are stored

35:14

So doc cloud is a folder

35:15

commands is another folder

35:16

skills is another folder

35:18

Inside each folder, there's obviously documents here

35:20

So this contains only two documents

35:22

which

35:22

we'll go through. And skills contains different folders

35:25

And each folder is a skill

35:26

There's a lot of folders and skills that

35:27

I've said so far

35:28

but essentially every skill that we add under

35:30

skills folder is a folder in itself

35:32

because a folder consists of documents

35:33

If that didn't make too much sense

35:35

I hope this will. So here we have

35:36

doc cloud skills. Instead of skills

35:37

let's say we have the LinkedIn writer

35:39

So let's say we want to make a

35:39

skill that writes LinkedIn posts

35:41

That's it

35:42

Inside each skill, we have three different things

35:45

okay?

35:46

We first have a skill.md

35:47

which is a process

35:48

step-by-step instructions that we tell Cloud Code

35:50

like, hey, here's how this skill is done

35:52

And then we have two more folders

35:54

Okay, so each skill, which is a folder

35:56

has two more folders. The first folder is called references

35:58

Now, just so you know

35:59

these two

36:00

they're not always going to be there

36:02

Like sometimes you need reference docs

36:04

or examples or templates. Sometimes you don't

36:06

right? So this is like if or or

36:07

you need them. But in the case that

36:09

you do need them

36:09

we'll have references, and this includes documents of examples

36:12

of templates, or reference docs

36:13

anything that you need there

36:14

On this side, we have scripts

36:16

Scripts is a Python code

36:17

which is again optional in case you need them

36:19

But essentially, if you're running some sort of automation

36:21

the code represents the automation itself

36:23

Okay, so this is the instructions

36:24

hey, here's how you do the task

36:25

References is like

36:26

here's supporting documents that you would need to

36:27

make the task happen. And the scripts will

36:29

be like the Python codes

36:30

which you can think of like the different N8N nodes

36:33

the different steps of the automation

36:34

which are code

36:35

right? That's exactly how it's stored

36:36

Then going back to my previous example here

36:38

of my YouTube workspace

36:39

which is where I store everything for my

36:41

kind of YouTube projects. If I go to doc cloud

36:42

this is the folder that contains

36:44

these skills folders as well

36:46

It also contains this right here

36:47

agent memory, agents and commands

36:49

which I'll walk you through later on

36:50

But skills is the one that we have

36:51

to worry about. And if I press skills

36:53

you can see that it has these many folders

36:54

which are all skills themselves

36:56

We have MD generator. We have comment analysis report

36:58

we have competitor analysis

36:59

escalator driver. We have a set of different

37:01

skills that do different things

37:03

and if I press on one of them

37:04

let's say comment analysis report

37:06

I'll be able to see the folder of references

37:09

the folder of scripts

37:10

and then the skill.md

37:11

which is a document that outlines everything

37:13

okay? And so skill.md

37:15

references and scripts are all folders inside the

37:17

main folder of LinkedIn writer

37:19

inside the doc cloud skills folder

37:21

So they're all folders

37:22

right? One folder here

37:23

one folder here, one folder here, one folder here

37:25

one folder here, and then we have the skill.md

37:27

Okay

37:27

but that's the

37:28

structure of different folders inside Cloud Code

37:30

And it is pretty much the same kind

37:31

of structure as a competitor analysis

37:33

but go here

37:33

I also have references and scripts

37:35

This also has the JSON file

37:37

which is just something that we use to store Google tokens

37:39

but some of them only have the skill.md

37:42

And so

37:42

that's why I say that this is mandatory

37:45

You always need to have the skill.md

37:47

The references, the scripts, you don't always need to have

37:49

because this right here

37:50

for example, we don't need any reference files

37:52

right? We just need a skill.md

37:53

which has a bunch of instructions as to

37:55

how to actually make the escalator

37:57

So here I have five different

37:58

what we call marketplaces

38:00

So the same way that you go to

38:01

Amazon to get a product

38:03

we go to this marketplace

38:04

to get a skill

38:05

The only difference is that these are free

38:07

which means you're going to have to pay

38:08

for the skill itself. At least as far as I know

38:10

I'm sure there's skills out there that you

38:11

have to pay for

38:12

but as far as these ones

38:13

you don't have to pay for them

38:14

Now the sites here

38:14

I'm going to leave the links all down

38:16

below so you can check them out

38:17

but I have AITMPL

38:18

which is a cloud code templates

38:19

which is the one that I personally use

38:21

pretty much most of the time

38:22

Then we have this right here

38:23

which is also a pretty decent one

38:24

Then we have this

38:26

a bunch of skills as well

38:28

Then we have this here

38:30

we have this as well

38:31

that contains different skills

38:33

If I go here

38:34

there we go

38:35

Now

38:36

it might seem a bit intimidating at first

38:37

when you look at this

38:39

just because it does look very

38:40

very techy. I mean

38:41

all of this didn't really make much sense

38:42

to me when I started

38:43

but it's actually very

38:44

very simple once you break this down

38:45

So let's say I go to the Cloud Code templates

38:47

okay? And I can mention

38:48

skills are just pre-built instructions

38:50

It's a prompt

38:51

right? It does a specific thing

38:52

So let's say I wanted to create a website

38:55

Okay, if I wanted to create a website

38:56

I would definitely use the front-end design

38:58

Why? It's because the front-end design creates distinctive

39:00

production-grade front-end interfaces with high design quality

39:03

Use a skill when the user asks to build web components

39:05

pages, artifacts, posters, or applications, dashboards, React components, HTML, CSS layouts

39:10

or when styling and beautifying any web UI

39:12

It generates creative

39:13

polished code and UI design that avoids generic AI aesthetics

39:17

okay? So this is great when you're building a website

39:19

when you're building a UI

39:20

which is a front-end

39:21

the thing that you see inside a software

39:23

and so we use this here

39:24

Now, the skill has different components

39:25

It has the name

39:26

right here. It shows you the amount of

39:28

downloads that a skill has

39:29

and typically the higher the downloads

39:30

the better the skill is

39:32

And front-end design is one of the most

39:33

used skill there is out there

39:35

Which is great

39:35

We have the install command

39:37

and what this is

39:38

is a command where you can copy this

39:39

and you can simply just paste it into Cloud Code

39:41

and I'll show you how to do that

39:42

and it will automatically install the skill without

39:44

you having to give it any instructions

39:46

okay? Then we have the description of the actual skill

39:49

we have the license

39:50

which is the, just the document to

39:52

I'm pretty sure it's to not use copyright

39:54

like to not copyright the actual skill

39:56

so you don't copy it

39:56

and then put it on your own marketplace

39:58

right? It's, it's going for this person

40:00

And then we have the preview

40:01

which is a preview of the actual skill

40:03

and what it is and what it does

40:04

And then, most importantly, we have the code

40:06

And the code really is a skill.md document

40:09

So when we spoke about this skill.md here

40:11

all it is, it's this

40:13

is this prompt right here

40:14

And so every skill is just a prompt

40:16

right? Don't make it more complex than it is

40:18

because that's all it is

40:19

just a prompt that tells Cloud Code what

40:20

to do

40:21

Now if I go here

40:22

just to show you how it looks like

40:23

on a different website

40:23

content ideas, let's do content, content creation

40:25

Content creation, let's do a random one here

40:27

And here we can see that we have

40:29

the name of the skill

40:30

we have the description

40:31

we have the command

40:32

which is what we talked about before

40:33

and then we have the file

40:35

In comparison to here

40:36

where we can and we are able to

40:38

actually copy the prompt and paste it to

40:40

to a code or do whatever

40:42

I mean, usually you just use a command

40:43

It's the easiest thing to do

40:44

Just paste the command

40:45

and it will do everything for you

40:46

Here you also have the command

40:47

but you have an option to download the documents

40:50

okay? Because this right here

40:51

will be a document for the skill

40:53

Okay, which is a skill.md document

40:55

And that's it. Instead of downloads

40:56

you all have stars

40:57

so 57,000 stars. Tells me that it's

40:59

pretty good and that a lot of people have downloaded it

41:01

which tells me positive things about the skill

41:04

If I go to here

41:05

I can see that I have

41:06

this is file search

41:08

Let's see what this shows us

41:09

One click install, one click copy

41:12

so we can copy this

41:14

File explorer with the skill.md

41:15

there's the full prompt

41:17

So they're all like

41:17

I mean, the websites look a bit different

41:19

but they all have the exact same thing

41:21

right? They have the description

41:22

they have the, the skill itself

41:23

The most important thing is you understanding

41:25

what skill do I need

41:27

to be able to do what I want

41:28

to do?

41:29

And then going into the marketplaces and looking

41:31

at the skill that you need

41:32

to then be able to copy and then

41:33

paste it. I'm going to copy this

41:34

and I'm going to go to

41:35

Cloud Code in a new version here

41:37

and I'm going to paste this here

41:39

Okay, and I'm going to go and run

41:40

Now it's asking me

41:41

if I can do this

41:42

I'm going to press yes

41:43

and actually turn this to bypass permissions

41:46

As you can see now

41:47

it made a folder called doc cloud

41:49

If I go inside

41:49

I can see that it made a folder called skills

41:51

and inside skills it made a folder called front-end design

41:54

and inside front-end design it had this skill.md document

41:57

which is a full prompt

41:58

that we saw here

42:00

The exaxim thing

42:01

And so, like I showed you before

42:03

I made the folder manually

42:05

but you can also have Claude make the

42:06

exaxim folder automatically for you

42:08

Now

42:08

let's say I want to use a different

42:09

skill. So let's say I went here

42:11

and I said, okay, let me use the senior architect

42:14

okay?

42:15

I go here, I copy this

42:16

and then I simply just paste it here

42:18

Actually, let me go to a new window

42:21

And then I can go and press done

42:23

I can go to bypass permissions

42:24

so it doesn't ask me every time to

42:26

approve

42:26

And now you'll be able to see a

42:27

new skill being added here

42:28

There we go. Senior architect

42:30

It's as easy as this

42:31

But senior architect, in comparison to front-end design

42:34

has references, it has scripts

42:36

and it has a skill.md

42:37

So sometimes it needs references

42:38

sometimes it needs scripts

42:39

to be able to fulfill the actual skill

42:41

but sometimes it doesn't

42:42

right? Sometimes you just simply have a prompt

42:43

which is just instructions

42:45

of what it needs to do

42:45

Now bear in mind that the skill.md

42:47

document is just a list of instructions

42:50

and it's standard across all the different LLMs

42:53

And so whether you're using Claude

42:55

Cursor, Codex, Gemini, CLI, or Copilot

42:58

or any other of these 20 different tools

43:01

you will be able to apply this exact

43:02

document to all of them

43:03

And that's why it's a very

43:04

very important document

43:05

and it also makes skills way more valuable

43:07

than people think. I'm going to go here

43:09

and delete this skill right here

43:10

because I'm not going to need it

43:11

But next

43:12

I want to show you how to build

43:13

a website and how Cloud builds a website

43:15

before skills and then after skills

43:17

Okay, so I'm also going to delete this front-end design

43:19

Actually, I'm going to delete the whole thing

43:20

Cloud folder

43:22

I'm going to go here

43:23

and I'm going to say

43:24

can you make a website for my agency

43:26

JM Solutions? It's an AI automation agency that

43:29

helps marketing agencies to build delivery systems and

43:31

helps them to double their client

43:34

delivery in 30 days. I'm not going to

43:35

put too much effort into this

43:36

I want to keep everything equal and just

43:38

show you the output

43:38

right? I'm also going to say

43:40

the color scheme is going to be dark blue

43:42

and yeah

43:42

give me the landing page in a

43:44

link that I can access on my web

43:45

I'm going to go here on plan mode

43:47

because I wanted to

43:48

kind of plan first before it gives me the actual website

43:52

and then we can go bypass permissions

43:54

to see the full thing

43:54

All right, so after a few minutes

43:55

we got the plan here

43:56

which is a plan on the landing page

43:57

dark blue, complex approach, color, sections

44:01

So we press yes and auto accept

44:02

and then we go to bypass permissions

44:04

So I don't have to do this every

44:05

single time

44:07

Bypass permissions, and that's it

44:08

All right, I just made the landing page

44:09

I'm going to copy this

44:11

I'm going to paste it here

44:13

And here we have the landing page itself

44:15

W client delivery in thirty days

44:16

So as you can see

44:17

this is the design style that it made

44:19

okay, with the actual Cal.com link and everything

44:21

okay, which is not bad

44:22

which is not bad

44:23

Now let's say I go back and

44:25

I delete this

44:27

Then

44:30

And I give it the exact same prompt

44:32

but now I'm going to use the skill front-end design

44:35

copy the command, and then paste it here

44:37

and now we're going to download the skill

44:38

and then do everything again and show you

44:39

the end result. Okay

44:40

so I just made the skill.cloud folder skills front-end design

44:44

and we have the skill.md

44:45

which is a prompt. Cool

44:46

Now I can go here

44:47

I'm going to paste the exact same prompt

44:48

The only thing I am going to say is

44:50

can you use a different localhost

44:52

number to give me the link?

44:53

Just because it gave me the link of localhost 8080

44:56

and I want to show you the difference

44:57

between this and that. So I can press this

44:59

I can go to plan mode

45:01

it'll make the plan

45:02

and then we go direct it to build the website

45:03

Okay, it just made a plan

45:04

I can press yes

45:05

auto-accept, and then go to bypass permissions

45:07

All right, now it made us a new link

45:09

so I can copy this as well

45:10

I can go here. You can see this

45:11

is the old website

45:13

I can paste the new link

45:14

and we'll have the new website here

45:18

Which is much nicer

45:20

It has numbers as well

45:20

It has testimonials in case we need any

45:22

And we also have the option to book

45:24

a call as well. So no obligation

45:25

no pushy sales

45:27

just a clear roadmap for your agency

45:28

So as you can see

45:29

with a skill right here

45:30

it has made the website a lot better

45:32

because of the prompt that we gave it

45:33

here. And you can apply this thing to

45:35

pretty much anything that you do with these

45:36

pre-built

45:37

kind of skills that you can see on

45:38

the marketplace. And now that you know exactly

45:39

how to go to a marketplace and use someone else's skills

45:42

in the next section

45:43

we're going to look at how you can

45:44

build your own skills step by step

45:46

I'm going to be building my own

45:47

to show you the full process

45:48

so eventually you can do the exact same

45:50

Now to be able for us to make the skill

45:52

you need to understand the foundation of how

45:54

a skill is actually made

45:55

right? Now you know

45:56

that a skill here has different folders

45:59

So for example, comment analysis report has this folder

46:02

this folder, and this document, okay?

46:03

Now this document is by far the most

46:05

important document because it has the instructions

46:07

Because not all the skills have the reference and scripts

46:09

but all of them do have an MD document

46:11

okay? Now let's go here to this one

46:13

for example. And let me actually pull up

46:14

another one as well

46:15

Let me actually go here

46:17

So you can see the

46:18

the same kind of structure here

46:19

So

46:20

first thing that we have with a skill.md

46:21

document is the name and description

46:23

So

46:24

the name and description are used because when you tell Claude

46:27

hey, I want to make an Excalibur diagram

46:29

or I want to fetch my competitors' comments

46:32

right?

46:32

What it does is that the first thing

46:34

it looks through all the skills that we have here

46:37

which are going to be like 12

46:38

Only then does it look at the skill.md

46:41

it looks at this first part

46:42

name and description

46:43

and then it chooses which skill it's meant

46:45

to

46:45

use. The reason why it does that is

46:47

because if you don't have this small snippet of name description

46:50

it will go through all the files of all the skills

46:52

and it will consume so many tokens

46:54

and it will cost you a lot more

46:55

And so here we have a preview saying

46:57

hey, Claude, if you want to use a skill

47:00

there's the description, and there's the name

47:02

If you mention something similar to this

47:03

let's use that. It also increases the accuracy

47:05

of which skill it uses

47:06

So first things first

47:08

Name, descriptions, we have the name of the skill

47:10

Excalidor diagram

47:11

then we have the description

47:12

to use when someone asks to draw a diagram

47:14

make an Excalidor diagram

47:16

or build an editable diagram

47:17

default for all diagram requests

47:19

okay? The whole file here is done in Markdown formatting

47:21

because that's the way the AI knows how

47:23

to put the text in hierarchy

47:25

So heading two, heading one, heading three, right?

47:27

So it knows which one is more important

47:29

right? So for this

47:30

it's a heading two

47:30

this is a heading three

47:31

and it goes through step by step what

47:33

the skill actually is. Okay

47:34

so step one, understand the request

47:36

Step two, research if needed

47:37

Step three, plan the layout

47:39

Step four is generate elements

47:40

Step five, save and deliver

47:42

And then we go through step six

47:44

handle and feedback, critical rules, design principles, layout systems, and then color systems

47:50

typography skills, element schema, and then we have the JSON file

47:53

which is the type of design that Excalidor

47:56

which is a diagram that we use to make different diagrams

47:58

makes the diagrams itself

47:59

okay? So that's the skill right here

48:01

On the other hand

48:01

let's say I go here

48:02

comment analysis report, which is a report that I personally made

48:05

right? That I actually made step by step

48:07

We have the name description

48:08

Then we have comment analysis report

48:10

prerequisites for the skill itself

48:12

This is what it does

48:13

We need this to start

48:14

so let's grab this first

48:15

Step one, fetch comments

48:17

So what it does is that inside the skill

48:18

it goes through the scripts

48:20

fetch comments.py. So scripts

48:22

fetch comments.py, which is the first

48:23

you can say, automation of the whole system

48:25

Fetch the top 200 comments by relevance

48:27

Look at these channels right here

48:28

Step two, analyze and write a report

48:30

Step three

48:30

generate PDF

48:31

And then it goes through error handling and

48:33

partial runs and all that stuff in case it breaks

48:35

okay? And so that's exactly how it's made

48:37

Now here I have a empty folder instead of plot code

48:40

which is general

48:41

which is the one that I used before

48:42

I'm going to go here and create a

48:43

dot cloud folder

48:45

Inside dot cloud folder

48:46

I'm going to create another folder here called

48:48

skills

48:50

And then I'm also going to create another

48:51

folder instead of dot cloud

48:52

which is called commands

48:54

which is a new thing

48:56

Okay?

48:57

So commands will be the thing that allows

48:58

me to call the skills itself

49:00

For example, if I want to call a comment analysis report

49:02

I don't have to type comment analysis report

49:04

I can do slash comment

49:07

Analysis report, and this is a slash command

49:09

So I don't have to type the name

49:10

it automatically goes there. It automatically calls the skill itself

49:12

which is a good thing to have when

49:13

we do our skills. So inside skills now

49:15

I'm going to create a skill

49:16

It's a new folder inside

49:18

I'm going to name this YouTube title generation

49:23

Boom. And inside of YouTube title generation

49:25

I'm going to create a file

49:27

which is called the skill.md

49:30

And the skill.md will be the list of

49:31

instructions of that specific skill

49:33

The first thing I'm going to do

49:33

I'm going to paste this here

49:34

which I already wrote before

49:35

and this will be the name and the

49:37

description of the skill itself

49:38

So we're making a YouTube title generator

49:40

We generate YouTube video title ideas

49:42

use when asked to brainstorm titles

49:44

come up with video names

49:45

or generate title options for a YouTube video

49:47

And then down here

49:48

I can paste this quick sentence saying

49:50

generate 10 title ideas for the given topic

49:52

Titles should be under 60 characters

49:55

create curiosity

49:56

and be relevant to a tech

49:57

AI YouTube audience. And I can also say

50:00

here's a few title frameworks

50:03

That work well. And I can paste this here

50:06

which are some title frameworks that we use

50:08

for YouTube videos

50:09

So for example

50:10

I failed that topic until I learned this

50:12

I never achieved

50:13

or I never achieved goal until I learned

50:15

this

50:15

Learn the skill and never unwanted action again

50:18

If you action like this

50:19

you're going to be negative situation forever

50:21

Okay, so these are all title frameworks

50:22

And so that's where he'll be the skill

50:23

that I'm going to use

50:24

I'm going to press command S to save

50:26

or you can go here file save

50:29

I'm gonna test it

50:30

okay? So I'm gonna go here and say

50:32

I want you to help me

50:33

with title generation

50:35

I wanna create titles for my new video

50:37

which is about how I broke past

50:39

the six-figure mark with my AI agency

50:41

step by step

50:42

I'm gonna press go

50:43

and now, what it will do is

50:46

it will think

50:47

it will look at the YouTube title generation scale

50:49

because that's the skill that it needs to use

50:50

and now it will give me 10 title

50:52

ideas for my video

50:53

okay? Which is great

50:54

And so, what did we do?

50:56

We created the skill manually

50:57

and now we're testing. So we're testing that

50:58

this actually works. And as you can see

51:00

it understood which skill to use

51:02

Of course, we only have one right now

51:04

And by the way

51:04

just because it has YouTube title generation

51:06

it doesn't mean that it's gonna use that

51:07

every single time. And so we said

51:08

hey, I wanna help with titles

51:10

I wanna create a title for my new video

51:11

and then it gave me 10 title ideas

51:12

which is the exact thing that I told

51:14

it to do in that skill

51:15

I never hit 100k

51:16

until I learned this skill

51:16

Give me 50 minutes and I'll show you

51:18

how to build a six-figure AI agency

51:19

The smartest strategy to hit 100k with my

51:21

AI in 2026. These are pretty good

51:23

titles. It's not fun

51:24

but it'll get your agency to 100k way

51:26

faster

51:26

I like that. And so now you saw

51:27

me actually build this. Skills step by step

51:29

So now that you know how to make it manually

51:31

there's actually a trick that we use when making skills

51:33

and that's using the skill

51:35

Skill Creator. So there's a skill that actually makes skills

51:37

which makes the whole process much easier

51:39

Now, why does it do this is because

51:40

I'm going to press go

51:41

So you can actually find the skill under manage plugins

51:45

and you'll be able to see Skill Creator

51:47

So you can find it here

51:48

I can actually just

51:49

saw it

51:50

I can look at Skill Creator

51:53

You can install it here

51:53

Okay, so now it's installed in your MacBook

51:57

Now it's asking you

51:58

do you want it available for all your

51:59

projects? Do you want it available just for

52:01

collaboration or just for you

52:02

only for this exact title or for this

52:04

exact project?

52:05

I like to have this skill installed across all my projects

52:08

which means that first of all

52:10

I won't be able to see it here

52:11

but it will be installed in a way

52:13

where it's always available to you

52:15

and you don't have to reference it or do anything

52:17

because it's automatically going to be

52:18

used whenever you're making a skill

52:20

And so here I'll say install for you

52:21

available in all your projects

52:22

and now whenever I'm making a skill

52:24

it will use

52:25

this skill to make that skill

52:26

Now where does the skill creator come from?

52:27

So the skill creator is something that Anthropic

52:29

actually came out and made

52:31

which is why we use it all the time

52:32

and it's a skill

52:34

that allows us to

52:35

be able to use whenever we're creating skills

52:37

So you can see skill creator

52:39

a skill

52:39

for creating new skills and iteratively improving them

52:42

At a high level

52:42

the process of creating a skill goes like

52:44

this

52:44

Decide what you want the skill to do

52:45

and roughly how it should do it

52:47

Write a draft of the skill

52:48

create a few test prompts

52:49

and run Claude with access to the skill

52:51

on them. Help the user Evaluate the results

52:53

both on a quality level

52:54

but also on a quantitative level

52:55

Rewrite the skill based on feedback

52:57

repeat until you're satisfied

52:58

and expand the test

52:59

and try again on a larger scale

53:00

And now it goes through the whole process

53:02

to make the skill

53:03

which is much more efficient and even a higher quality

53:06

rather than us making it manually

53:07

So we're going to make a lead generation system skill

53:10

okay?

53:11

And this lead generation system skill will start by the input

53:16

which is niche, location, and amount of results

53:22

And that will be me just giving it to cloud

53:24

like on a chat basis

53:25

which is fine. And here we can use Apify

53:28

Google Maps scraper. By the way

53:31

if you don't know what Apify is

53:32

it's like the Amazon for scrapers

53:33

You go inside, you look at one scraper

53:36

which allows us to be able to go in Google Maps

53:38

extract the results based on what we want

53:40

So in this case

53:41

we want, let's say

53:41

roofing companies in North Carolina

53:44

It will scrape all of them

53:45

obviously within one of the results

53:46

And then it will go

53:49

actually filter

53:50

Leads by website exist

53:53

and then it will go inside the website

53:58

and then it will

54:01

give me a personalized icebreaker that I can use for LinkedIn

54:10

okay?

54:11

So this is an automation

54:12

right? Something that you would build in 2010

54:13

we're going to build it

54:14

here on Cloud Code

54:16

And we can simply call it every single

54:17

time that we want

54:18

So the first step I'm going to do

54:19

is explain to Cloud Code roughly what I want

54:21

okay?

54:22

So I'm going to say this

54:22

hey

54:23

so the task of today is to make

54:24

a lead generation system. The first step is

54:26

me giving you a niche that I want to target

54:28

the location of that niche

54:29

and the amount of results that I want

54:31

The second step is you have to go

54:32

to Appify and use the Google Maps scraper

54:35

to be able to do this

54:35

Then when you get the list of leads

54:37

you have to filter

54:37

by which leads have a website

54:39

The ones that don't have a website

54:40

just discard them. Then the ones that have a website

54:43

just go inside the website

54:44

scrape the website. And then finally

54:46

based on a personalized template that I'll give you

54:48

you have to make a personalized icebreaker

54:51

that I can use to connect people for

54:53

LinkedIn. The output will be a Google Sheet with company name

55:01

location, website, phone number

55:05

and it's fine if they don't have a phone number

55:07

and then I Breaker. Okay

55:09

that's it. Ask me any questions if you have any

55:11

and let's go from there

55:13

Okay, that's the prompt. Now

55:15

bear in mind, I didn't set up API

55:16

I didn't set up anything

55:17

So we're starting from scratch

55:18

right?

55:19

I'm going to go to plan mode

55:20

so we can plan things out

55:21

And as you can see

55:22

I didn't tell it to create the skill yet

55:23

because I'm going through everything step by step first

55:25

and then I tell it at the end

55:27

hey, now that you've done the whole thing step by step

55:29

you know what went right

55:29

you know what went wrong

55:30

just make the skill

55:31

Okay?

55:32

I'm going to press go

55:33

So how do you want to interact with

55:34

the Google Maps API?

55:36

Do you already have an API key or

55:37

service set?

55:38

I'm going to say API

55:40

How should the output Google Sheet be created?

55:42

Google Sheets API

55:43

Let's see if we do a CSV file

55:44

to make things easier. Tech stack

55:46

What language framework do you want this built

55:47

in? Python scripts. Yeah

55:48

I definitely do not want to be building an Anytime workflow

55:50

or else that would not be the point

55:51

of this video. Submit answers

55:53

And if you're wondering whether something like this

55:54

will be built in an attend versus Cloud Code

55:57

I would definitely build this in Cloud Code

55:59

because it's much easier as well

56:00

and we don't have to

56:01

pretty much worry about much

56:02

Now it's asking me

56:02

do you already have an API

56:04

key for API? So I don't

56:06

so I can actually go find it

56:08

Dot com. And by the way

56:09

if this is the first time that you hear about Appify

56:12

I would definitely recommend checking out this video up here

56:14

which goes through the full tutorial

56:15

on how it works. I'm saying that because

56:17

it is one of the most

56:18

used softwares that I use when building automations

56:21

So here I am on the dashboard

56:22

I can go down to settings

56:23

I can go to API and integration

56:25

and I can copy my key

56:27

And by the way

56:28

you get $5 of free credits when you apply

56:30

or when you actually get into Appify

56:32

I have $29 because that's what I paid for

56:34

but you have the free version

56:35

which is $5. And if you're anything like me

56:37

and you want to rinse the software

56:38

just make five different emails

56:40

and then go through them

56:41

So for the LinkedIn icebreaker template

56:43

do you want to provide it now

56:44

or should I build the system first and

56:45

you plug in later?

56:47

Let me give it to now

56:48

So instead of doing LinkedIn icebreaker

56:50

why don't we do this?

56:50

Since we're going to get their

56:52

their number as well

56:53

right? And by the way

56:53

you can filter by people who have a number

56:55

like a phone number itself

56:57

I want you to find from their website

56:59

so still do the exact same thing I asked you to

57:01

but I want you to find from the

57:02

website the value proposition of the business

57:05

what makes them unique

57:06

and give me a

57:07

case study Of a client that they've helped

57:10

if you can find it

57:10

And this is part of testing

57:11

working with call code

57:12

and so on, which is good

57:13

For scraping websites to personalize icebreakers

57:15

how deep should the scraping go?

57:17

Homepage only, homepage and about, homepage about services

57:20

I'll go homepage about services as well

57:23

Because I want to know pretty much everything

57:24

about it. It'll be a bit more expensive

57:26

sure, but that's fine. I can submit answers

57:28

and now we're going to go ahead

57:30

Now it will remake the plan because I

57:31

told it to change a specific thing

57:33

about

57:34

the process. So now it's saying that the

57:35

input will be niche location

57:36

number of results, the APIFY Google Maps scraper to find businesses

57:39

filter keep only leads with both a website

57:41

and a phone number. Cool

57:42

Scrape each website, home, about, and services, extract value proposition, what makes them unique

57:47

and case study of clients who helped

57:48

if findable, and output will be a CSV file

57:50

Great

57:51

perfect. Because now it's asking me as well

57:52

for the anthropic API key to analyze websites

57:55

So I can go to the console here

57:57

which is the anthropic

57:58

platform.cod.com

58:00

I can log in

58:04

I can go down to API keys

58:06

I can make a key

58:07

name it cloud

58:10

cloud code, copy the key

58:13

and paste it here

58:14

By the way

58:14

the reason why I'm pasting it into others

58:16

is because if I press yes

58:17

I have one

58:17

I then have to wait a minute for

58:18

it to understand that I told it yes

58:20

I have one, and it will ask me

58:21

I just might as well just give it

58:22

the API key itself. Okay

58:24

cool

58:25

The plan is ready

58:25

here's a summary, and it has everything here

58:27

So as you can see

58:28

we have the lead generation system

58:29

implementation plan, context, pipeline overview, file to create, implementation details

58:34

step one, Google Maps scrapings, set to his filter

58:36

website scraping, cloud API analysis, CSV output, error handling dependencies, verification

58:42

Great, amazing

58:43

I can press yes

58:44

auto accept

58:45

And now I can go to bypass permissions

58:48

and it will

58:48

start building it up step by step

58:50

As you can see now

58:50

it's starting to make different files

58:53

and it's not creating its skill

58:54

it's just creating the actual file itself

58:56

because we didn't tell it yet to create

58:57

the skill. It said it created the files

58:59

Great. Want me to run the test?

59:00

Yes. Run the test with roofing companies

59:05

And

59:06

I don't know why I'm very

59:07

adamant on roofing companies in North Carolina

59:09

I've never been there in North Carolina

59:11

and I want, and I want hundred results

59:14

Let's just go fifty

59:15

actually forty, to make things faster

59:17

Actually, I'll do thirty

59:18

because I'm not trying to wait all day

59:19

So there you go

59:20

run the test

59:21

Now it will start working on it

59:22

So now it's running the actual

59:24

lead generation automation. If I go to Appify

59:26

I can see that it just scraped thirty-four results

59:28

right? It just scraped it

59:29

and now it's going to use those results

59:31

to then go and scrape each website of those results

59:33

obviously assuming that they have the website and

59:35

phone number. All right

59:36

so you just finished running

59:37

We have the leads.csv document

59:39

which is this one right here

59:41

I can actually command A

59:43

copy

59:44

I can go to a new Google Sheet

59:45

right here. I can go here

59:47

I can paste this

59:47

and then I can go to data

59:49

to text and columns. There we go

59:51

And now we can have Different

59:54

Thanks

59:54

sir. So I can

59:55

make this

59:56

both a little bit of formatting here and

59:57

there

59:58

Make this white. Sorry

59:59

guys

59:59

I have to do this every single time

60:00

because I also can't work with Google Sheets

60:02

And now we have the company name

60:04

location, the website, we have the phone number

60:07

value proposition. So they do comprehensive roofing

60:09

gutter solutions with a family-owned approach that prioritizes

60:12

customer satisfaction and personal attention

60:14

They offer financing options. Okay

60:15

cool

60:16

Unique differentiation. So this company is led by

60:18

John Davis with 30 years plus of

60:20

industry experience

60:21

They maintain an A plus BBB rating

60:23

and don't consider a job complete until customers

60:25

are 100% satisfied. Great. Now

60:27

as you can see for case studies

60:29

interesting, most of them didn't have case studies

60:31

which is fine. And so this right here

60:32

is just an extra thing that we can

60:34

do. I'm going to leave it here

60:35

right, because some of them do

60:35

So we have a client testimonial from Tara

60:37

Manson details on how Brian completed multiple home improvements

60:40

projects inside a building. And so

60:42

the reason why I asked this is because

60:43

let's say we're cold calling the business

60:45

We can see something like this

60:47

We can commend the business on having a

60:48

case study and we can reference the name

60:50

of the person who gave the case study

60:52

right, which is always good

60:53

Personalization over anything is great

60:55

And so that right there is the list

60:56

right, that we have here

60:57

which is great. What I can do now

60:58

is this. I can say

61:00

okay

61:00

great. Now what I want you to do

61:01

here is the whole process that I run you through

61:04

I want you to create a skill

61:05

All right, so now in terms of improvements

61:06

I want to do a few things

61:08

For example

61:08

location sometimes gets mixed up with company name

61:11

Website sometimes get mixed up with the location

61:14

Phone number

61:14

and this is pretty much all shifted

61:16

for no reason. So

61:17

I can go here and I can say

61:18

there was one problem

61:19

like for example, Guardian Roofing of North Carolina

61:22

the location

61:23

is

61:24

LC. The website and then the value proposition

61:26

is the phone number and so on

61:27

right? And so we are shifting everything

61:29

by one. So I want to make sure

61:31

that when you give me the CSV

61:32

you check that all the columns that we

61:34

have are clearly placed

61:36

right? There are a lot of them that

61:37

don't have a location

61:38

which is fine

61:39

so we can leave it there

61:40

and you put things in the right place

61:41

But there are some of them as well

61:43

that have different things in different places

61:45

So I want you to fix that and

61:46

then give it back to me

61:47

And this is part of iterating the actual

61:49

skill itself. All right

61:49

so now it told us that it updated

61:50

the leads.csv

61:52

I can go here

61:53

I can Copy this and go back

61:55

I can make a new sheet and I

61:56

can paste it here and I can go to data

61:59

split text to columns

62:00

and as you can see here

62:01

now it fixed

62:02

the area that we had before

62:04

Which is perfect

62:05

There we go

62:06

Now, with all this said

62:07

we can go to the next step and

62:08

actually create this skill that we need

62:11

I'm in a new tab

62:11

and I can say

62:12

okay, now that you've generated the lead generation kind of workflow

62:16

I want you to create a skill

62:18

so that when I call the skill

62:20

it basically runs through the whole process that

62:22

we just went through

62:23

okay? So let me know if you have any questions

62:24

but create a skill called lead generation system

62:26

and then, add it in the

62:28

in the folder itself. I can press go

62:30

I can put plan mode

62:31

but essentially now it will use the skill

62:33

of skill creator

62:34

to then be able to create the skill

62:35

Okay, so now I can press yes

62:36

and then

62:37

I can go from plan mode to bypass permissions

62:39

All right, so now I've finished creating the skill

62:41

I can go here to lead generation system

62:43

and I can see that I have a document called skill.md

62:45

and this contains a full instruction manual for

62:47

that specific skill. So lead generation system

62:49

description, instructions, and we go on and on

62:52

The only thing they didn't do

62:53

is that it didn't make a folder here

62:55

called script and references

62:57

because this right here and this one here

62:59

they're both references, even this

63:01

and then this right here is a script

63:02

So what I can say here is

63:03

Inside the skill folder for the lead generation system

63:06

can you also add a references and

63:09

scripts folder to store the appropriate files?

63:12

So the lead generation.py needs to be in a scripts folder

63:15

and in the rest as well

63:16

like

63:16

based on what you know and

63:18

obviously your definition of scripts and references

63:20

just include the files there to make things

63:22

organized. And I can press go

63:23

As you can see now

63:24

I just made two different folders

63:25

references and scripts. And inside scripts

63:27

we can see that we have lead generation.py

63:29

and inside references, we can see that they have env

63:31

requirements, text, and sample output CSV

63:34

which is great. One thing I want to do

63:35

though, is I want to delete these ones

63:36

So I can say

63:38

okay

63:38

now that you added those inside the references and scripts folders

63:41

can you delete the other ones that we're

63:42

not going to use? As you can see

63:43

it deleted both of them

63:44

and it's also asking me

63:45

hey

63:45

do you also want to delete dot env

63:47

and leads dot csv in the output files?

63:50

Yes, delete them. There we go

63:51

So now we have a references folder

63:53

we have a scripts folder with the automation

63:54

In this case, the Python file

63:56

which doesn't make any sense to anyone here

63:57

It's just a file that wants the automation

64:00

right? Because behind every automation in any 10

64:03

make the com It's all Python

64:04

so it's all JSON. Because as you may know

64:06

all the platforms in N10

64:08

or make the column behind the automations

64:10

it's all code. And so

64:11

this is the code itself that runs the

64:13

actual system. And as you can see

64:14

we start to add more and more skills

64:15

with more and more folders

64:16

Now there's one more thing that I want to do

64:17

and it's this

64:18

which has to do with commands

64:19

Now

64:20

I have a skill called lead generation system

64:22

I want to create a command

64:24

so that I can put like slash lead generation

64:26

and then it calls that specific skill automatically

64:28

Can you make that command?

64:29

I can go bypass permissions

64:30

and I should be able to see

64:31

a command be made here

64:32

All right

64:33

so now I've finished making the command

64:34

As you can see

64:34

we have the lead generation.md here

64:37

So if I go to a new chat

64:38

I can press slash lead generation

64:41

and then that will

64:42

activate the lead generation

64:44

system we have in the skill itself

64:46

There you go

64:47

And now it's going to ask me

64:48

hey, what niche do I want

64:49

location, number of leads, and output file?

64:51

I was going to say roofing companies again

64:53

Let's buzz in London

64:56

and I want to find 30

64:59

and I want to name the file MedSpa

65:05

London, thirty

65:08

Boom

65:09

And I can go to bypass permissions

65:11

Okay?

65:11

So that right there

65:12

is the full process of creating a skill

65:14

step by step and actually testing it and

65:16

making a command so that whenever I go to Cloud Code

65:19

I can simply just press slash the skill

65:21

name and it will automatically call the skill itself

65:23

All right, as we can see here

65:24

we have the result

65:25

MedSpa's in London, I can make a new sheet

65:27

data

65:29

and then we have the full

65:30

list right here

65:32

And we just call this simply by pressing

65:33

slash command

65:35

Of that specific skill. So now you know

65:36

exactly how to build your own cloud code

65:37

skills. What if I told you

65:39

that your cloud code skills could actually improve

65:41

themselves 24-7

65:42

overnight without you having to do anything?

65:44

This is where we get to self-improving skills

65:46

Like we saw earlier

65:47

the first way of actually self-improving your own

65:49

cloud code skills is to do it manually

65:51

So we gave it the first basic version

65:53

it made this right here

65:54

Then we said, okay

65:55

this right here is a bit messed up

65:56

this right here is a bit messed up as well

65:58

can we change this? And then it came

66:00

and did this as well

66:01

And so what it did is that it made it better

66:02

it improved itself. And this is great because

66:04

it allows us to make a near perfect

66:05

score or the output for our specific skills

66:08

But if you want to take your skills

66:09

to the next level and actually measure how

66:11

much it improved by

66:12

well

66:12

this is where we get to the skill

66:13

evaluation. So the first step of skill evaluation

66:16

is making sure that you have

66:18

the skill creator installed. So you need to

66:20

have this installed

66:20

And this is something that we spoke about

66:22

in one of the first steps

66:23

Then once we have the skill involved

66:24

we then have to tell Claude to evaluate

66:26

our skill. And so here I can say

66:27

create an eval for my lead generation system

66:30

skill

66:30

I'm just going to do the YouTube title generation

66:32

just to keep it more simple

66:33

So for my YouTube YouTube title generation skill

66:37

and I can press go

66:38

make sure it's bypass permissions

66:39

As you can see here

66:39

it created two eval files

66:41

One is eval.json and one is trigger eval.json

66:44

Basically

66:44

the whole point of evals is to make our skills self-improving

66:47

and to make the skills itself

66:49

improve themselves over time. The first thing it

66:51

made is three test use cases

66:52

So if I go here to the skill itself

66:55

it made a folder called evals

66:56

and we have two. So the first thing

66:57

is test cases

66:59

So it uses specific inputs to run the

67:01

skill. So let's say we're using the prompt

67:02

of cloud code beginner video

67:03

what a test is framework usage

67:04

character limit

67:05

curiosity variety. Then it uses VS Code versus

67:07

cursor versus windsurf comparison. So basically

67:09

these are three video ideas

67:10

and this is what it tests for

67:12

right? Each eval has five to six expectations checking

67:16

Exact count, time titles, character limit, 60

67:18

framework adherence, topic relevance, curiosity, engagement, and variety

67:21

Whilst on this side

67:22

we have 20 trigger tests

67:23

10 should be triggered

67:24

10 should not. Should trigger is give me

67:26

the YouTube title ideas

67:27

brainstorm video titles, help me name my video

67:29

etc. Should not trigger is writing scripts about YouTube

67:32

video script, thumbnails, SEO analysis, blog titles

67:34

channel growth. Advice adjacent but distant tasks

67:36

And so now what it'll do

67:38

is it will test the actual skill based on this criteria

67:41

and it improves

67:42

if it finds that this output isn't good

67:43

enough. So it will test all the three use cases

67:45

and it will test all these 20 triggers

67:47

With this said, I can now say

67:49

run the eval for my huge petal generation skill

67:51

and I can go press run

67:52

What this will do now

67:53

is it will go and run this too

67:55

and then give us a kind of a report

67:57

All right

67:57

perfect. Now it's finished with the evaluation

67:59

It's telling us that with the skill

68:00

it did 100% correct. Without the skill

68:02

33%. With the skill

68:04

100%. And then average will be 50% here

68:06

and average will be 100% here

68:08

And it also gave us full document here with the prompt

68:11

with the skill, the outputs that we got

68:13

four more grades, so six out of six passed

68:15

zero failed, which is perfect

68:17

Then we also have the benchmark comparison here

68:19

which gives us all the data of

68:21

what the output looks like with the skill

68:24

and what the output looks like without the skill

68:26

okay?

68:26

And so here we have the pass rate with the skill

68:28

without the skill. Obviously here we have with the skill

68:30

100%. Without the skill

68:32

50% plus minus 17. So as you can see

68:34

with the Skill is a perfect score of

68:36

100. And so what it did here

68:37

is that it evaluated 1

68:38

2, and 3, so that we have the Cloud Code Beginner

68:41

which is the first topic that it used

68:42

the first test use case

68:44

With the skill, without the skill

68:44

you can see all the numbers here

68:46

Assertion

68:46

so all the different things that we needed inside that specific

68:50

title. With the skill

68:50

it got everything right. Without the skill

68:52

only two. And it did that for all

68:53

the three topics. So topic number one

68:55

topic number two, and topic number three

68:57

And then we have the analysis notes

68:59

So with the skill

69:00

achieves 100% pass rate across all three evals

69:03

versus a 50% average without the skill

69:04

The biggest skill advantage is in the format compliance

69:07

Exact count, 10 titles, and character limit, baseline consistency, overproduced

69:10

and exceeds length limits. And so all this

69:12

is doing is it is testing three different

69:14

use cases for our skill

69:15

and it's telling us

69:16

hey, out of 10 times that we run this

69:17

how many times fail? How close is it

69:19

to the desired output that we asked it

69:20

to do? And so you can always have

69:22

this evaluation for every single skill that you do

69:24

you'll be able to test it

69:25

Now, of course, YouTube title generation is a very easy system

69:27

but if we were to do

69:28

something with the lead generation system

69:30

it would work really well

69:31

just because there's a lot of moving components

69:32

in the actual skill itself

69:33

And so at this point

69:34

you could Tell it to improve the skill

69:35

run the eval again

69:36

and see the numbers and how it's improved

69:38

And so at this point now

69:39

we have a hundred percent accuracy rate

69:40

but let's say we had eighty percent

69:42

So what we would do at this case

69:43

in case it has eighty percent

69:45

or anywhere less than a hundred

69:46

I can tell it

69:46

hey, it's missing X, Y, and Z

69:48

can you improve the skill?

69:49

It then improves the skill

69:50

it tells us, hey, it's done

69:51

and then we ask it to give us

69:52

the evaluation again. And this is a whole

69:53

repetitive loop that it does

69:55

to make sure that these skills are self-approving

69:57

One more thing I wanted to go through

69:58

when it comes to skills

69:59

is the ability to call them using slash skill

70:03

right? So in this case

70:04

in the dot call folder

70:05

we have commands and we have skills

70:07

I've gone through skills

70:08

but commands are the things that we use

70:10

to call that specific skill

70:11

right? So for example

70:12

let's say I wanted to run

70:14

YouTube title generation

70:15

Instead of saying, hey, can you run YouTube title generation

70:19

I would say, slash YouTube title generation

70:22

and it will simply run automatically

70:24

right? Just because we've added the command

70:26

which allows us to call it directly

70:27

Same way that if I wanted to run

70:29

the views to clicks

70:30

I would go slash views to clicks analysis

70:33

and it will run the skill itself

70:35

Now

70:35

this is great for us because

70:36

we simply don't have to manually

70:39

write the skill itself. We can just put slash

70:41

call the skill, and you're good to go

70:42

And if I go to slash

70:43

I can see down to the commands

70:47

I can go down to the skills right here

70:48

and I can see all my skills that

70:50

I can call automatically

70:51

and it will simply trigger that specific automation

70:54

And the way that we did that is

70:55

let's say we have XcaliDraw diagram

70:57

and we have YouTube title generation

70:59

If I wanted to call this one here

71:01

I would have to put slash XcaliDraw diagram

71:03

and I simply don't have the command there

71:04

So what needs to happen because of that

71:06

is I would have to explicitly say

71:08

hey, can you run my skill called the XcaliDraw diagram

71:15

right? Which is quite inefficient

71:16

especially when you have twenty to thirty Skills

71:19

that you want to browse through and just pick one automatically

71:21

right? And so in this case

71:22

to turn that skill into a command

71:24

I can say, hey

71:25

can you turn my two skills into two commands and then

71:29

so that I can call them using slash

71:30

the name of the skill

71:31

and then add the file

71:34

inside the commands folder under the doc cloud folder

71:37

All right

71:37

now I can press

71:38

plan mode because I wanted to make a

71:40

plan first. All right

71:41

so here we have the plan itself

71:42

turn skills into slash commands

71:44

We have two skills and we're going to

71:46

create the commands as well

71:47

I can just press yes and auto accept

71:49

You can see here that it's going to

71:50

create a file called under the commands

71:53

under here

71:54

Same thing with this

71:55

YouTube title generation and Excalidraw diagram

71:57

I can press yes

71:58

auto accept

71:59

I can go to bypass permissions

72:01

and I can let it do its thing

72:02

right here. All right

72:03

so after a few minutes

72:04

it made the two different commands

72:05

which are two different files

72:06

So we have this one here

72:07

which says when someone calls slash Excalidraw diagram

72:11

use the Excalidraw diagram skill to handle the

72:13

request. Same thing with the YouTube title generation

72:15

So when someone calls slash YouTube title generation

72:18

Use the skill right here to handle the request

72:21

That way

72:21

if I go here and I can put

72:22

slash YouTube title generation

72:25

it'll run the skill itself right here

72:27

As you can see

72:27

it's looking at all the different references

72:29

Then it should ask me

72:30

what video

72:31

do I want to make a title for?

72:32

As you can see now

72:33

it's asking me for the video topic

72:34

angle

72:35

number of titles. And then if I go

72:36

to a new conversation

72:38

I can do the exact same with Excalidraw

72:40

So I can put slash Excalidraw diagram

72:42

and it will start running the Excalidraw skill

72:45

And now it's going to ask me what

72:46

concept or system do I want visualized

72:48

and what level of detail

72:49

And the next level would be

72:50

obviously having 10 to 15 different skills

72:53

with 10 to 15 different commands

72:55

so that we're able to call the skills

72:56

directly

72:57

Without us having to

72:59

type the skills off or have a conversation with Cloud Code

73:01

we can simply just filter by the skills

73:03

You can go here

73:04

and we can look at all the skills

73:05

that we have and call them directly right here

73:09

Right

73:10

so now you understand how to build

73:12

custom skills and slash commands

73:14

that make Cloud Code do

73:16

what you need in one line

73:17

But what if sometimes

73:18

you want Cloud Code to do things automatically

73:21

without you even having to ask?

73:22

Well, that's exactly what hooks are for

73:24

and that's what we're covering in the next

73:26

video. Let's dive in. Right

73:27

now the one feature that I'm talking about is

73:29

of course, Cloud Code hooks

73:31

If you've never heard of Cloud Code hooks

73:32

they are a bit more on the technical side

73:34

but they're actually very

73:35

very easy

73:36

and very powerful when you understand how to

73:38

use them. So what they are is

73:39

automatic actions that run at specific moments

73:42

like every time X happens

73:44

automatically do Y

73:45

and it's almost like setting rules that your

73:47

assistant always follows. Now

73:48

the only difference between Cloud Code hooks

73:50

and Cloud Code skills is that hooks

73:53

are

73:54

things that Cloud Code uses

73:56

regardless of whether you call them or not

73:57

They're on the organization level

73:59

right? So you set a rule

74:00

so every single time Cloud Code runs

74:02

it always uses these hooks

74:04

okay? And they're automatic things that I'll walk

74:06

you through what they

74:07

what they actually are. For example

74:08

So you give CloudCode a task

74:10

you ask CloudCode, CloudCode does the work

74:12

and it uses these hooks

74:14

right, whenever it does the work

74:16

For example, we have pre-tool use

74:18

so we have

74:19

hooks that we have to use

74:20

before the tool runs. Post-tool use

74:22

which are hooks that we have to use

74:23

after the code is written

74:24

Stop when CloudCode finishes

74:26

or notification when CloudCode needs you

74:28

So these are the four different types of

74:29

hooks that you can have

74:31

and they're just examples of what it could look like

74:33

right? So you have a hook

74:34

that runs every single time the workflow finishes

74:37

or CloudCode starts, or when it stops

74:39

or when it needs notifications

74:41

And they're almost like triggers

74:42

what starts the actual hook itself

74:44

Now, I'm here on the YouTube workspace

74:46

and if I go to my doc loft folder

74:48

which is where the hooks are going to be stored

74:50

there's a file called settings.json

74:52

If I press on settings.json

74:54

you'll be able to see the full code

74:56

Now, the code itself is actually very

74:59

very easy to understand

75:00

right? Now, don't worry about the code too much

75:01

because I'll explain exactly what it is

75:04

but essentially

75:04

it's stored in a file called settings.json

75:07

And so here I am on a new

75:08

project. I can go to new folder here

75:11

I can name it doc cloud

75:12

And by the way

75:13

doc cloud is also the folder that we

75:14

use to store skills

75:16

to store commands

75:17

to store different things of cloud code

75:18

And inside doc cloud

75:20

we want to add a new file called settings.json

75:24

Now, a few things

75:25

First of all, that whenever we're making a hook

75:29

a hook

75:29

some sort of like automation or some sort of plugin

75:31

we

75:32

are all making it

75:33

under the same exact file

75:34

So this is not like a skill where

75:36

you would need to add a new

75:37

file for every single hook

75:39

We use settings.json as the main file to

75:41

store all the different hooks that we have across our organization

75:45

okay?

75:46

Second thing is that when we are

75:48

installing hooks in settings.json

75:51

you will be able to use this hook

75:53

across different projects. This will make more sense

75:55

when I show you

75:55

but essentially I understand those two rules

75:57

First thing

75:58

you will be able to write all the hooks here

75:59

and then second thing

76:00

just as if you were able to

76:01

to write all the skills in one file

76:03

without you having to make different folders

76:05

but in this case

76:06

skills are different. And then you

76:08

need to understand This is on a organization level

76:10

not on a product level

76:11

With this said, let's go to here

76:13

and the way to

76:14

actually use hooks is you can either press slash hooks

76:18

which

76:19

you will be able to only to do using your terminal

76:21

as you can see right here

76:23

terminal

76:24

Let me go here to terminal

76:26

new terminal, and I can press cloud

76:28

So here we are on a terminal view using cloud

76:30

and I can put slash hooks

76:32

view hook configuration for tool events

76:34

I can go enter

76:35

and here we have zero hooks configured

76:37

So we have zero hooks that are configured

76:39

right now in our organization

76:40

okay?

76:41

And here

76:42

we have a read-only menu to add or modify hooks

76:46

okay?

76:46

And so in the easiest way

76:48

what this is, is you being able to say

76:50

okay

76:51

I want to run some sort of

76:53

notification before the tool execution

76:55

And so these are all triggers

76:57

right? When I say trigger

76:58

I mean

76:58

this is a trigger before the tool execution

77:00

This is a trigger after the tool executes

77:02

This is a trigger post tool use failure

77:04

so when the tool fails

77:06

So let's say you want to send yourself

77:07

an email whenever a tool fails

77:09

you use that

77:10

Permission denied, after auto mode classifier denies a tool call

77:13

then you use the hook

77:15

then you can make a hook there

77:16

And so you have different options

77:18

that you can work with whenever

77:20

you have these kind of

77:21

statuses inside of Cloud Code

77:23

Let's say I go to notification

77:26

press enter

77:27

It says, we currently have no hooks configured

77:30

To add a hook

77:31

you either need to edit settings.json directly

77:34

or you need to ask Claude

77:35

And now, let's go to the normal view

77:38

And I start the clock up

77:39

Say, okay, hey

77:41

I want to make a hook

77:43

So here it's asking me

77:44

what kind of hook do you want to

77:45

do?

77:46

Here are some common hook events

77:47

And I have running before the tool is called

77:49

run after a tool completes

77:51

run when cloud sends a notification

77:53

run when cloud finishes a response

77:54

or run when sub-agents finishes

77:56

Let's do this. Let's run a hook where

77:57

when cloud finishes a response

77:59

it sends me a notification

78:01

Okay

78:01

so that's going to be the hook that

78:02

we're going to make

78:03

So when Claude finishes doing something

78:05

it actually gives me a notification

78:07

I can press yes

78:08

As you can see now

78:09

it's running the hook. And by the way

78:10

when you see something like this

78:11

always in Claude code

78:13

it just means that this is the previous version

78:15

this is the updated version

78:16

And it's saying

78:17

do you

78:18

agree that we can change this exact file?

78:20

And I can say yes

78:22

And now it's going to add the hook

78:23

instead of settings.json here

78:25

And it says that the hook is live

78:26

Every time I finish a response

78:27

you'll get a macOS notification

78:29

saying that Claude has finished responding with the

78:31

title Claude code. And I can review or

78:33

edit at any time via hooks

78:35

So in theory now

78:35

what we made is a hook

78:37

that alerts me every single time

78:38

Claude code finishes doing something

78:40

So that I'm able to then do whatever I want

78:42

work on different things simultaneously

78:44

and I'll know when I have to go

78:45

back to Claude code to see the output

78:47

So here I can say

78:48

what's the capital of Italy?

78:49

And I will get a notification when it

78:51

finishes running. Now this is the exact notification

78:53

that I just got. The reason why I'm

78:54

showing you here is because I'm using my

78:57

my monitor

78:57

so it doesn't actually show you the notification

78:59

on the monitor as well

79:00

But essentially what would happen Is that

79:02

you would be talking to Cloud Code

79:04

at the same time you could be doing something else

79:06

and you will know whenever you have to

79:08

go back to Cloud Code

79:09

because of this notification right here

79:11

And there, there is a hook

79:12

right?

79:13

And this will run

79:14

every single time that you talk to Cloud

79:15

Code. Regardless of whether you have all these

79:17

documents in your project

79:18

it will always run

79:19

right? These are rules that will always run regardless of X

79:22

Y, and Z. So in this case

79:23

when Cloud Code finishes

79:25

then we get a notification

79:26

All right, so another example here could be that

79:29

let's say we had a.env document

79:31

So this.env document is full of API keys for OpenAI

79:34

Anthropic, Stripe, all these things

79:36

And by the way

79:37

they're fake, so don't try to copy them

79:39

Fake API keys. Let me just put

79:40

API keys

79:41

None of these are real

79:42

And

79:42

what I want is that whenever I talk to Cloud Code

79:45

maybe I could be doing a complete custom code

79:48

full stack project, or maybe I'm doing a full website

79:50

or maybe I'm building some

79:51

some automations, right? Ideally

79:52

we wouldn't want Cloud Code to impact the.env document

79:56

because once this is impacted

79:57

let's say it deletes everything by mistake

80:00

we have to reconfigure Everything

80:01

And all the automations break because they're all

80:03

reliant on these passwords

80:05

right, to access the softwares for the automations

80:07

So what we can do here is I can say

80:10

I want to make a hook

80:12

So in this case

80:13

we want to tell Cloud Code

80:15

hey, don't edit the dot env at all

80:18

right? And so in this case

80:19

it would be a pre-tool hook

80:21

So we would say

80:22

yeah, so this is a pre-tool hook

80:23

Essentially what we want is

80:26

I want the Cloud Code

80:27

to not touch the dot env documents

80:29

So don't ever touch the dot env documents

80:31

regardless of what it is that we're doing

80:32

So if you have to edit the dot env

80:34

then you can't do it because you have

80:36

to stop yourself because it's sensitive information

80:38

So now we're making another hook

80:39

I can press yes. I'm going to go

80:40

to bypass permissions

80:42

so I can

80:42

let it create the own hook

80:45

There we go. So now it added the

80:46

pre-tool use onto our settings of JSON

80:49

So if I go here

80:50

I can see now that I have the stop hook

80:52

which means that whenever the Cloud Code stops

80:54

it gives us a notification

80:55

So we get a notification here

80:57

when Cloud has finished responding

80:59

And then we also have the write edit one

81:01

which is

81:02

the one that allows us to block env

81:06

And then we have the pre-tool use hook

81:08

which is the one that we use

81:09

to be able

81:10

to not let Cloud Code touch Our dot env document

81:13

because of course it has all the passwords

81:15

So now here I can say

81:16

create a dot env file with API key

81:19

test123. And you can see that it says

81:21

I can't do that. Your settings.json has a

81:23

hook that blocks me from writing to the dot env files

81:26

since they contain sensitive information

81:28

You can create it yourself by running echo this

81:30

this, this, right? And so this is extremely important

81:32

right? I wish I knew this sooner

81:33

because there are many

81:34

many times where I told Cloud Code to do something

81:37

and it deleted something from here

81:39

or it added something from here

81:41

which messed up my whole project

81:42

And so in this case

81:43

what we're saying is

81:44

whatever happens, don't edit this, right?

81:47

What I would recommend

81:48

is you add all the API keys that you have

81:50

all the passwords that you have

81:51

that Cloud Code needs to have access to

81:53

whenever it needs to run a specific automation

81:55

And then once you have those

81:57

put this hook here

81:58

make sure that now

82:00

Cloud Code is not able to edit the dot env

82:02

to be able to make sure that you're

82:03

safe whenever you're building the project

82:05

And last example

82:05

let's say I am building a project for

82:07

a client. At the end of every session

82:08

I want to know

82:09

exactly what files Cloud Code touched and when

82:12

Which is an audit trail

82:13

And so what I want is a log

82:14

of all the

82:15

kind of edits of different files that I made

82:17

with the timestamps of when it actually touched

82:19

them. So here I can say

82:20

add a post tool use hook

82:21

because post tool is going to be

82:23

the trigger. After Cloud Code finishes

82:25

it logs

82:26

every file Cloud Code edits or writes to

82:28

a file called cloud slash log dot txt

82:30

with a timestamp

82:32

And we can press go

82:32

And this is great because a lot of

82:34

the times that I'm building with Cloud Code

82:36

I am making stuff

82:37

and then I'm forgetting which files I edited

82:39

And sometimes things just get deleted

82:42

And so

82:42

this allows me to know exactly what got

82:44

edited and what got deleted

82:46

or what timestamp. All right

82:48

so it just finished running

82:48

It said the post tool

82:50

use hook is now configured

82:51

I can see here that I have the write edit

82:53

type command

82:54

and I have the status login file change

82:56

to

82:57

All of this makes no sense to me

82:59

It should log everything to a cloud

83:01

slash log dot txt document

83:03

okay? So now if I go here and say

83:05

can you make a new

83:06

markdown format document

83:08

and name it YouTube title generation?

83:11

As you can see now

83:12

it's getting the tasks

83:14

but it's also updating the cloud slash log.txt

83:16

If I go here

83:17

I can see that 174409

83:20

we wrote YouTube channel generation.md

83:23

Let's say I wanted to say

83:24

can you also make another document or another five documents

83:27

naming them, hook number one, hook number two, three, four, five

83:31

So essentially, it's whatever change you make

83:32

it will log it here

83:33

So as you can see

83:34

it made five different documents

83:35

If I go here

83:36

I can see that we now have

83:38

five different logs. So one

83:40

two, three, four, five. And it'll be hook one

83:42

hook two, hook three, hook four, hook five

83:44

That way you can have a log of

83:45

all the edits that you've made

83:46

so you can always come back to them

83:47

in case it went wrong in some way

83:52

Alright

83:52

so now Cloud Code can actually run custom

83:54

commands and run actions on its own

83:56

But there's still one limitation

83:58

and that is that

83:59

it can only work with what's inside your

84:01

machine. What if you want Cloud Code to

84:03

talk to your browser

84:04

talk to your databases

84:05

talk to your APIs

84:07

or any external tools? Well

84:08

as of right now

84:09

it can't really do that without this one thing

84:12

which is called Model Context Protocol

84:14

and it stands for MCP

84:15

and it completely changes what's possible

84:17

Let me show you. Alright

84:18

so the best way to explain MCP is they're just plugins

84:21

right? Essentially, other people build tools

84:24

and you just plug them in

84:25

So you have Cloud Code

84:26

which talks to the MCP server

84:28

which is running in the background

84:29

and the MCP server will be the one

84:30

controlling the external apps

84:32

So your browser, your databases, maybe it's Notion, maybe it's Superbase, maybe it's your Gmail

84:37

right? Whatever that may be

84:38

And it makes it so much faster for

84:40

us to be able to use the softwares

84:42

because it is trained on how it should

84:43

talk to the API documentation

84:45

And the API documentation is essentially

84:47

the documentation which shows how you can actually

84:49

automate something with the software

84:51

So you can install the MCPs

84:52

they run as servers

84:54

right? And Cloud gets new tools

84:55

and talks to real apps

84:56

So you basically don't have to configure things

84:58

yourself. And so how are they different from

84:59

skills and hooks?

85:01

Well, skills are just instructions

85:03

We say do this task like this

85:04

right? We have step-by-step instructions that Cloud could

85:07

use to fulfill a certain task

85:08

Now I made a full video on skills

85:09

you can check it out up here

85:10

Then we have hooks

85:11

which are guardrails. So this is giving it

85:13

rules as to saying don't touch this file

85:15

It's sort of like the security guard

85:17

right? Which is, hey, before this happens

85:19

make sure this is in place

85:20

After this happens

85:21

make sure that is in place

85:22

Different rules

85:23

I've also made a full video on hooks

85:25

you can check it up here

85:25

And then we have the MCP

85:27

So the MCP gives us new abilities

85:29

right? So we say now

85:30

you can use curl. Now you're able to

85:31

connect to Notion and you know exactly what you should do

85:34

right? And this is a new pair of

85:35

hands that we give to Cloud Code

85:37

which is why I say that it completely

85:38

changed the way that I use Cloud Code

85:39

because it allows me to build things that

85:41

previously

85:42

either taken a lot more time to set

85:43

up and things that previously were not possible

85:46

So again, skills are instructions, hook are guardrails

85:48

and MCP are new abilities

85:50

They work together, not again

85:51

Against each other. So they're different things

85:52

right? But when you combine all those three things together

85:55

that's when Cloud Code starts to become really

85:57

really

85:57

really powerful. And so here I am on

85:58

my main YouTube workspace

85:59

which controls my main YouTube content generation system

86:03

And inside this workspace

86:04

you can see that I have a file

86:05

called.mcp.json. So

86:08

.mcp.json is going to be

86:10

the file that we use

86:11

to add the MCPs. As you can see here

86:13

if I zoom out a little bit

86:14

here we have an MCP server

86:16

called

86:16

Chrome DevTools. And so what does this do?

86:18

This allows us to be able to use

86:20

the web and access different things outside of Cloud Code

86:23

right? This doesn't mean scrape a website

86:24

This means going inside a website and taking a screenshot

86:27

for example. Now, if you look at this

86:29

you're like, what the hell am I looking at?

86:31

That's okay

86:31

because you don't really have to understand the

86:33

code. You just have to be able to

86:34

talk to Cloud Code to ask it what you want

86:36

right? As you saw

86:37

to set up the MCP

86:38

we have to make a file

86:39

But before we do that

86:40

let me show you why we need it

86:41

in the first place. Hey

86:42

can you go to apple.com?

86:44

Can you take a screenshot of the

86:45

main page?

86:46

And can you

86:47

then give it to me here?

86:48

So in theory

86:49

this should be able to then go to

86:50

apple

86:51

Take a screenshot of the page and then

86:52

bring it back here as an image

86:53

As you see though

86:54

it says I can't take screenshots of website

86:56

I don't have a browser automation or screen

86:58

capture tools available

86:59

I can only fetch the page content

87:01

read screenshot files that already exist on your

87:02

machine. So this is why we use MCP

87:05

And so what I can do here is

87:06

I can go to file

87:07

I can write.mcp.json

87:09

and

87:10

inside here I'm going to paste the code

87:12

that we had before. File

87:14

save

87:15

And by the way

87:15

I'll show you how to get this code as well

87:17

so don't worry. Once we have this here

87:18

let me ask you the exact same question

87:20

in a new tab

87:21

and you'll see how now it's going to go to Apple

87:23

take a screenshot, and then bring it back

87:25

So you can see here it's saying let

87:26

me navigate to apple.com

87:27

and take a screenshot for you

87:29

I can say yes

87:30

let me do bypass permissions

87:31

so it doesn't ask me

87:32

It said it successfully navigated to apple.com

87:35

it took a screenshot

87:36

and now we have the screenshot of the page

87:37

which is this one here

87:38

Okay? And so just in a matter of seconds

87:40

we were able to unlock a new feature

87:41

in Cloud Code

87:42

just by adding this MCP.json file here

87:45

You can start to see the different applications

87:47

that come into this

87:48

This is a very

87:48

very simple example. One of the most used

87:50

examples is Chrome DevTools

87:52

which is the ability to go to websites

87:53

take screenshots, and access the web

87:55

but we also have other ones as well

87:56

And by the way

87:57

just so you know

87:58

it actually opened up a new Chrome tab here

88:00

which is how it was able to take

88:01

the screenshot

88:02

and then bring it back to the workspace itself

88:04

All right

88:04

so to find the snippets of code that

88:05

I just showed you

88:06

which allow Cloud Code to have new features

88:08

plugins, we have to go to the mcpservers.org

88:11

Now, if you're aware of skills

88:12

which is the ability for us to

88:14

add instructions to Cloud Code and do different things

88:17

skills also have something similar

88:18

So if you go to skills dot SMP dot com

88:22

we have what we call a workspace full

88:23

of skills that we can copy and paste

88:25

into Cloud Code. So this right here is

88:26

Amazon for MCP servers

88:28

And so as you can see

88:29

we have the Chrome DevTool

88:30

MCP, which is

88:31

an option that we can use

88:32

We have Firecrawl, we have Superbase, we have Playwright, we have Xcode by MCP

88:36

and we have a lot more as well

88:38

right? Across official, search, web scraping, communication, productivity, and so on, right?

88:43

So let's say I go to featured

88:45

let's say we go to the Chrome DevTool

88:46

that we had before

88:47

If I go here

88:48

I'll be able to see the name

88:49

the

88:50

description as well. So this is the

88:52

tool that allows your coding assistant

88:54

such as Gemini, Cloud Cursor, or Copilot

88:56

control and inspect a live Chrome browser

88:59

It allows to control

89:00

Chrome

89:00

and go to a new browser and do

89:02

whatever it is that you want to do

89:03

It looks at the key features

89:04

the disclaimers, usage statistics, update checks, requirements

89:08

and we have the code itself

89:10

So here we can copy this code

89:12

and that's the code

89:13

that you'll be able to paste right here

89:15

Go here, you can paste it

89:17

And then make sure you press command S

89:19

then it pushes everything back

89:20

That's why it kind of changed the structure

89:22

But it's the exact same code

89:23

Now, bear in mind, with MCPs

89:24

not all of them are vetted

89:25

which means that they weren't completely verified

89:28

so

89:28

user judgment as to what it is that

89:30

you can actually be using

89:30

If I see that this is in the featured MCPs

89:33

then it's probably something that

89:35

I am going to be able to use

89:36

because it's trusted. But let's say I go to cloud service

89:39

right, and find a random one

89:40

let's say AKS MCP

89:42

I don't really know

89:43

whether this is good unless I look at the stars here

89:46

the GitHub stars. The higher the stars

89:47

the better it is

89:48

If it has two stars

89:49

then I'm most probably not going to be

89:50

able to use this

89:51

because I don't know the impact it will

89:52

have

89:53

on my cloud code and

89:54

the things that it does

89:55

And if you ever copy some code and

89:56

you have no clue what it actually means

89:58

you can simply just say

89:59

what is the MCP.json that I have right

90:01

now? What does it do?

90:02

It tells you

90:02

and then I also told it to give

90:03

it to me in one sentence and make it very

90:05

very simple to understand. And here it says

90:07

that it lets me control your Chrome browser

90:08

clicking, typing, taking screenshots, and inspecting things

90:11

so it can help you test and debug

90:12

websites directly from this chat

90:14

Now, one use case for this specifically is

90:16

let's say you are Making a landing page

90:18

And let's say you wanted to look at

90:19

the top landing pages from five different websites

90:22

And you wanted to take inspiration as to

90:24

the top header

90:25

that was on the landing page specifically

90:26

So here we have Apollo

90:28

we have Mail.so, we have Email List Verify

90:31

which are all different SaaS

90:32

so softwares, that verify different emails or give you some data

90:35

right? And let's say I wanted to make

90:36

a new tool

90:37

and make a landing page similar to these

90:39

ones. And so here then I can say

90:41

can you go to each of these websites

90:42

can you take a screenshot of the landing page

90:44

and then bring it back and make a

90:45

new folder under our workspace

90:47

name it screenshots of landing pages

90:49

and leave the pictures there

90:51

And I can press go

90:51

I can put this as bypass permissions

90:53

Okay, so now it's going inside Apollo.io

90:55

It's going to Mail.so

90:57

Email List Verify, the next one is ZeroBounce

90:59

then we have Verifylia.com

91:01

and then finally we have Hunter.io

91:03

As you can see

91:03

it just finished everything

91:04

It made the folder of screenshots of landing pages

91:07

and here we have Apollo

91:09

Email List Verify, Hunter, Mail.so

91:11

Verifylia, and ZeroBounce, right?

91:14

And so all of these could be used

91:15

then when you're making your landing page

91:17

Inspiration. You can say

91:18

hey, look at these images on landing pages

91:20

kind of emulate their style

91:22

and make me a new landing page

91:24

and call it XYZ name

91:25

So another one that I use often

91:26

is the one of Superbase

91:28

So Superbase is a software that allows us

91:30

to be able to store data

91:31

It's sort of like a Google Sheets on steroids

91:33

right? It allows us to add data to

91:35

a table that we can then fetch

91:37

using its API. So you throw in a question

91:38

all the information is stored in Superbase

91:40

and Superbase gives you the answer

91:41

So here

91:42

what I can do

91:42

is I can just configure this here

91:44

So I can copy this

91:45

which again usually is in the bottom

91:47

Right? Here what I can see is the

91:48

name

91:50

We can see the demo of this

91:52

of how it works

91:52

Then it's giving me the actual setup

91:54

What I would have to do here

91:55

is I would have to go

91:57

and copy this

91:58

You can also look at the setup documentation here

92:00

which goes through everything that you have to do

92:02

but it's pretty much the same as the

92:03

one we had here. So I can go here

92:04

and by the way

92:05

you don't actually have to make a new file

92:07

you can use the exact same one

92:08

So now we have this

92:09

and then I can say

92:11

can you add this to the MCP?

92:12

I can give it the new MCP

92:13

of Superbase. When I press go

92:16

Put bypass permissions

92:17

it will now add the MCP to our new MCP.json

92:21

which will be here. As you can see

92:23

it's making the change. I can say yes

92:24

And now we have the Chrome DevTools

92:26

we have Superbase

92:27

and we have more and more and more

92:28

and more as we start to add new

92:29

things to our file. Okay

92:31

let's say I wanted to have this document right here

92:33

which is a fake SOP

92:34

for an e-commerce store

92:36

I wanted to add it to my Superbase

92:38

so that we can store all of this in a table

92:40

and then we can then fetch

92:41

can ask questions about the thing itself

92:43

like a chatbot. I can go here

92:44

I can download as a PDF

92:46

and I can upload it right here

92:48

And then I can say

92:49

hey

92:49

I uploaded the

92:50

PDF

92:51

of the e-commerce store SOP and knowledge base

92:53

Can you please now make a vector database instead of Superbase

92:56

and can you add all the information there

92:59

vectorize it, and then, yeah, make a new table

93:01

and call it SOP. I'm gonna make this plan mode

93:03

so plans

93:04

What it's now going to do

93:05

is it's then going to use the MCP

93:07

of Superbase because it has to access the

93:09

software itself. As you can see

93:11

it's reading the mcp.json file because that's where

93:13

we have the actual MCP itself

93:15

Now it's giving me a plan for creating

93:17

the local vector database in Superbase

93:19

Sure

93:20

Looks all good

93:21

I understand half of it

93:23

but Not even halfway

93:24

but it's saying

93:25

all the things that I wanted to do

93:26

I can press yes

93:27

auto accept

93:28

and then I can put papa's permissions to

93:30

let it run on its own

93:31

So it's going to take the PDF

93:32

we call it chunk it

93:33

so it's going to chunk

93:34

meaning that it's going to take the actual document

93:36

it's going to split it up into different parts

93:38

and then put it all into the

93:40

vector database itself. And the vector database

93:41

by the way

93:42

is just

93:43

a Google sheet which stores all the information

93:45

So it just pulled up this page right here

93:47

where I have to authorize the MCP connection

93:49

So MCP client connection

93:51

authorize API access, I can say yes

93:54

choose organization, authorize cloud code, authentication successful

93:57

I said done, you found my project

93:59

which takes two minutes to set up

94:01

make an account, and then pretty much make an organization

94:03

and that's it. And now it's going to

94:05

create the table in the Superbase account

94:07

As you can see

94:08

it made a table called SOPs

94:09

and we should be able to see the

94:11

information starting to come in

94:12

And as you can see now

94:13

we have the information that we had on the Google Doc

94:15

we have it here in the Superbase itself

94:18

right?

94:19

And the way that this works

94:20

again, is that we have each text

94:22

which is an embedding

94:23

an embedding, Just a number

94:24

and when we say query the vector database

94:26

when we ask questions

94:27

it matches the

94:28

numbers of the questions or the letters that we

94:30

that we actually give it to the numbers here

94:32

and it gives us the answer

94:33

right?

94:33

But essentially what it did is that it

94:35

used the MCP

94:36

of Superbase itself to then send all the

94:38

information here

94:39

without us having to set up anything apart

94:41

from that MCP code itself

94:42

Now, although MCPs are great

94:44

there is one fallback

94:45

which is the cost

94:46

And so if I go here to slash context

94:50

I'll now be able to see

94:51

how many tokens the MCPs actually take

94:54

So here I can see that I'm using Cloud Opus 4.6

94:57

I'm using 15.6k out of 1 million tokens

94:59

so 2%, and only two MCPs use 10,800 tokens

95:04

which is quite a big amount considering the

95:06

fact that we only added two

95:08

And you can see here that the MCP tools

95:10

right, so each tool

95:10

which means that each type of action that it can do

95:14

is, requires tokens to run

95:16

And so we have click

95:16

which is 163, close page, 112, drag, 155, emulate, and so on

95:21

right? And it keeps them going and going

95:23

So one important thing is that the more MCPs you add

95:25

the more money you have to spend on

95:27

tokens

95:28

So make sure to not use

95:30

15 or 20 MCPs because

95:31

they will take up a lot of your

95:33

tokens

95:34

and context. Just use the top two to

95:35

three that I showed you

95:37

and make those your main priority ones

95:39

whenever you're using Cloud Code

95:40

And so with the fact that you are

95:41

spending a lot of tokens with the MCPs itself

95:43

usually what we do is that we install the MCPs here

95:46

right? So let's say we install Chrome DevTool

95:48

or let's say we install the Superbase

95:49

Check if it works

95:50

which means that you test it

95:51

see if it did the right thing

95:53

and then you turn that

95:54

MCP itself into a skill

95:56

because skills require less tokens

95:57

And so with the healthy setup

95:59

you typically have a system prompt

96:00

which is the prompt of the actual AI itself

96:02

Cloud Code. Then you have built-in tools

96:05

and then you have MCP tools

96:06

You want to have a few

96:07

and then you have your conversation

96:08

which means it's a lot of room

96:09

Because the more you speak to Cloud Code

96:11

the more the tokens

96:12

or what we call the context window

96:14

increases

96:14

which means that

96:15

you can't be having a conversation with Cloud Code forever

96:17

because the tokens, the space of conversation

96:20

ends

96:21

right? And so

96:21

you want to have as much space as

96:23

possible for the conversation

96:24

instead of having the system prompt

96:26

built-in tools, and having 20 different MCP servers

96:28

which take up so much tokens and so

96:30

much space and context

96:31

where

96:32

to the point where you can't actually have

96:33

a proper conversation with Cloud Code before the

96:35

tokens runs out. So you want to be

96:36

as efficient as possible

96:38

with your tokens that you are using in

96:39

Cloud Code. And by the way

96:40

if you want all the resources

96:41

that I'm going to be talking about in this video

96:44

plus being part of a community of over

96:45

9,000 people who are actively building with AI

96:48

then check out the second link down

96:49

Hello. All right

96:55

so congrats on finishing module two

96:57

At this point, you have fully customized ClockCode

97:00

You have custom skills

97:01

you have automated hooks

97:02

and it's connected to every single tool through

97:04

the use of MCP

97:06

So now you have a pretty powerful setup

97:08

But if there's one thing I learned after

97:09

working with ClockCode for hundreds of hours is

97:11

that it's not so much about your setup or your tools

97:14

but it's about how you speak to ClockCode

97:16

which is the prompting. And that's exactly what

97:18

the next module is all about

97:19

which is the prompting strategies

97:21

that will 10x the quality of everything that

97:23

you do with ClockCode. Let's dive in

97:24

Okay, so the first hack is called splitting prompts

97:27

Now

97:27

one of the biggest mistakes that I see

97:28

when people use ClockCode is them adding

97:31

huge and humongous prompts with a ton of

97:33

information and a ton of tasks all in the same prompt

97:35

and then expecting ClockCode to give them the

97:37

output of day one. An example of this

97:38

could be them saying

97:39

build me a CRM with auth

97:41

database, API, dashboard, emails, and deploy it

97:43

which are basically seven different tasks

97:45

right

97:45

or four different tasks within the same kind

97:47

of major thing that they want to build

97:48

And there's a few problems with This

97:49

the first one being that there's a lot

97:50

of context or

97:51

tokens used

97:53

in that same session that we're using Cloud

97:54

Code. The second thing is that the output

97:56

is not going to be the best

97:57

just because you're feeding it and you're telling

97:59

it 20 different things at once

98:00

The same way that you tell an employee

98:01

hey, can you build me a full software with this

98:03

this, this, and that

98:04

without giving the person any pure structure

98:05

And the issue here

98:06

even when I started

98:07

was that I thought Cloud Code would build anything

98:09

right? It could build anything and anything if

98:11

I give it one prompt and that's it

98:12

But it's much better to split prompts

98:14

and take a more structured approach

98:15

So instead of putting this prompt right here

98:16

with seven different tasks

98:18

we split the tasks up

98:19

right? And we talk to Cloud Code task

98:21

by task. So we look at this and we say

98:22

okay

98:23

the first task is setting up the project

98:25

Then

98:25

it will set up the project and then

98:26

we review whether it did a good job

98:28

or not. If it did a good job

98:29

then we go to the next thing

98:30

If it didn't do a good job

98:31

then we go back and fix it

98:33

Then we go to task two

98:33

which is knowledge base

98:34

and then we review if it worked

98:35

If not, then we look back

98:36

And we do this for the other tasks

98:38

as well. And the reason this works a

98:39

lot better is because when we have to fix task four

98:42

we don't have to fix the whole thing

98:43

because we reviewed each one and it worked very

98:45

very well. And so we go task by task

98:46

making sure that everything works until the end

98:48

Now

98:48

is it going to take more time to

98:49

Clock code at the start

98:50

yes

98:51

but actually you

98:52

spend more time fixing this

98:53

after it didn't do a good job

98:54

rather than setting things up properly from

98:56

from the start. So to make this more practical

98:58

I can give you like an actual prompt

98:59

that I built. So you can see the

99:00

bad approach and then the good approach that

99:01

we take. So on the one hand

99:03

we have a prompt that says

99:04

build me a full customer support chatbot with Next.js frontend

99:07

a knowledge base that uses embeddings

99:08

escalation logic that routes tickets to Slack

99:10

an admin dashboard analytics

99:11

user authentication with OAuth

99:13

and deploy the whole thing onto Vercel

99:14

Use Tailwind for styling and make it responsive

99:17

Now

99:17

I'm not going to run this for the

99:18

sake of not having to wait three hours

99:20

but this right here

99:21

it will overwhelm clock code with so much

99:23

context and so many things to do that the output

99:26

don't get me wrong

99:27

it will be okay

99:28

but it won't be as good

99:30

as if you split things up step by step

99:31

right? And so what we could do here

99:33

is we have the first prompt

99:35

which would be this

99:36

let's say, prompt one

99:38

set up the Next.js project with the TypeScript and Tailwind

99:40

create the basic folder structure with components

99:42

and we give it the first part

99:44

So when prompt one is finished

99:45

it did a good job

99:46

then we go to prompt two

99:47

right here

99:48

Then once prompt two is finished

99:49

so this is like The knowledge base

99:50

it goes to prompt three

99:52

and then we keep going until prompt four and prompt five

99:55

right? So prompt four will be this

99:56

So all we did here is took this prompt right here

99:58

which has Next.js front end and knowledge base

100:00

escalation logic, it has admin dashboard

100:03

it has auth, it has deploying, right?

100:05

So Next.js front end will go here

100:06

then we had knowledge base will go here

100:08

then we had escalation logic will go here

100:10

then we had dashboard which will go here

100:12

and then finally we'll have another one

100:14

which is deploying the whole thing

100:15

And so you can see that it's kind

100:16

of like a spider web

100:17

and the reason why we do this is

100:18

because you take a more structured approach

100:20

right? And so I'd walk you through how

100:21

you review it and so on

100:22

but essentially you would run this

100:23

right? And don't worry about all these terms

100:24

because they don't mean much to me either

100:26

but it's more the structure that we have to understand

100:28

right? And so you have something simple with seven different tasks

100:31

you split them up into different prompts

100:32

prompt one, prompt two until prompt seven

100:33

you review them individually

100:35

to make sure that when you get the output

100:36

it's going to be correct

100:37

you don't have to pre-prompt Claude all over

100:39

again and fix it and debug it and spend two

100:41

three hours asking it why it did the

100:42

wrong thing. The next hack is using plan

100:44

mode. Now I know plan mode is not something revolutionary

100:46

but it's a big misconception as to what

100:48

it actually does and when it's

100:50

So whenever we have a new task in Cloud Code

100:51

we ask ourselves whether it's hard or not

100:53

hard. If it's hard

100:54

so if it requires different steps

100:55

then we use plan mode

100:56

then we review the plan

100:57

and then we approve it

100:58

and then we execute the actual thing to

100:59

give us the output that we want

101:00

right? An example of this could be an

101:01

API with five endpoints

101:02

or when you're building a full workflow

101:04

and I'll show you exactly what that looks

101:05

like. Because it has different steps to it

101:06

and you want it to plan accordingly

101:08

take the time to show it

101:09

to think through how it's going to build

101:10

it before it builds it

101:11

On the other hand

101:12

if you have something very easy

101:13

like fixing a typo or changing some sort

101:15

of color in a button that it made in a software

101:16

then you can just execute directly without needing the plan mode

101:19

because there's not much planning to it

101:21

It's just a very

101:21

very easy task. All right

101:22

so here I'm on my main hub in Cloud Code

101:24

which is called the YouTube workspace

101:26

I automate pretty much 30% of everything that

101:27

I do on a day-to-day basis

101:29

with like YouTube. I would

101:30

even go as far as saying like 40%

101:32

Now here under dot cloud

101:33

and by the way

101:34

if you're new to Cloud Code

101:34

if you have no idea what I'm talking about

101:36

if this looks very confusing to you

101:37

watch this video up here

101:38

It will give you a much better idea

101:39

But if you understand skills

101:40

which I'll walk you through in the last hack as well

101:43

it's got a draw diagram

101:44

So that skill right there allows me to

101:45

be able to make these diagrams automatically

101:47

To break down the topic for you guys to understand

101:49

I You break it down in a visual way

101:51

but also make it practical

101:52

So you see how I'm showing you the diagram

101:54

plus also how you can use

101:56

the actual concept here. And so in this case

101:58

let's say I wanted to make a video

101:59

about anti-gravity versus cloud code

102:01

By the way, if you want the video

102:02

let me know in the comments down below

102:03

if you're interested. If not

102:04

also let me know

102:05

But

102:06

if I wanted to do a video on

102:07

anti-gravity versus cloud code

102:08

well, I would want to basically

102:10

first of all

102:10

explain

102:11

what anti-gravity is and how cloud code is

102:13

different from anti-gravity. And so

102:14

I would want to make maybe three diagrams

102:16

I'd say, using Excalidraw, which is this platform right here

102:18

which allows me to explain my concept

102:21

the differentiation between anti-gravity and cloud code

102:23

Okay

102:23

So I told it to research anti-gravity

102:25

I told it to research cloud code

102:27

I wanted to look for the differences

102:28

I wanted to make three visuals

102:29

and then give me back the visuals

102:30

Okay. So a task like this has a plan

102:33

right? And so instead of going bypass permissions

102:36

which is basically where we go YOLO mode

102:37

we say, hey, take care of it yourself

102:39

don't ask me for any help

102:39

don't make any plan

102:40

just go after it

102:41

right? We can go

102:42

and use plan mode

102:44

to make a plan first before

102:46

it gives us the output

102:47

Okay, so now I'm going to press run

102:48

and you know it's on plan mode

102:49

When it's blue right here

102:51

bypass permissions is going to be red

102:53

Now it's saying that it's going to research

102:54

both products in parallel to understand what we're working with

102:56

then plan the diagrams. Okay

102:57

so that's how it plans through

102:58

that's how it thinks through actually planning the

103:00

the whole concept here. Okay

103:01

there you go. So after a few minutes

103:02

it made the plan

103:03

which is Google's anti-gravity versus cloud code

103:05

three visual comparison diagrams

103:07

It gives me the context

103:08

so Michele wants to create XYZ

103:09

research summary, so the summary that it made

103:12

three key differentiators, so interface

103:13

agent orchestration, verification, and feedback, and deliverables

103:16

which is the three things that we want

103:17

right, with color system, the layout, diagram one, diagram two, diagram three, diagram four

103:21

and the implementation steps

103:22

and the files and verification

103:23

One more reason why I do also like

103:25

this in specific is because when I use plan mode

103:27

I can also add comments

103:28

So I can say interface paradigm

103:30

Paradigm is a very strange word

103:32

make it beginner-friendly. And it was a very stupid comment

103:34

but let's say you add the comment

103:35

When we press approve plan

103:36

then we will approve the plan

103:38

but it will also take the comments that

103:39

we add in the plan as context to

103:41

change whatever it is that we told it

103:42

to change. But also I get to see

103:44

exactly what it came up with

103:45

Because if I didn't like interface paradigm or

103:47

if I didn't like agent orchestration

103:48

I could say, hey, Can you find something else

103:50

right? And we can go back and forth

103:52

Now let's say I like the plan

103:53

everything's good

103:54

Now I can

103:54

press send feedback and keep planning

103:56

So send feedback because we

103:57

we added one comment

103:58

And if you add more comments

103:59

so if you say here

104:00

let's say orchestration, you add, turn this to organization

104:03

turn this word

104:05

So we're just playing around with the names

104:07

I can go here and I can send

104:08

feedback and keep planning. Or

104:09

actually now that I thought of it

104:11

color system, I know anti-gravity is not purple

104:13

so we can do use dark blue

104:16

Cloud code will be not blue

104:19

it will be, use orange

104:23

That's it. Now I can go back

104:24

It will say we added four comments

104:25

It will take those four comments and we

104:26

can send feedback and key plan

104:27

okay? So now it will take the feedback

104:30

it will apply the feedback

104:31

and then give us back another plan

104:32

with the updated feedback. There's a lot of feedback there

104:34

but you get the point here

104:35

It's basically crafting the whole plan before it

104:37

gives it to us. All right

104:38

so here we have the new updated plan

104:40

If I go back here

104:40

we can see that we have dark blue

104:42

orange, we have interface style

104:43

and we have agent organization

104:45

So these things changed. It updated the actual

104:46

plan. And now I can say yes

104:48

not accept. And now instead of making it edit automatically

104:50

I can now put bypass permissions because

104:53

I know that the plan is good

104:54

I know that it's smart enough to be

104:55

able to do whatever I want it to do

104:56

and now it's just about doing it

104:58

All right

104:58

so just finished making the three diagrams

105:00

Now let's look at the output

105:01

So I can look at the first one

105:02

I can go here

105:03

I can import scenes

105:04

So anti-gravity, I believe this is an interface

105:07

Let's open it up

105:08

There we go

105:09

Nuts

105:10

Crazy

105:10

Insane. But here we have the actual

105:12

diagram that it made

105:13

Google Antigravity and Cloud Code

105:14

This is the

105:15

diagram itself to show the differentiation

105:17

in the interface. The next one is agent organization

105:20

which is this one right here

105:21

which is on the left-hand side is Antigravity

105:23

on the right-hand side is Cloud Code

105:24

showing the difference

105:25

And finally, we have the agent verification

105:27

which is this one right here

105:28

So the verification and how they both work

105:30

Now the output was significantly better

105:32

because we took the time to plan things

105:34

ahead. On the other side

105:35

you could have just used bypass permissions

105:36

and it would have given you something

105:37

and I've tried it before

105:38

and the output just wouldn't be as good

105:39

as you'd want it to be

105:40

So if a task requires more thinking

105:42

if it requires a plan

105:43

just make sure to use plan mode

105:45

every single time

105:48

But the difference between a kitchen sink versus

105:49

a clean session is

105:51

the amount of tokens that it uses

105:52

So if you didn't know

105:53

a Cloud Code session

105:54

right, one session means one kind of conversation here

105:56

figure here, I just press this

105:57

it will compact it. I'll explain what that means

105:59

But essentially

106:00

each conversation has X amount of tokens

106:02

So if I go here and I put context

106:05

We get the context usage

106:06

right? So we get how many tokens have

106:07

we used so far. So we have a

106:09

limit of 200,000 tokens

106:11

and we have used about 83,900 tokens

106:14

in one session. So the whole session that we just did

106:16

it used about 83,900

106:18

which is about 42% of each one

106:19

And it also gives us the actual like

106:21

percentage broken down into like what actually takes

106:23

the most amount of tokens

106:24

So the system message is 3.8

106:26

the system tools is 9.3

106:27

the MCP is 5.7

106:29

the system tools is 10.8

106:30

and so on

106:30

right? And so you have

106:32

all the tools broken down and you can

106:33

see everything here. My point is that

106:34

prompts are better if you separate them in different sessions

106:37

right? So instead of fixing a bug

106:39

making a new feature

106:40

refactoring something, or writing a doc all in the same session

106:42

right, even though they have nothing to do with each other

106:44

you would create a session for each one

106:45

And when I say session

106:46

I mean a new session here

106:47

So you press plus

106:48

this is a session. Conversation

106:49

You can call it conversation

106:50

really, not a session. So conversation one

106:52

conversation two. Because the more tokens you use in a session

106:55

the more stupid the clock would becomes

106:57

The more it starts to think less

106:58

and the more it starts to get overwhelmed

106:59

with information

107:00

And so split them up in different conversations

107:02

and then you can start over

107:03

Now

107:03

I call this prompt hack because Sometimes you

107:05

think that the problem is your prompt

107:06

but really it's the session tokens that you're

107:08

using. So right here

107:09

we basically made three diagrams

107:11

for a video that I'm planning to make

107:12

Now we're using about 42% of the actual

107:14

usage. If I keep going with this conversation

107:16

we could do that

107:17

right? But the

107:18

level of

107:19

intelligence of Cloud Code will decrease the more

107:22

you increase the tokens in that conversation

107:24

So if I wanted to deploy another workflow

107:25

let's say common analysis

107:26

it goes to my competitors

107:27

it goes through all their comments

107:28

see what people want

107:29

and it gives me back a report

107:31

Is it better for me to keep going on this session

107:33

or is it better for me to

107:34

go to a new session and start from there?

107:35

Well

107:35

obviously it's better to go to a new

107:36

session and start from there

107:37

because you start from scratch

107:38

and if I go here and I put context

107:41

Then this will now give me the context

107:42

and we're only using about 9% token usage

107:44

So now it has a fresh mind

107:46

and it can start doing the tasks that

107:47

you want it to do

107:48

without being overwhelmed with all the information from

107:49

the previous conversation that we had

107:51

which did something completely different

107:52

The next hack is using cloud.md

107:54

So let's say you wanted to make an app

107:55

some sort of software app

107:56

okay? And you wanted the app to use TypeScript

107:58

you wanted all the words to follow camel case

108:01

you wanted to use Tailwind

108:02

you wanted to use Superbase

108:04

and you wanted to use an app router

108:05

not pages

108:06

Now

108:07

don't worry if you're not really understanding what

108:08

all these words mean

108:09

but understand that these are all rules

108:10

right? And so these rules

108:12

you can either tell Cloud Code every single

108:14

time to do all these things

108:15

or you can add them to something that

108:16

we call a cloud.md document

108:18

where you basically build a one-time prompt

108:20

and it will auto-load that prompt every single time in Cloud

108:23

so it has context

108:24

and we don't have to repeat everything over

108:25

and over again. So Cloud already knows

108:26

you just ask. So if you have to

108:28

make a PDF document

108:29

and we wanted the color to be blue

108:30

we wanted the letters to be small

108:32

we wanted the, I guess, the font to be italic font

108:35

right? We can either tell it individually every single time

108:38

or we can add it to a prompt like this

108:39

and then it will use that every single

108:41

Time without us having to repeat ourselves

108:43

Now, lucky for you

108:43

I made a skill in Cloud Code called the cloud.md generator

108:47

and this includes a pull prompt on how

108:48

to make the best cloud.md document

108:50

plus it gives you the example structure of one of them

108:52

for example, here, right?

108:54

So all you have to do is download the skill

108:55

which you can find in the second link

108:56

down below. And here we can say

108:58

hey, I want to create a cloud.md prompt for my project

109:01

for my app that I'm building

109:02

I just want you to ask me any

109:03

questions of what

109:04

you need from me to be able to

109:05

make this document. And we can press go

109:06

So now what it's going to do is

109:07

it's going to look at the skill of

109:08

all the requirements that it needs to be

109:09

able to make the cloud.md document

109:11

and then we're going to start working on

109:12

it. So as you can see here

109:13

it says

109:13

I have found the cloud.md generator skill

109:15

Now let's walk through the process

109:17

So now it's going to ask me where

109:18

I should save the cloud.md file

109:19

I can press, recommended, it's fine

109:21

And then tell me about your project

109:22

So what is the product name?

109:23

What does it do? And who is this

109:24

for?

109:24

I can press other and I can start

109:26

talking to it. So let's say here I say

109:27

okay, so I want to build a website

109:29

So this is for a website and the

109:30

website needs to be for my company

109:31

GM Solutions

109:33

we're an AI automation agency that helps

109:35

marketing agencies to be able to build

109:37

their backend delivery systems. And

109:39

yeah, that's what, that's what the actual thing

109:41

So you give a context

109:42

and then the text that I will use is

109:44

I'm not too sure

109:45

Whatever you think is the best for build a website

109:47

let's just do that. It can submit the answer

109:49

and now it will start thinking about what

109:50

to put in the Cloud.md document

109:51

Okay, so it's not Jam Solutions

109:53

it's Jam. But now when it makes the Cloud.md document

109:55

it will be an overview

109:56

of what it is

109:58

that we want to build

109:59

For your Jam Solutions website

110:00

what kind of pages

110:01

features do you want? Landing plus services

110:03

keep it simple, keep it simple

110:06

Actually, for conventions, actually, let's keep it simple

110:09

And rules is going to be any hard

110:10

rules Cloud must always or never do in this project

110:13

I have specific rules

110:14

There you go. And now it's most likely

110:15

going to ask me for rules

110:16

Because all these rules

110:17

I don't want to repeat myself every single

110:19

time I use Cloud Code

110:20

I want Cloud Code to use

110:22

the Cloud.md document when it's actually building the

110:24

thing that I want it to build

110:25

What are your hard rules?

110:26

Mobile first design, no inline styles

110:28

There you go

110:29

Submit answers

110:30

And also other, my company name is JM Solutions

110:34

not JAM

110:36

Okay

110:36

so here we have a plan to create

110:37

the cloud.md document. We have the context

110:39

project with the James Solutions

110:41

AI Automation Agency website

110:42

helping marketing agencies to build backend delivery systems

110:45

The tech stack is going to be this

110:46

pages, conventions, file structure, rules, actions, verification

110:51

and we're good to go

110:51

Yes, and auto-accept. All right

110:53

so here it made the cloud.md document

110:55

If I open it up

110:56

I can see that we have a full prompt

110:58

which is the one that we had before

110:59

So these are information

111:01

basically a summary

111:02

of what it is that we want to

111:03

build. And this is great for us because

111:04

every single time that we have to go

111:05

out of cloud code and come back in

111:07

we don't have to re-explain to cloud code

111:08

what we want. We can just say

111:10

it knows

111:11

that every single time that we start talking to it

111:12

it will always look at the cloud.md document

111:14

it will get context

111:15

and then only it will respond to us

111:17

Sort of like an employee having context of

111:19

what it is that it has to do

111:20

every day. I mean

111:20

imagine you give the AI the task to build a website

111:22

you give it all the different features and

111:24

what it needs to do

111:25

then you go and close cloud code

111:26

you come back to it later

111:28

and then you tell it

111:29

hey, keep going, and then it's going to ask you

111:30

keep going with what? What it is you

111:31

want me to build? And you have to

111:32

re-explain everything again. That's what this prompt is

111:34

for. The last prompt hack is Using skills

111:36

So we call it reusable prompts

111:38

So initially, we had the manual way

111:39

which is same effort every time

111:41

So we had session one

111:42

where we write a full prompt

111:43

Session two, we write another full prompt

111:45

And we keep going session three

111:46

session four, and session five

111:47

And each session will take about five minutes

111:49

and we take about 25 minutes

111:50

that we wasted writing every single prompt manually

111:53

The second way is using skills

111:55

So we write the skill once

111:56

and we use it forever

111:57

So if you remember before

111:58

I showed you how

111:59

I made

112:00

a Excalibur diagram using what we call skills

112:02

The skills are preset

112:04

instructions that we give Cloud Code

112:06

to perform a specific thing

112:07

And so all we have to do is

112:08

create a skill one time

112:09

which will take about five to ten minutes

112:10

and we also have a skill to create the skill

112:12

which is

112:13

awesome. And so whenever you want to do a specific thing

112:15

you can always create a skill

112:16

and then call the skill whenever you want

112:18

to do that specific thing whenever you want

112:20

And so let's say we had the Excalibur

112:21

diagram thing. So inside here

112:23

we have a prompt. This is a prompt

112:24

that explains how to make the Excalibur diagrams

112:27

So this is the full prompt

112:28

Now

112:28

I could either copy and paste this prompt

112:30

every single time into Cloud Code into a new session

112:32

or I could always just

112:34

save it here under the skill

112:36

And it could always read this

112:38

when it needs to make the actual thing

112:39

Now this is great for us because it

112:41

saves me so much time whenever I want

112:42

to make any of these

112:43

to be honest. So for example

112:44

if I go back here and I go to the command

112:46

I can see here that I can call the CloudMD generator

112:50

I can call the common analysis

112:51

I can call the comment analysis report

112:52

competitor analysis

112:53

which are all skills which perform different things

112:56

Now I've covered skills step by step in my colloquial tutorial

112:58

so if this is the first time that

112:59

you hear about this

113:00

just check out this video up here

113:01

it will give you much more context

113:03

But if you already know what skills are

113:04

then

113:04

let's actually build that step by step and

113:06

show you how it works

113:07

Now let's say I wanted to go on

113:08

YouTube every day

113:09

and check the amount of views that I

113:10

have for the videos. Now I could either

113:12

prompt colloquial to do this every day

113:13

or whenever I want to

113:14

And by the way

113:15

before we build the skills

113:16

you can find all of them at aitmpl.com

113:19

These are preset skills

113:21

where we have the skill creator

113:22

which is a skill that makes skills

113:23

we have the senior front end

113:24

we have the UI design

113:25

we have all these different skills that do

113:27

different things. But sometimes I want to make my own skill

113:29

which is my own kind of task

113:30

And so here I can say

113:31

hey, I want to be able to make a skill

113:34

and the skill needs to be

113:36

A skill that goes on the internet every time

113:38

and it finds the top news

113:40

or top features that came out about Cloud Code

113:42

then

113:43

it goes on YouTube and it checks the

113:44

highest viewed video on Cloud Code in the moment

113:47

and then it gives me

113:48

content ideas

113:49

that I could be using for my channel

113:50

for Cloud Code

113:51

Make a plan for it first

113:52

and then we can

113:54

go ahead and make a skill

113:55

So this is the initial thing that we

113:56

given

113:56

I can just press go

113:57

and now it will start making the plan

113:59

for the skill that it needs to do

114:00

Alright, so as you can see here

114:01

it made the plan for the skill that it's gonna make

114:03

and it's gonna be called Cloud Code Content

114:04

Ideas. So what it does is that it

114:05

searches the internet for latest Cloud Code news

114:07

features

114:08

it searches YouTube for the highest viewed recent Cloud Code videos

114:11

it then generates 10 to 15 ranked content

114:13

ideas across reference against Michaelis Content Pillars

114:15

and then it outputs a Markdown report plus

114:17

style PDF

114:18

to outputs

114:19

So now it's gonna be making all the

114:20

skills for that specific thing

114:22

and then we can see phase one

114:24

phase two, phase three, phase four, phase five, phase six

114:27

and that's it. So from a single prompt

114:29

it made this full plan of execution

114:31

that we can now simply press yes

114:32

no, or accept

114:33

and I'm gonna press bypass permissions

114:35

Everything on its own. So as you can see here

114:37

it's going through the to-dos

114:38

tasks that it needs to make

114:39

and then it's adding the folder called cloud code content ideas

114:43

which is a folder where it's going to

114:44

store our skill that we can always call

114:46

when we want to

114:47

to use it. Right

114:47

so as you can see here

114:48

it finished making the skill

114:49

If I go to cloud code content ideas

114:51

have the references

114:52

so this will be the report in PDF

114:54

that it needs to give us

114:55

which is the output

114:56

Then the scripts, which is the actual Python scripts

114:58

it's kind of the

114:59

the automation that it needs to execute

115:01

generate PDF, and search YouTube videos

115:03

and then it has the skill.md

115:05

which is a main document that explains to

115:06

it exactly what we want to do

115:08

So, with that said, I can Go here

115:12

and I can say slash cloud content ideas

115:16

It should also make a command

115:17

and the command allows us to be able to

115:19

to just find it easily

115:20

So I can press cloud code content ideas

115:21

I can put this in plan mode

115:23

and now it will start looking at all

115:24

the different skills here

115:26

and it will go through each skill.md right here

115:28

and it will know

115:29

that it needs to use the cloud code

115:30

content ideas because

115:32

it's associated with that command in specific

115:34

And so what it will do here is

115:35

it will start calling the cloud code content ideas

115:37

and it will start executing everything that it

115:38

needs to do. It says I have all the context needed

115:41

let me write the plan and exit plan

115:42

mode so we can execute

115:43

All right

115:43

it made the plan of what it needs to do

115:45

all the steps, the key files, verification

115:47

We can go and press yes and auto accept

115:50

and we can put this on bypass permissions

115:52

So the idea behind this is that whenever

115:54

now we have to go back to cloud code

115:55

instead of prompting it again saying

115:56

hey, can you go on the web

115:57

can you search this

115:58

can you search that

115:59

we can simply just press slash

116:00

and call the cloud code content ideas

116:02

and it will execute that thing that we made previously

116:05

which again will take us 10 minutes at the start

116:07

but it will take us three seconds to

116:08

be able to set this up

116:09

or to call the tool

116:11

Whenever we have to run it

116:12

All right

116:13

so

116:13

now it gave us the top five content

116:14

ideas. It finished running the actual thing

116:17

agent teams for agency operations

116:18

Cloud Code HTTP hooks

116:20

why don't delete my cloud.md

116:22

skills, build a business command center

116:24

and Cloud Code for non-coders

116:25

And if I go to outputs

116:26

I can see that it also made

116:28

let's go to here

116:29

It also made a report in PDF format

116:31

And the PDF looks like this

116:32

So we have news and updates

116:34

the loop command, which came out in Cloud Code

116:36

the Cloud API skill

116:37

the effort levels in Authory Think

116:38

auto memory, security, agent teams, enterprise AI agent ROI

116:43

and we have

116:43

the top YouTube videos latest by 30 days

116:46

So Nick Straub's crushing it

116:47

Cloud Code work, there we go

116:48

Dan Martell, he's a legend

116:50

Brian Cassell

116:51

and we have the other ones as well

116:52

And it keeps on going and going and

116:53

going. And we have the patterns observed

116:55

content ideas, and so on

116:56

So it took us like 10 minutes to

116:57

be able to make this whole skill

116:58

but

116:59

now it's going to save us a lot

117:00

more time when we have to rerun it

117:01

again. Because now when I go to a new Cloud session

117:03

I can simply go and press

117:04

the Cloud Code content ideas

117:07

And when I press this

117:08

it runs it over and over again

117:09

So instead of writing the prompt over and over again

117:11

we have At one time

117:12

and we save ourselves an eternity

117:14

every time we have to run the whole thing

117:15

All right, congrats on finishing module three

117:22

which is the prompting guide

117:23

Now you know how to

117:24

actually communicate with Cloud Code to make it

117:26

do things that you want it to do in a very

117:28

very effective way. And that matters because

117:30

every single thing that we're going to cover

117:32

from now on requires that skill

117:34

So now we're going to be moving into agentic workflows

117:36

which is sub-agents, agent teams, workflow hacks

117:38

and the deployment side. And this is where

117:40

you get Cloud Code to stop acting like

117:42

a normal coding agent

117:43

but start becoming the

117:45

system that runs the entire operations for you

117:47

And this is the highest leverage thing

117:49

that you can build with Cloud Code right

117:50

now. So let's dive in

117:51

Okay

117:52

so think of Cloud Code as the business

117:54

owner. Right now what's happening is that you're

117:55

asking Cloud Code to do everything

117:58

Research, write, test, organize, right?

118:00

And that's like running a one-person business forever

118:03

And so you can think of sub-agents

118:04

like hiring different employees

118:06

Each one has their own brain

118:07

each one has their own job description

118:09

their own tools, and each one reports back

118:11

To the main boss

118:12

which is Claude in this case

118:13

right? And so what happens here is that

118:14

you tell Claude a task

118:16

you give it a task of what it has to do

118:19

and then Claude delegates

118:21

which means that it gives the work to the right employee

118:23

to the right agent in this case

118:24

And the employee

118:26

works in their own space and then hands

118:27

back a clear summary to the actual agent

118:30

which then gives it back to you

118:31

So you can think of this as a research assistant

118:33

a copywriter, and a QA employee

118:35

And so these are three different employees

118:37

which are able for us to split up

118:38

the tasks across different people

118:40

so that each one is specialized in their own thing

118:42

and they all report back to the main

118:44

agent here. So now that you know exactly what it is

118:46

how do you actually create sub-agents?

118:48

So there's the first way

118:49

which is using Cloud Code

118:50

and the second way

118:51

which is doing it manually

118:52

And we're going to look at both ways

118:53

Okay, so here I am on Cloud Code

118:55

I made a new folder called sub-agents

118:57

and for us to be able to create the sub-agent

119:00

I can type slash agents

119:02

okay?

119:03

Now, if you're on this view right here

119:05

you will then see a symbol

119:07

which is this one right here

119:08

This symbolizes that to run this exact command to make Agents

119:12

you have to run it through your terminal

119:13

And to run it through your terminal

119:15

got it here, you can go up terminal

119:17

you can press new terminal

119:18

and then you will be able to see

119:19

the terminal here. And I can press cloud to initiate cloud

119:22

to actually make cloud work

119:24

and then I'll have this screen right here

119:25

Let me actually go Maxed out

119:28

and then I have this page right here

119:29

So now we're going to be working with

119:30

this page specifically to make the actual sub-agent

119:33

okay? Now that we're here

119:35

we can press slash agents

119:36

even though the spelling's wrong

119:38

and press enter

119:39

And now it's telling us

119:39

hey, agents

119:40

do you want to create a new agent?

119:42

I can press enter again

119:43

which means yes. It's asking me the location

119:44

so project or personal

119:46

I can do project

119:48

that's fine. So now it's asking me how

119:49

I want to create the agent

119:50

In this case, I'm going to do generate it with cloud

119:52

and now I have to explain

119:54

what it is that I want

119:55

For the sake of just showing you how

119:56

it actually creates the agent

119:57

I want to keep it very simple

119:59

I want a sub-agent that I can review

120:01

or that can take an idea of a YouTube video

120:04

and then it can spin up 10 different YouTube ideas

120:08

give me the 10 ideas plus

120:10

hooks that I can use for that specific video

120:11

Now

120:12

I'm using just a very simple content creation

120:14

kind of use case

120:14

but at the end of the video

120:15

I'll also show you

120:17

three different use cases

120:18

where you can actually apply this inside a real business

120:20

right? Now that we have this here

120:21

I can press enter

120:22

and now it's going to generate the agent from description

120:25

and we should be able to see

120:26

the files in the folder

120:27

Being created here

120:28

which is basically how it stores the actual agent itself

120:30

Okay, so now it's asking me to select the tools

120:32

and I think of this as like the

120:34

level of accessibility that you give to the

120:36

agent itself. So for example

120:37

do you want it to give it access

120:38

to all the tools

120:39

read-only, edit tools, execution tools, other tools?

120:42

I'm going to press continue

120:43

because I

120:44

can give pretty much access to everything

120:46

It's going to ask me to choose

120:48

the model

120:49

that we're going to use to be able

120:50

to run the sub-agent

120:51

and I'll walk you through like the model

120:52

and which one should you choose in the

120:53

first place

120:54

I'm going to go for Sonnet

120:55

because it says balance performance

120:56

best for most agents. Let's go

120:58

And then it's asking me for a color

120:59

so let's make this nice and pink

121:02

Then it's asking me

121:03

where do I want to configure agent memory

121:05

Keep it very, very simple

121:06

Use product scope

121:08

because then it's going to add it to your actual project

121:09

and then

121:10

now you can see that it made a folder called doclog

121:12

and then I can press S to save

121:15

If I go here

121:16

I can see that it now made the agent memory

121:18

which doesn't really have much in it

121:20

and then it made the agent itself

121:22

So the agent itself

121:23

just so you know

121:23

each sub-agent

121:25

all it is, is a prompt, right?

121:27

The same way that You know skills

121:29

sub-agents are just prompts as well

121:30

And the prompts is a

121:31

list of instructions of what the agent should actually do

121:34

Okay, so now that we've made a sub-agent step by step

121:37

let's go through the structure of the actual sub-agent

121:38

Okay, I'm going to press X here

121:39

I'm going to expand this

121:41

And here you can see you have the name

121:42

description, model, color, memory

121:45

and then we have a whole description of

121:46

the actual

121:47

sub-agent and what it needs to do

121:48

At this point, you have it ready

121:50

go to your chat window

121:51

and then you can press file

121:53

you can press save

121:53

or you can press command S

121:55

and it's saved to your computer as well

121:57

And so what happens is that you'll start

121:58

to have more and more agents

121:59

which is more and more prompts

122:01

right?

122:02

And so you can run agents simultaneously if

122:04

you want. So now you know how to

122:05

make it automatically using Cloud

122:06

But just like everything

122:07

if you don't really understand the fundamentals of

122:09

how it all works

122:10

then

122:10

if you're stuck on something or if Cloud

122:12

doesn't do the right job

122:13

you won't know

122:14

because you don't know what good looks like

122:15

So for us to create the sub-agents

122:18

we first have to create a folder

122:20

called the dot cloud folder

122:21

Now in case you haven't watched the skills video

122:23

check it out up here

122:24

But I use that exact folder itself to

122:26

store skills as well. And so you You have agents

122:28

you have skills, you have agent memory

122:29

you have commands

122:30

you have all these things in the same

122:32

exact folder. And then we have to make

122:33

a folder called agents

122:34

make a folder called agent memory

122:36

which is the memory of the actual agent

122:37

and then we have the actual markdown file

122:40

which is.md

122:41

because it's the way that it's written like this

122:43

with hashtags, with asterisks, to organize the text

122:47

which is the instructions of the actual agent

122:49

So I deleted the previous subagent that we

122:51

had. First things first to make our own

122:52

subagent is make a folder called the

122:54

.cloud folder. And inside this folder

122:57

we can make another folder called agent

123:00

slash memory

123:02

And then inside the.cloud folder as well

123:04

we can put the agents

123:06

And so this looks like the exact same

123:08

thing that we had before

123:08

And then inside the agent here

123:10

we can then press new file

123:12

and call this name of the actual agent

123:14

So let's say we're doing email writer.md

123:18

Okay

123:19

And then you have the actual marathon file

123:22

where you'll be actually running these instructions

123:24

Now

123:24

the instructions for a sub-agent prompt are divided

123:27

into different parts. We have the name

123:30

we have the description

123:30

we have the tools

123:31

the model

123:32

and then we have the actual instructions here

123:33

First things first, we have the name

123:35

which is email writer in this case

123:37

It's what you call this agent

123:38

It's how you'll be able to refer to it later

123:40

Okay, so remember the name

123:41

Then we have the description

123:42

So in this case

123:43

write professional client emails based on the context

123:45

you give it. You tell Claude when to

123:46

use this agent

123:47

Claude reads this

123:48

to decide if it's the right fit

123:49

And the description is the most important part

123:51

because you tell Claude

123:53

when to use this agent

123:54

So when you tell it

123:55

hey, I want to do X

123:56

Y, Z

123:56

Claude will read this exact description to decide

123:59

if this is the right agent

124:01

for your request. Then we have tools

124:02

In this case, we just wrote reader write

124:04

The tools are the permissions

124:05

right? So what this agent is allowed to do

124:07

or it's like onboarding someone into your own project management system

124:10

but not giving them access to everything

124:11

Then we have the model

124:12

which is in this case

124:13

Sonnet

124:14

which is the brain

124:15

that we're going to use

124:15

So which cloud code brain do you want

124:17

to use for this exact

124:18

Now bear in mind that we have three

124:19

different models. We have Haiku

124:21

which is fast and cheap

124:22

We have Sonnet

124:23

which is why it's at balanced recommended

124:24

Most agents use this. And then we have Opus

124:27

which is the most powerful

124:28

Bear in mind that the most powerful they are

124:30

the more costly they are

124:31

Haiku is very, very fast

124:32

but it's only used for things like reading files

124:34

searching code, or quick lookups

124:36

So looking for different things that don't necessarily

124:38

require a lot of brain power

124:40

Then we have Sonnet

124:40

which is typically good for most things

124:42

It's medium cost, so it's cost effective

124:44

hence why we say balanced

124:45

But it's good for writing code

124:46

code reviews

124:47

and most of the daily tasks that you do out there

124:49

like writing emails or making content

124:51

And then we have Opus

124:52

which is the most powerful

124:54

You're rarely going to use this

124:55

but you would use this if you're doing very hard debugging

124:57

if you're making some sort of full stack application

125:00

and you need a agent that can go there and debug

125:03

meaning fix the code that you're writing

125:05

or if it needs really

125:05

really complex reasoning for it

125:07

right? So basically the overall summary here is

125:09

use Haiku for simple tasks

125:10

Sonnet for most work

125:12

and Opus only when you need the big

125:13

brain. And once we're done with the model

125:15

then we have the instructions

125:16

So this is the rules and personalized

125:18

And guidelines of the actual subagent

125:20

So for example, we have

125:22

you are a professional email writer for an

125:23

AI automation agency

125:25

When asked to write an email

125:26

you ask who it's for and the goal

125:28

you write a clear professional email

125:29

and you keep it under 150 words

125:31

Match the tone to the situation

125:33

always end with a call to action

125:34

So this right here is the personality and

125:36

step-by-step instructions. Think of this

125:38

as an employee handbook. You tell it exactly

125:40

how to do its job

125:41

right? So these are the instructions that you

125:42

give it

125:43

And so that right there is the structure

125:44

of the.md file

125:46

which is the

125:47

file where we're going to give everything that we need

125:50

to the subagent, to Claude in this case

125:52

to know that, hey

125:53

this is the subagent that we want to

125:54

run. I'm just going to paste the

125:55

example that I already showed you

125:57

We have the name

125:57

description, tools, model

125:59

and then we have the instructions here as

126:00

well

126:01

So now we can press command

126:02

S to save it

126:03

We can go to Claude

126:04

and I can paste this prompt right here

126:05

which is

126:06

write a follow-up email to my client John

126:07

about the chatbot project. Now what it should

126:09

do is it should look at the email writer MD

126:12

which is this right here

126:13

It will read through the instructions

126:15

It will look at what it has to

126:16

do. In this case

126:16

the first thing it should ask me is

126:18

Asked who the email is for and what

126:19

the goal is. In this case

126:21

we already told it who it's for

126:22

not entirely because it doesn't have the email

126:24

so it should ask us the email itself

126:26

As you can see

126:27

it just made the email called email john follow-up dot md

126:30

where we have the two

126:31

john, and by the way

126:32

we could actually make it so that it

126:34

can send the emails for us

126:35

but in this case

126:35

I just wanted to

126:36

to write the email

126:37

We have the subject line

126:38

checking in chatbot project and next steps

126:40

Hey john, hope you're doing well

126:41

I wanted to touch base on the chatbot

126:42

project and to see how things are progressing

126:44

on your end

126:45

We made solid progress so far

126:46

and I want to make sure that we

126:47

keep the momentum going. If there's any questions

126:48

blockers, or feedback you've been sitting on

126:50

now's a great time to surface them

126:51

I'd love to align on the next steps

126:52

and make sure we're on track with the

126:54

timeline. Would you be available for a quick

126:55

call this week or early next week?

126:57

Happy to work around your schedule

126:58

just let me know what works best for

126:59

you. Looking forward to hearing from you

127:01

best Mikalec. Yeah, very, very good email

127:02

And so as you can see

127:04

we just gave it a simple prompt

127:05

to write a follow-up email to my client

127:07

john about the chatbot project

127:08

Then it knew that it had to use

127:09

the agent email writer dot md

127:11

to fulfill the actual task

127:12

Now in the grand scheme of things

127:13

you will have the output

127:14

So some of them are outputs

127:15

right? Which is this right here

127:16

which is the output of the

127:17

task that we But sometimes it can get very

127:19

very messy. So what we usually do

127:21

is we make a folder

127:23

which is not in the doc cloud folder

127:25

called outputs

127:26

and then we store this right here

127:29

Move. So that all the outputs go

127:31

under outputs folder. It's all to keep things very

127:33

very structured. With that said

127:35

you got to see how I built it

127:36

plus how I ran it step by step

127:37

Now, in terms of triggering

127:39

which is like how do you start the agent itself

127:40

you saw me talking to the

127:42

cloud agent

127:43

and not really referencing the name of the

127:45

actual agent itself. But that doesn't work too

127:46

well when it comes to you having skills that are very

127:48

very similar. And so

127:50

the main ways that we actually

127:51

run the actual agent is three different ways

127:54

We have the natural language

127:55

which is similar to what I just did

127:57

but in this case

127:58

we ask it to use the code reviewer

128:00

to check my changes

128:01

use the name of the actual subagent to do X

128:03

Y, Z, to run the task

128:05

So cloud figures out which agent to use

128:06

and then it runs the task

128:07

The second and the most common one is

128:09

mentioning. So you're going to use at

128:11

the name of the actual subagent to

128:13

tell it to initiate X

128:14

Y, Z, or to do whatever it has to do

128:16

right? Create the task. Now this is guaranteed

128:18

to trigger because you're referencing that specific agent only

128:21

right?

128:22

And then finally we have session wide

128:24

which is use one agent for everything

128:25

So use cloud dash slash agent code reviewer

128:28

So my advice is to start With the first one

128:30

natural language, because that is the easiest to start with

128:32

and then move on to the second one

128:34

mentioning

128:34

right? The only reason for that is because

128:36

most times you guys are not going to

128:37

have 35 different sub-agents that you have to manage

128:41

right? And so you want to keep it

128:42

as simple as possible. So right here

128:43

to reference the sub-agent

128:45

you can just say

128:45

can you make an email using the

128:47

email writer agent to Jessica

128:50

telling her we have dinner tomorrow

128:51

and then you put the actual name

128:52

right? You can also not put the slash itself

128:55

it's just something that

128:55

I like to do to keep it more

128:56

precise. And in this right here

128:58

we'll know

128:58

that it needs to use the agent itself

129:01

It's going to look through the agents here

129:03

Of course, there's only one, right?

129:04

And it drafted the email itself

129:06

The second way, like I mentioned, is you go to tag

129:08

and you can see the different options

129:10

and you would use the email writer.md

129:13

And then you can say

129:14

can you make an email for James and Jack

129:16

telling them

129:17

thank you for the business meeting that we had before

129:19

and tell them that the next steps is

129:20

to have an onboarding call tomorrow

129:22

5 PM EST

129:23

to go through the next steps

129:24

And so because that was very specific

129:26

and turn this off

129:26

so you know that you're not reading it

129:28

I'm not cheating. But we were exactly referencing

129:30

the agent that we want to use

129:31

and so

129:31

there's no way that it will fail

129:33

in using the agent because we gave it

129:35

the exact instructions to use that agent

129:37

And as you can see

129:38

it made the email itself

129:39

James and Jack, saying, great meeting, onboarding call tomorrow 5 EST

129:42

and then it made the whole thing

129:43

right?

129:44

Obviously

129:44

you could change this by putting your name

129:46

I didn't really give it the main folder

129:47

which is a cloud.md folder

129:49

just because I didn't want to spend too

129:50

much time on that

129:51

but you can give it more information as well

129:53

Now

129:53

in case that you are using the terminal to run cloud

129:55

you can simply go to slash agents again

129:57

and then you'll be able to see the email writer agent

129:59

which is this one right here

130:00

And you can simply press enter again

130:02

you can view, edit the lead agent

130:04

you can back agent

130:05

and you can do all sorts of things here

130:07

And again, you can simply just give it natural language

130:09

or you can press at

130:10

email, writer agent

130:12

and then you can tell it whatever you

130:13

want. Make an email to James tomorrow

130:15

And you can run this

130:16

and it pretty much does the exact same

130:17

thing. I just want to show both things

130:18

because I know some people get

130:20

when I see some people

130:20

I mean me at the start

130:21

I got a bit confused when it came

130:22

to using the

130:23

terminal versus using the normal cloud chat and

130:25

how those actually work. So as you can see now

130:27

one thing we didn't have before

130:29

I told Claude that if I don't tell

130:30

her who it specifically I have to send

130:32

the email to and what the goal of the email is

130:34

then it should ask me

130:35

And so here it's saying

130:35

who is James and what's the purpose of

130:37

the email? James is a client and the

130:38

purpose of the email is follow up after

130:40

a meeting. WhisperFlow has changed my life

130:41

guys. You should definitely try it out

130:43

As you can see

130:43

we have the full email

130:44

James, subject, and the body itself

130:46

Now

130:46

one more thing that I spoke about before

130:48

is restrictions and safety. Because the thing about

130:50

sub-agents is like they get really

130:51

really powerful because you can control what they

130:54

can

130:54

and cannot do. The same way that you

130:56

wouldn't give an intern access to the whole company database

130:59

here's the exact same thing

131:00

And so we have the read-only agent

131:01

we have the balance agent

131:02

and we have the full access agent

131:04

Now, when you said read-only

131:05

it can look at your files

131:06

it can search through the code

131:07

it can explore kind of what you have

131:09

but it cannot edit or write

131:11

right? It can look but can't touch

131:12

It's perfect for research

131:13

that you have to do without it actually

131:15

doing

131:15

something on your documents or changing anything on

131:18

your computer. Then we have the balance agent

131:19

it can read search

131:20

it can also edit and run commands

131:22

but it cannot delete files

131:24

right? It can build things but can't break

131:25

things. It can only add but it can't remove

131:27

which is great because a lot of times

131:29

When you have an agent that's out of hand

131:30

that's going everywhere

131:31

it can delete things that later on you can't get back

131:34

right? And so you want to keep it

131:36

normal and clear. And then we have the

131:37

full access agent. It can read

131:39

edit

131:39

write. So full power it only uses when

131:41

you really trust the tasks you're doing

131:42

Table key is for tasks that are easy

131:44

that don't require a lot of brainpower

131:46

that are very, very structured

131:47

You have a step-by-step

131:48

it's done this way

131:49

and it can be done pretty fast

131:50

So here you decide what each agent can

131:51

do. The less access to sit for it is

131:54

the more access, the more powerful

131:55

but the riskier it is for you

131:57

as the person who's actually talking to the

131:58

agents. And the way that you would change

132:00

the actual permission is if I go here to the MD

132:02

you would have tools

132:03

right? The tools is the permission

132:05

So you can literally think of this

132:06

as if the words permissions

132:08

but we use tools

132:09

right? Now, if you only want this to read

132:11

you can copy this

132:13

Read, grep, and glob

132:15

That's it. Now, if you want a balanced

132:17

which is the most common

132:18

then you want this here

132:20

Read, edit, bash, grep, and glob

132:21

It's also going to give you suggestions

132:23

but

132:23

I'm going to keep it normal for now

132:24

And then if you want the full access

132:26

you don't have to specify tools

132:28

It Parrots everything. So if this is empty

132:30

then it will use any tool that it

132:32

has access to. It's only if you specify

132:34

the level of permission that you want the sub-agent to have

132:36

that it actually uses it

132:37

per se. Now, when it comes to sub-agents

132:39

there's one issue

132:40

which is the issue of memory

132:42

Because every single time that you have a

132:43

new session with Cloud Code

132:45

so a new conversation

132:46

the sub-agents start from scratch

132:48

which means that they don't have a previous

132:49

memory of the conversation that you had before

132:51

Now, as you can see here

132:52

we have a folder called agent memory

132:54

and this is where this comes into play

132:56

Because now we're able to actually have the

132:57

sub-agent remember the previous conversations

133:00

and we don't have to re-explain things again

133:02

So

133:02

all we have to do

133:03

is go to the actual agent itself

133:05

the prompt, and then we can simply

133:07

right below model and above this

133:09

we want it to include

133:10

but all you have to do here is

133:10

put memory colon project

133:12

okay? And you press save

133:14

So let me show you how it actually

133:15

works. So if I go to a new tab

133:16

of course, again, save the project

133:18

so it's here. So I can go to a new session

133:20

and I can say

133:21

email writer.md, that's fine

133:24

I can paste this right here

133:25

which says, make an email for John

133:27

saying that we have dinner tonight

133:28

or tomorrow night. 5 PM

133:29

and make it really

133:30

really, really, really

133:31

really happy with an exclamation mark

133:32

Also add the tone to the memory

133:37

and I can press go

133:38

And so now what it does

133:39

it's going to launch the email

133:40

right, to actually make the email that we asked it

133:41

but it's also going to save the tone preference in parallel

133:44

so at the same time

133:45

Write the email and save the tone into

133:47

the memory itself. So you can see now

133:48

it's starting to work with memory.md

133:49

It's going to save the tone preference to the memory itself

133:51

which is this folder right here

133:53

I can press accept. That's fine

133:54

Accept again. So as you can see here

133:56

it made the email dinneremail.md with all the exclamation marks

133:59

it's on a voice. And then we have

134:01

the agent memory. The agent memory has two

134:03

different things. Has feedback tone matching.md

134:05

and it has memory.md

134:07

which is the main one

134:07

So feedback tone matching means this is the

134:10

explanation

134:11

of the actual tone

134:12

of the email that we need to use

134:13

So that we don't have to re-explain to

134:15

it again to use exclamation mark

134:17

to use X, Y, and Z

134:18

This list of

134:20

tone for emails is going to be used

134:22

again and again when we're making new emails

134:23

right? And then we have memory.md

134:25

which has a memory index

134:26

which means that every single time that we

134:27

give feedback to the Actual agent itself

134:30

it's going to store the feedback here

134:31

And so we'll have more and more feedback

134:33

so that it remembers the previous things and

134:35

it gets better over time

134:36

Self-improving

134:37

sub-agents. And then we have the memory and

134:39

the prompt itself

134:40

I can go here and I can say

134:41

I can reference the email writer.md

134:44

Is it this? Yep

134:45

And I can say

134:46

can you

134:47

send

134:47

an email to Jack saying that we have football tomorrow 5pm

134:51

bring the boys over as well so all

134:53

of them can have a good time

134:54

And yeah, make it exciting

134:55

So as you can see now

134:56

as it's running

134:57

it's going to read the memory.md file

134:59

And it's going to do that every single

135:00

time because it's with this specific agent itself

135:03

The more memories you have

135:04

so the more self-improving the system will become over time

135:06

right? And the more effective and the accurate

135:08

the results are

135:09

when it comes to doing the exact task

135:10

It's also looking at the feedback tone matching.md

135:12

And then we have the actual

135:14

email itself with exclamation mark here

135:16

here, here

135:18

Kind of looking at the matching here as well

135:23

Right

135:23

so you can now spin up sub-agents to

135:25

handle specific tasks in parallel

135:27

working at the same time

135:28

But what happens when you need

135:29

multiple agents working together on a bigger project?

135:32

That'd be coordinating

135:33

sharing context, dividing the tasks, right?

135:36

That's exactly what agent teams is all about

135:38

And the way that I like to think

135:39

about it is that agent teams

135:40

is like sub-agents on steroids

135:42

So, let's dive in. Right

135:44

so here I broke down cloud chat versus

135:45

sub-agents versus agent teams

135:47

what they all are and how they're different

135:49

All right

135:49

so the first one here is cloud chat

135:50

This is you talking to cloud directly with no tools

135:52

no code, just a conversation, simple back and forth

135:55

and nothing in between. And typically you use

135:56

cloud chat only when you have a question

135:58

or want some help writing something

136:00

just a normal conversation. Maybe you can write

136:01

your emails. An example of this could be

136:03

that you're building automation in

136:04

let's say, an 810. You have an error

136:06

you drop the error message into cloud code and say

136:08

hey, can you explain this error message?

136:10

Or can you help me write my cover letter

136:12

brainstorm video ideas? Basically the way the ninety-nine

136:14

percent of people are using AI right now

136:16

it's using cloud chat. Then we have sub-agents

136:18

and sub-agents just means that you have a boss

136:20

and the boss has three employees

136:22

But, this is a but

136:24

the employees cannot talk to each other

136:25

okay? So if employee A

136:26

is making something that then depends on employee B

136:29

well that's a problem because they can't speak

136:31

to each other and they can't communicate

136:33

right, when making a task together

136:34

which could decrease the quality of output

136:36

and it just wouldn't work well

136:37

The results can only return to parent agent only

136:40

and they have a shared context window

136:41

So they have a shared context window

136:43

which means that they all operate under the same window

136:44

the same kind of tab in Cloud Code

136:46

And when do we use sub-agents?

136:47

Well, sub-agents are used when you're building code

136:49

or you need Cloud to research

136:50

explore

136:51

or plan different things

136:53

in parallel at the same time

136:54

This could be add authentication

136:55

write test, right

136:56

where you have two different things

136:58

which are not dependent on each other

136:59

running at the same time

137:01

or explore this code base for me

137:02

plan this feature, then build it

137:03

And when you're building with sub-agents

137:05

you ideally want to have two different tasks

137:07

or three or four or five

137:08

whatever it is that you want

137:09

but make sure that these tasks do not

137:11

correlate with each other. So for example

137:13

let's say you have a YouTube transcript

137:14

and you want to repurpose that YouTube transcript

137:16

into a LinkedIn post

137:17

a blog post, and maybe a Twitter post

137:19

right? All these three things

137:21

you might think They're not dependent on each other

137:23

but they are. Because the style of writing

137:25

the tone of voice

137:26

has to be the same

137:27

across the different content that you make

137:29

And so with sub-agents

137:31

you're not able to have the LinkedIn post

137:33

talk to the blog post

137:34

talk to the Twitter post

137:35

to make sure that they all have

137:36

the same exact tone of voice

137:38

And then we have the one and only agent teams

137:41

okay?

137:41

So the way the agent teams works

137:43

is that you have a team lead

137:44

So you give team lead a task

137:45

say, hey

137:46

can you make this team of people that can do X

137:49

Y, and Z

137:49

right? What this does is that it spawns a team

137:52

so it creates a team

137:53

and assigns tasks across the different teammates

137:56

And just like sub-agents

137:57

it will start creating agent one

137:59

agent two, and agent three

138:00

and these agents all have the same shared task list

138:03

but the agents communicate with each other

138:05

and each one has its own context

138:07

So it has its own context window

138:08

and operates under its own kind of environment

138:10

right?

138:11

And so

138:11

you have team at A talking to team at B

138:13

talking to team at C

138:14

which can all see the same tasks that they have here

138:16

and the team lead is the one that

138:18

kind of operates the whole

138:19

the whole team. And this would be like an actual company

138:21

right? All the employees are talking to each other

138:23

same departments, maybe different departments

138:24

but they can still speak to each other

138:26

correlate, and communicate with the head of department

138:29

kind of in that role

138:30

Now

138:30

you only use agent teams really when you

138:32

need multiple cloud working on separate repos or

138:34

separate kind of instances

138:35

or when you have a big tasks

138:37

right? Usually it's development-based

138:38

so it's more technical. For example

138:40

you can build a front-end plus back-end plus

138:41

tests. Front-end does depend on the back-end

138:43

and it does depend on the tests

138:45

You want to have two different agents under

138:46

the same team

138:47

that are building the exact thing that you want

138:49

but bear in mind they can also communicate with each other

138:51

so the output will be much higher

138:53

right? Because obviously the front-end will depend on the back-end working

138:56

and the test will depend on front-end and back-end working

138:59

so they want them to communicate with each

139:00

other at the same time

139:01

Or migrate five services at once

139:02

but also one cloud orchestrating the rest

139:05

Now, in terms of the actual build itself

139:06

there is a repeat and improve process

139:08

which is a four-step cycle that we go

139:09

through every single time when it comes to the communication

139:12

when it comes to speaking to the agent

139:13

team for whatever it is that we want to build

139:14

right? The first thing is collect

139:16

So we want to brainstorm ideas

139:17

and it's going to list every tasks and set clear goals

139:20

just like a head of department would

139:21

or just like CEO would

139:23

Then we organize

139:24

So we rank by priorities

139:25

we set deadlines

139:26

and we block the time slots for the

139:27

agents itself. Then we execute

139:28

so each agent will work through the tasks

139:30

We'll track the progress

139:31

and we'll have focus on what we have to do

139:33

because each agent has its own task that

139:35

it has to do. And then review

139:36

which is checking what's done

139:37

know what's left

139:38

and adjusting the plan

139:39

in case we need to change

139:40

And then we repeat and improve

139:41

and we keep going on and on and

139:43

on. And that's exactly how the team actually operates

139:45

just like you would for a real company

139:46

And one more thing before we get to the build

139:48

is that this is

139:49

great because we have what we call

139:51

parallelization advantage. Because a traditional workflow would have six tasks

139:55

and let's say they all take one minute

139:57

right? So you have six minutes total

139:58

When you have an agent team

139:59

we have what we call parallel processing

140:01

So instead of each task being done individually at different times

140:05

here you'd have three different agents

140:06

all working

140:07

at the same time on their own tasks

140:09

Task one, task three

140:10

and task five will be done at the exact same time

140:12

as well as task two

140:13

task four, and task six

140:14

which means that instead of spending six minutes

140:15

or waiting six minutes

140:17

then you'll wait two minutes

140:18

because we have what we call parallel execution

140:20

Now these are all fancy terms

140:21

Just to say that this basically saves us

140:23

a lot of time. I'm speaking in

140:24

in like minutes

140:25

but sometimes it can be hours

140:26

that you would wait for a workflow to

140:28

to work. And here you would literally get it within minutes

140:31

which is kind of mind blowing

140:31

And I think that's where the future of

140:33

our gentle workflows are going to be

140:34

It's having one main agent

140:36

which controls all the others

140:37

and having

140:38

them

140:39

do the tasks for you without you having

140:40

to be there. Okay

140:41

so here I am in anti-gravity

140:43

our IDE that we're going to use

140:44

And by the way

140:45

anti-gravity VS Code, basically the exact same

140:48

You can pretty much use VS Code as well

140:49

but

140:50

I just chose anti-gravity because that's what I

140:51

have here. First thing I'm going to do

140:52

here is press open folder

140:53

So I typically like to have a Cloud

140:55

Code main folder and then have different folders inside

140:57

just something that I do

140:58

And I can press new folder here and name this agent

141:02

so Cloud

141:03

Agent Teams. And I can press open

141:06

I already downloaded the extension

141:07

but you can go here to extensions

141:08

You can look for Cloud Code

141:11

Make sure it's the one by Anthropic

141:13

which should be the one here

141:15

Anthropic

141:16

And then

141:16

you can install it and then you'll be

141:17

able to see a little button here

141:19

which has Cloud Code open

141:20

and you can press this button

141:22

X out of here

141:22

X out of here

141:23

and you go back to here

141:24

Okay

141:24

so here I am on the Cloud Code

141:25

documents. This is the official Cloud Code documents

141:27

which tells us how to

141:29

basically start the team. I found this by

141:30

going to code.cloud.com slash docs slash en slash agents slash teams

141:35

and you'll see run agent teams right here

141:37

And if I scroll down

141:39

you can see when we use it

141:40

when we don't use it

141:40

how it works. And then here

141:42

it says to enable agent teams

141:44

So to enable agent teams

141:45

we have to make a file called settings.json

141:47

and then we have to put

141:48

this right here inside. I'm going to go back here

141:51

I'm going to press new file

141:52

I'm going to name this settings

141:55

is it settings or settings?

141:56

Settings.json, settings.json

141:59

and you'll see that this will now be turned into JSON

142:01

because that's the data structure

142:03

And if I go back

142:04

I can copy this

142:05

and then copy it back

142:07

And I can press file

142:08

save

142:09

What you could also done

142:10

is you could just actually copy this

142:12

or you can just go command A

142:14

copy the whole thing

142:15

and then you can paste it into Cloud Code and say

142:17

hey, based on this documentation

142:19

can you please enable agent teams?

142:20

And it will do it for you

142:21

But I like to be different

142:22

so there you have it

142:23

Okay, so settings.json we have this here

142:25

and now

142:27

this is where we can prompt

142:28

Cloud Code to be able to make the agent

142:29

All right

142:29

so I already have my prompt here that

142:31

I'm going to use

142:32

You can find it in the second link

142:33

down below in my school community

142:34

So what this prompt says is

142:35

create a multi-agent team that turns

142:37

sales discovery call transcript into useful sales assets

142:40

When I drop the transcript into the slash input

142:43

the system to generate the client summary

142:45

the product proposal, the deal updates

142:46

and follow-up emails. So we'll have four different

142:48

agents. The first agent is call insights agent

142:51

So we analyze the transcript and extract the client company summary

142:54

core problem pain points

142:55

desire.coms timeline, and the operable B a client summary.md

142:59

which will be a file

143:00

which will give us a summary

143:01

Then we have agent two

143:02

which will be a proposal builder agent

143:03

So this will generate a proposal

143:05

And then we have CRM agent

143:06

which will create a structured JSON CRM update

143:09

including Company contact name problem

143:11

and a bunch more other things

143:12

And here we'll have the output

143:14

which will be deal up to JSON for

143:15

the CRM

143:16

And then we have the follow-up email agent

143:17

which will create a follow-up email

143:19

So basically

143:20

what I'm telling the agent to do

143:21

is I am going to drop a sales

143:23

discovery call transcript

143:24

So let's say I had a call with a client

143:25

I dropped the transcript of the call

143:27

so what we said on the call

143:28

And then the team will be consisted of four different agents

143:30

one, two, three, and four

143:32

So this agent will make the client summary document

143:34

this agent will make the proposal

143:35

this agent will make the CRM kind of update

143:37

and this agent will make a follow-up email

143:39

With that said, I'm going to copy this

143:40

I'm going to go back here

143:41

and I'm actually not going to be using this chat

143:44

but I'm going to be using the terminal

143:47

Now

143:48

I was a bit confused when I saw

143:49

people using the terminal versus using the chat

143:51

The reason why we use the terminal here

143:53

is because we can see more

143:54

So if I press cloud here

143:56

my cloud will open up

143:57

Like this tells us everything

143:58

This shows all the settings

143:59

it shows all the small things

144:00

It might look a bit more complex

144:01

but it's very

144:02

very easy when you start using it and you

144:04

you get the hang of it

144:04

I'm going to press X

144:06

I'm going to make this full screen

144:08

and we can use the terminal

144:09

And by the way

144:09

you can use the other one as well

144:10

it's just that I can't see all the

144:11

features that I want to show you

144:13

but it still works the exact same

144:14

I'm going to go here

144:14

copy this, and then paste the prompt here

144:17

I'm going to press enter

144:18

and now

144:19

cloud code is going to start working on

144:20

this. Test tool loaded

144:22

We have this setting.json

144:23

which means that we are enabling agent teams

144:25

and so it knows that it has to

144:26

make a team of agents

144:28

It's now saying, hey

144:29

we have to create the input and output

144:30

directory structure

144:31

I can just say yes

144:32

proceed. I could also say yes and don't ask again

144:35

but I want to show you all the different

144:36

things that it asked me to do

144:38

which is fine, so I can press yes

144:39

As you can see

144:40

it made the output and the input

144:41

The input is where we have the transcript

144:43

so we'll be adding the files here

144:45

and the output is where we get the CRM

144:46

the emails, the insights

144:48

and the proposals as well

144:49

for each transcript that we use

144:51

Now it's saying that it's going to create

144:52

a sample transcript and orchestration script in parallel

144:55

so it's also testing that it does work

144:57

Now it's asking whether we can

144:58

make a transcript.md

145:00

document to basically test

145:02

a transcript of a call and make sure

145:03

that all the agents work simultaneously

145:05

I can press yes

145:06

This is now writing the transcript.md

145:07

which I believe is going to be here

145:08

Yeah, the input, because that's the input

145:10

and then it's saying

145:11

do you want to make this edit to

145:12

run pipeline.sh? I can just press yes

145:15

So now it's asking me all the different

145:16

small things that I typically would not want to be asked

145:19

because it's all code

145:20

Now I can go to yes and don't ask again

145:22

Okay

145:22

now it's also making the folder called agents

145:25

So all the different agents that we have

145:27

the four different agents

145:28

will all have a md document

145:29

So md just stands for markdown

145:31

it's the way that they basically

145:32

put titles and subheadings and

145:34

bullet points and so on

145:35

to structure the information

145:36

And so each agent will have an md document

145:38

and the md document

145:39

it's simply just instructions of what the agent

145:41

is

145:42

So then when we give it a transcript

145:43

it basically goes back to the md documents and says

145:46

okay, what does this agent have to do?

145:48

What about this agent? What about that agent?

145:49

And the last agent as well

145:50

And it will start running all of them

145:51

at the same time. Okay

145:52

there we go. So this is the structure

145:53

we have dot cloud slash agent

145:55

Then we have agent one

145:57

Let me just take this out here

145:59

We have agent two

146:00

We have agent three

146:01

And then we have agent four to make the proposals

146:04

Okay, so I'm pulling back up the terminal

146:05

I kind of Actually just

146:07

clear this conversation

146:10

Clear

146:11

I can start from zero

146:11

So now what we'll do

146:12

is we'll actually test it

146:13

So we have everything done

146:15

We have the cloud agents set up

146:16

We have the input and the output folders

146:18

We have the script

146:19

and then we have the

146:20

the thing that allows us to be able

146:21

to make the team. So I could just

146:22

go back here

146:23

I can then

146:24

go to this example discovery call

146:26

This is an example

146:27

like example input that I would give the

146:29

actual agent

146:30

And I can say

146:31

hey

146:31

can you use this discovery call transcript

146:34

to be able to activate the team and

146:36

yeah, activate the whole team

146:37

agent team

146:38

in Cloud Code and use this transcript as

146:39

the input

146:40

Not a question. Paste it here

146:42

And now we press run

146:43

So this basically pasted the whole

146:45

the whole call that we have

146:47

It's a short call

146:47

so nothing crazy

146:49

but it's enough for it to be able

146:50

to make all the things that it wanted

146:51

to make. So it's saying

146:52

I can see the full agent team pipeline

146:53

Let me first update the transcript file with

146:55

the callants marketing discovery call

146:56

and then run the pipeline

146:58

Pipeline just means run the agent team

146:59

Now it's saying, hey, can we replace the transcript MD

147:01

which is the one here

147:03

with the transcript that you just gave us?

147:04

I'll just say yes

147:05

allow all edits during the session

147:06

just because I don't have to then

147:08

prove everything. So it

147:08

it uploaded here, and now it's saying

147:10

hey, can we run the pipeline?

147:12

I can just press yes and don't ask

147:13

again

147:14

Now let's see, hey, let me kick off agent one

147:16

call insights agent first

147:17

since the other three depend on its output

147:19

And this is great because you get to

147:20

see exactly how agent one

147:22

now talks to the other agents

147:23

Agent one is the call insights

147:25

and based on the call insights

147:26

we then feed it to the other agent

147:27

to make the email and the proposal

147:28

and the CRM as well

147:29

Okay

147:30

so now it's telling me that it made the client summary

147:31

company overview

147:32

core problem, main pain points, desired outcome, timeline, red flags, objections and concerns

147:37

and key quotes

147:38

I can say, yes, I want to proceed

147:40

and now I should get the insights here

147:41

client summary.md. And by the way

147:43

I'm doing this obviously manually to show you

147:44

how to test it and how everything works

147:46

but you can pretty much

147:47

upload this or publish this to a platform like modal

147:50

and then all you have to give it is the input

147:52

and then the whole agent team will work

147:54

on its own

147:55

And one thing I'll also cover later on is the pricing

147:57

because you can see I'm getting

147:58

completely rinsed with my tokens

148:00

It's going faster than

148:02

than I've ever seen. So

148:03

this is one drawback of agent teams

148:04

but

148:05

we'll get into that in the later part

148:06

So agent one was complete

148:07

It then ran agent two

148:08

three at the same time

148:10

so two agents finished

148:11

It built the product proposal from insights and

148:13

the CRM deal for JSON

148:14

From insights

148:15

And now that agent two and three are done

148:17

now we're going to be launching agent four

148:19

which is going to be creating the follow-up

148:20

email. So it's saying now that all four

148:21

agents have been completed

148:23

here's a summary of what was generated

148:25

So we have the call insights agent

148:26

proposal, CRM, and follow-up emails

148:28

We have the files

148:29

which you can find emails here

148:31

CRM here, we have insights here

148:33

and we have proposal here

148:34

which I'll go in just a second

148:36

And it says all outputs files are ready

148:37

in the output directory

148:38

Want me to open them for review?

148:40

I can just say

148:41

yeah, let's see what it does

148:43

Yes, open them

148:46

By the way

148:46

you can also edit the permissions

148:47

So right now, if you press shift tab

148:49

shift tab, you can see that I'm

148:52

I'm changing this plan mode accepts edits

148:54

You can also do bypass permissions

148:56

which is letting it just

148:57

do everything on its own

148:58

So client summary, this is a document

148:59

product proposal, CRM deal update, and follow-up emails

149:02

So I'm gonna make this smaller so you can see

149:05

right? And I'm gonna go through each one

149:06

So CRM

149:07

This is the

149:08

deal update that you uploaded in your CRM

149:10

to be able to include all the fields

149:12

So this included the company name

149:13

industry, size indicator, growth rate, contact name, title

149:16

email. There was no email

149:17

so we couldn't find it

149:18

The deal stage, discovery, summary, main problems, pain points, desired outcomes

149:22

and a bunch more other things as well

149:24

Then we have the emails

149:25

So this is the follow-up email that I made

149:26

It said, to Sarah Collins from Wren GM Solutions

149:30

Subject line, your roadmap from 18 to 40 clients

149:33

proposal inside

149:33

Thank you for the candid conversation on Wednesday

149:35

The way you described the gap between where

149:37

Collins marketing is today and where you want it to be

149:39

40 clients without the operations breaking

149:41

gave us a clear picture of what needs

149:43

to happen. A few takeaways I want to

149:44

make sure we capture correctly

149:45

Onboarding is the biggest time drain

149:47

An hour per client with manual folder creation

149:49

document sends, and tool setups across Google Drive

149:51

ClickUp, and Slack

149:52

and things still slip through the cracks

149:54

Reporting is costing your VA three to four

149:55

hours per week. The knowledge bottleneck is real

149:57

I've attached a proposal which outlines a two-phase eight-week engagement

150:01

and talk soon

150:02

Wren. We also ask you to book a call

150:04

Now, if I go to client summary

150:06

I can see the insights

150:06

So this is the

150:07

Summary of the actual conversation that we had

150:09

on the call. And then proposal

150:10

we have the proposal that we have

150:12

that it just made based on the conversation

150:13

So understanding your challenges

150:15

manual client onboarding, reporting tools

150:18

knowledge, proposed solution, and we have scope of work

150:21

deliverables, timeline, investment, Why us, next steps

150:25

and so on

150:25

And these are all things that you could

150:27

easily turn into like a PDF

150:28

So let me go here to Cloud Code

150:30

I'm going to say

150:31

can you turn the

150:32

project proposal.md file

150:33

into a PDF that I can use?

150:36

Make sure the formatting is correct

150:37

Obviously, if you look at this

150:38

you'd be like, okay

150:39

how do we use this now?

150:40

You could easily

150:41

copy that into Notion or copy that into Google Docs

150:44

but

150:44

I just want to show you

150:45

how Cloud Code does it for me

150:46

All right, so I just finished making the product proposal PDF

150:49

so we can go here

150:50

This is the Markdown format

150:51

and it turned it into a PDF

150:53

Of course, if you look at it this way

150:54

it also wouldn't make sense

150:56

but because it's a file on our computer

150:58

I'm just going to open it up

150:59

and we get this right here

151:00

which is PDF version with the executive summary

151:03

understanding your challenges, then we have tools

151:05

proposed solutions, scope of work, deliverables, timeline, investments

151:08

It also made the table

151:09

very nice formatting, the price, why else

151:12

and next steps

151:12

And so that's exactly how the whole agent

151:14

team ran through to make the output

151:16

which you can then use for whatever it

151:17

is. And this right here is a 7

151:19

out of 10 use case

151:20

I would say

151:20

because you could still do it with sub-agents in some way

151:22

but the output is better here

151:24

Just because you have the client summary

151:26

which is made

151:26

and then passes the client summary to the

151:28

proposal and to make the email

151:30

and the CRM

151:31

to then give you the full output

151:32

All right

151:33

so there's one aspect of this that I briefly mentioned

151:35

but I didn't really dive deep into

151:36

which is price. Because if you saw before

151:39

for me to run all the four agents

151:40

it cost me 10,000 tokens

151:42

and it was increasing

151:44

way too fast for my liking

151:45

And so here we have token cost

151:47

and here we have speed

151:48

Obviously, regular cloud will take a lot more time

151:50

it'll be cheaper. The faster you get

151:51

and the more expensive it will get until

151:53

you get to agent teams

151:54

which is five minutes and 80 bucks

151:56

This is because

151:57

each employee in the agent team

151:59

has its own context. It has its own

152:01

kind of place where it runs

152:03

And so

152:04

it will cost you a lot more to

152:05

be able to run this agent team

152:06

But if you are building something a bit more complex

152:09

where let's say you're building a software application

152:11

and for the software application

152:12

you want the main

152:14

you know, team leader

152:15

and then you want the different agents to

152:17

be able to talk to each other

152:18

Because when you're making the front end

152:19

the back end, the privacy, the, the OAuth, all these things

152:23

you'd want the Different agents

152:24

to communicate

152:25

to build a much better output

152:27

rather than using sub-agents. But for the most part

152:30

if you're working with Cloud Code and you

152:31

want to run different tasks in parallel

152:33

you're better off going with sub-agents

152:35

rather than spending six times more

152:37

but also getting six times the speed

152:41

All right

152:42

so you now know how to build sub-agents

152:43

and agent teams inside of Cloud Code

152:46

But after having built hundreds of agent workflows

152:48

I've discovered specific techniques

152:50

hacks

152:51

really that make your Cloud Code

152:53

dramatically and exponentially better. And I've divided these

152:56

hacks into eight different hacks

152:58

which is why in this video

152:59

I'm going to break them all down step

153:00

by step

153:01

With that being said

153:02

let's dive in. All right

153:03

so we're going to divide these hacks into

153:04

three different types of hacks

153:05

We have system design hacks

153:06

then we have execution and performance hacks

153:09

and then we have control and optimization hacks

153:11

All right

153:11

so the first hack is actually setting up

153:13

the folders for your project

153:14

All right, so here we have Cloud Code

153:16

and the mistake that people make when starting

153:18

out to build is that they go straight to building

153:20

right?

153:20

But the problem is that right here on the left-hand side

153:23

where we store all the folders and the

153:24

files and all these things

153:25

we have to structure

153:26

Cloud Code in a way

153:27

where it knows where to store all the

153:29

different folders and the files

153:30

which is where we go to the three-layer agent architecture

153:33

which is a way

153:34

for us to be able to structure

153:35

the AI agent to be way more reliable

153:37

The way that we do this is

153:38

dropping a file inside of Cloud Code

153:40

which is called the claude.md file

153:42

That's where here I can think of it

153:43

as a blueprint that you get to Claude code

153:46

and it contains a three-layer architecture

153:48

a three-layer structure

153:49

that Claude code needs to use

153:51

to be able to then

153:52

build these things and store all the files

153:54

in the same place. The first layer is the directive

153:56

so this is what to do

153:57

So we give it SOPs

153:59

we define objectives, we list the inputs and the outputs

154:01

we specify which tools do we have to use

154:03

and we cover all the edge cases of

154:05

the workflow. It's like a brief that you

154:06

give to someone working in your company

154:08

Then we have layer two

154:09

which is the orchestration

154:10

and this is for the decisions

154:12

So the actual LLM lives inside this folder

154:15

So this is good when giving it instructions

154:16

as to where to store the right scripts

154:18

right? How to handle errors

154:20

how to ask questions when it's stuck

154:21

It's self-healing, right? That's what we call Claude code

154:23

because when something breaks

154:25

it knows exactly what's broken

154:26

and then it goes back and fixes it

154:28

so you don't have to

154:29

And this is the glue between what you

154:30

want to build and what's actually built

154:32

And then layer three is the execution

154:35

This is doing the work

154:36

So this includes the Python scripts

154:37

the API calls

154:38

all the data that's going within the automation

154:40

All the files, all the databases

154:41

all the techy stuff that

154:43

it needs to do to actually do the

154:44

thing. And this is all code

154:45

right? Of course

154:46

cloud code is just building code that we use

154:48

Now, instead of doing this

154:49

why don't we let the LLM do everything

154:51

and structure everything?

154:52

It's because LLMs are about 90% accurate per step

154:56

right? It sounds good until you realize that

154:57

when you put step over step

154:59

you have 0.9 to the power of 5

155:01

which is 59% success rate

155:03

And it's simply not good enough at all

155:04

if you have anything below a 99% success

155:07

which is simply not good enough when we're

155:08

building agenda workflows. And so the solution is

155:10

to give it more frameworks

155:11

more guardrails in cloud code

155:13

to be able for it to understand how

155:14

to structure things when it's building

155:16

okay? Now, if that didn't make too much sense for you

155:18

I'll show you what it means

155:19

So right here

155:20

I have the cloud.md

155:21

file that I just mentioned

155:23

I'm going to copy this

155:24

And by the way

155:24

if you want access to this full prompt

155:25

second link down below in my community

155:27

you can find that in the classroom section

155:29

So we go here and we can make

155:30

a file and name this cloud.md

155:33

And then you can paste this little prompt

155:35

right

155:35

which again

155:36

is just a verbal explanation of what I

155:38

just said. So layer one

155:39

layer two, and layer three

155:40

And then you want to press

155:41

file and save

155:43

And then you can say

155:45

initialize this project based on the cloud.md file

155:48

that I just put in the project

155:49

or in the files

155:50

There we go

155:51

And I put plan mode

155:52

and then I press run

155:53

Okay, so now what it's going to do

155:55

it is going to go inside the cloud.md file

155:57

it's going to read it

155:58

and then it's going to start structuring all

155:59

the different folders of what it needs to do

156:01

All right, so it just made a plan here

156:02

plan to initialize project structure from cloud.md

156:05

the context, the changes. So if you see this right here

156:08

don't get too intimidated

156:09

What this means is the

156:10

structure of all the folders here

156:12

Right, so this is the project root

156:13

we have directives, we have execution, we have.tmp, and we have.env

156:17

and we have git ignore

156:19

right? These are all different things that are

156:20

part of the actual structure that it would

156:22

need to do

156:22

You can just press yes and auto-accept

156:24

Now it will have tasks

156:25

and then on the left-hand side here

156:26

you can see that it's adding the different

156:27

files and folders that it's going to use

156:29

when we're going to be able to make the actual workflow

156:31

to make things more structured

156:32

Okay, there we go. It just finished

156:33

So all the folders are here

156:34

we have directives, we have readme.md, which is a

156:37

basically a file that explains exactly

156:39

what this folder is

156:40

So this folder contains marked on SOP

156:42

standard operating procedures, the defined objectives, inputs, tool and scripts, outputs

156:46

and edge cases. And the easiest way possible

156:48

in this way, a second architect going to build a house

156:51

right? Either he goes to build a house

156:52

and starts building immediately

156:53

As soon as he goes on site

156:54

or what he does is that he sits down

156:56

he basically makes a roadmap or a blueprint

156:58

as to how he's going to build the house

156:59

and then we get to the building part

157:01

which is exactly what this is

157:02

This is the

157:03

structure and the blueprint that we're going to use

157:06

well

157:06

the cloud code is going to use

157:07

when we have to make all these things

157:09

To give you a practical use case

157:10

I actually built this YouTube strategy system for myself

157:13

and I made a full video as well

157:14

you can check it out up here

157:15

But essentially, we have the cloud.md file that we propped

157:18

which is the exact same one that we

157:19

did for the other one

157:20

and then it started making all the folders

157:22

and if you go to directives

157:23

we get to see all the different kind of SOPs

157:26

right, explaining the different automations

157:28

So for example, this right here, analytics tracking

157:31

So it's an md file

157:32

so it's marked on file

157:33

marked on because we have hashtags

157:34

it's just a way of writing

157:35

so AI or computers can understand better

157:37

But it's a marked on file that

157:39

tells the actual cloud code exactly what to do

157:42

the SOPs

157:44

This is what the objective of the thing

157:46

is

157:46

This is the objective of what it is

157:47

that we want to do

157:48

What is the input and what is the output

157:49

right? The input is the channel ID

157:51

and the output is performance dashboard data

157:53

So With all the data and all the different things

157:55

Now, to be fair

157:56

it might seem a bit intimidating when you're

157:57

starting out and you're wondering

157:59

still wondering

157:59

why is it like this or why is

158:00

it like that? But just understand that this

158:01

is the way that we structure all the

158:03

different folders and files

158:04

The second hack is using Cloud Code skills

158:06

Now

158:06

I've made a full video breaking Cloud Code

158:08

skills down in much more detail

158:09

but essentially what they are

158:10

is instruction manuals that Cloud reads before doing a task

158:13

Right

158:14

so the same way that you give instructions

158:15

to a chef to make some sort of food

158:17

you give Cloud Code the skills to then make better PDFs

158:19

better websites, better presentations, and so on

158:21

And so, for example

158:22

let's say you give Cloud Code a task

158:23

to make a website

158:24

Cloud Code, which is the agent

158:25

will look at its own skill that we

158:26

just added to build effective websites

158:28

and it will give us a much better

158:29

result rather than us not giving it a skill

158:32

right? Because without a skill

158:33

Cloud guesses, and it's inconsistent in the results

158:35

whilst with a skill

158:36

it follows a pattern and it follows instructions

158:38

and it gives us a much better output

158:40

And you can find these skills at AITMPL.com

158:43

I'm going to leave the link down below

158:44

in case you want to check it out

158:45

And here, we have over 683 skills

158:48

and there's more added every single day

158:50

And these skills, like senior backend, web app testing

158:53

Architect, code reviewer, git commit helper

158:55

are the

158:56

attachments or the things that we can add

158:58

to Cloud Code for it to be able

158:59

to do a better job at the thing

159:00

that it needs to do

159:01

So I'm going to be working with this

159:02

agent of work that I just mentioned

159:04

which gives me a full presentation of all the competitors

159:06

the analytics, and everything like that

159:08

And so I'm going to be using this tool right here

159:10

which is Canvas Design

159:11

which allows me to be able to give

159:12

Cloud Code

159:13

the

159:14

extra skill of making better presentations and better

159:16

PDFs and visuals for any kind of thing

159:18

that we're doing. So I actually already added

159:19

this skill. I'm just going to go ahead

159:21

and delete it

159:22

I'm going to delete the folder as well

159:23

because then you can see the output before

159:25

and after

159:26

I'm going to say

159:26

can you please make the presentation and give

159:29

me back the result

159:30

analyzing my competitors and the analytics and everything

159:32

like that

159:33

I'm going to press go

159:33

and now it's going to start to do

159:34

what it needs to do

159:35

and I'll catch you when it gives me

159:36

the output. All right

159:37

so I just finished making the presentation with no skill

159:39

and as you can see

159:40

it's very plain and simple

159:41

no effects, no colors, and it's all right

159:44

The actual structure of it is a bit off

159:46

but nothing crazy, right? And so ideally

159:49

we want something that's branded

159:50

that's better in terms of structure

159:52

I feel like I want something Affects your

159:54

I want some colors to make it easier on the eyes

159:56

but also to make it appear better when I'm using it

159:58

with my team and so on

159:59

So ideally, we want something that's a bit more professional

160:02

that has colors, structure

160:04

and the layout looks a bit better

160:05

So what we're going to do here is

160:06

go back to the Cloud Code templates

160:07

I'm going to scroll down to the Canva

160:08

design. And by the way

160:09

if you're looking for a PDF or for a MCP builder

160:12

which I use all the time

160:13

and for other things

160:14

you can just simply search it here and

160:16

you'll find it. You can

160:17

for example, we can look for a website

160:19

web design reviewer, Google Analytics, SEO optimizer, web experience

160:22

a lot more things that are there

160:24

So I can look for Canva design

160:27

Canva design. And if I go to view details

160:30

I will get to this page right here

160:31

to find out that the skill is actually

160:32

a prompt

160:33

So every skill is a prompt

160:35

and the prompt is very detailed

160:36

which allows Cloud Code to be able to

160:38

build better presentations

160:40

based on this full prompt that we have here

160:41

which

160:42

someone made. Now I would recommend looking at

160:43

the amount of downloads

160:44

because if it has good downloads

160:45

then obviously it's good. Not all the skills are great

160:48

right? Depends on the prompt

160:48

and prompting depends on the person

160:50

So all we have to do here

160:51

is copy this code

160:52

I can go back to Cloud Code

160:54

I can clear conversation

160:55

and I can say install this skill here

161:00

And I can press go

161:01

And now what this will do

161:03

is it will start

161:04

adding the skill inside of Cloud Code here

161:05

We just have to approve a few things

161:07

And as you can see

161:07

it's saying that the canvas design skill has

161:09

been installed successfully

161:11

I can go to Cloud

161:11

I can go to Skills

161:12

and here we have canvas design

161:14

And if I go to skill.md

161:15

here

161:16

we have the full prompt that we had

161:17

here on the code. Let's run it again

161:19

So let me go to clear conversation

161:21

and I can say

161:22

okay

161:22

can you now make the presentation

161:24

for the analytics comparing me and my results and my competitors

161:27

and give me back the full presentation

161:29

on a special. I'll press plan mode

161:31

It will now start working on the presentation

161:32

All right, and after a few minutes

161:33

it made this full presentation right here with colors and designs

161:37

and it looks a lot more cleaner

161:39

than the one that we had before

161:40

We have colors, we have different types of colors

161:42

we have a better structure now

161:43

better way that we can look at things

161:45

we have headers

161:46

which is a lot cleaner than the one

161:47

we had before

161:47

I can see there's a few mistakes though

161:49

on the top as well

161:49

but it's something that we can fix just by prompting it

161:52

to do a better job in those specific areas

161:54

and we have until the end

161:56

Now

161:56

Cloud Code skills are by far one of

161:57

the most important things You need to know

162:00

when it comes to Cloud Code

162:01

because Cloud Code is great

162:02

but it's even better when you add enhancements

162:04

It's sort of like the Iron Man suit

162:06

which is amazing

162:06

but it's even better if it has the controller

162:08

if it has the laser

162:09

if it has all these things that make

162:10

it 10 times more powerful

162:11

and it makes you build things at 10 times the speed

162:13

instead of you having to go back and

162:15

forth and prompting it saying

162:16

hey, it's wrong, it's not wrong, make it better

162:18

and things like that. The next hack

162:19

is using the spec to do and code workflow

162:22

Okay

162:22

so now that we added all the structure of the files

162:24

do we go straight to building?

162:25

Well

162:25

no

162:26

We have to plan even more

162:27

if we want to build something that's actually

162:28

scalable and that's more structured

162:30

because yes, you will spend more time now

162:33

but it will save you a lot more

162:33

time later on

162:34

when you want it to actually work

162:36

And the problem with most people using Cloud

162:37

Code is that they paste the prompt like this

162:39

which is, Claude

162:40

build me a family activity finder app

162:42

AI is great

162:43

right? But if you're not specific about what you want

162:45

then it's not going to do a good

162:46

job. And so

162:47

we replace this prompt with something that looks like this

162:50

which is asking it to make a spec.md

162:52

for a family activity finder app

162:54

We have the requirements

162:54

which is parents, enter city, kids ages, availability, distance, willing to travel

162:58

The app returns Five activities with title description

163:01

and emojis. The tech stack that we're going

163:02

to use is simple HTML

163:03

CSS, and JavaScript. And we have milestone one

163:05

which is building the UI with dummy data

163:07

and milestone two, which is connect to real activity API

163:09

Now

163:09

the reason why we do this is because

163:11

we want a detailed blueprint with requirements

163:13

tech choices

163:14

and clear milestones as to what it needs

163:16

to do

163:16

to be able to build this thing

163:17

before we even start building anything

163:19

And so I'm going to press go

163:20

And as you can see here

163:21

the spec.md is sort of like an SOP

163:24

And we told previously to Cloud Code that

163:26

all the SOPs

163:27

should be stored under the directives

163:29

Now it's going to ask me to accept

163:30

this plan. I'm just going to press

163:32

out

163:32

just because I don't want it to do

163:33

anything. And if I go to the directives

163:35

I should be able to now to see a spec.md file

163:37

which is a full file

163:37

Let me zoom out a little bit

163:39

It tells us the overview

163:41

of the exact app that we're going to use

163:43

right? So we have the overview

163:45

tech stack

163:45

user inputs, output, UI layout, milestone

163:48

and we have dummy data examples

163:49

milestone two, and so on

163:51

This gives a lot more clarity and structure

163:53

to Cloud Code later on when you are

163:54

building something. Because bear in mind that when

163:56

you talk to Cloud Code

163:57

you're talking here

163:58

Then when you leave Cloud Code

163:59

When you come back

164:00

this will be empty

164:01

and so it will need some sort of

164:02

memory

164:03

to understand exactly what you were doing in

164:05

the first place. And so you always want

164:06

to make sure that you're as structured as

164:07

possible when building the actual thing

164:09

Now we're done with step one

164:10

which is creating a detailed spec

164:12

Now we're going to go to step two

164:13

which is generating to-do lists

164:14

And so now I can say

164:15

create a to-do list for milestone one based on the spec

164:18

and we'll be able to go through milestone

164:19

by milestone to make sure that everything is

164:21

all right and everything is structured

164:23

and it doesn't build a full thing

164:24

and then we're like

164:24

hey man, not the thing that we wanted

164:26

It's going to make to-dos

164:28

for that specific milestone

164:30

Now it has made all the to-dos for

164:31

the first milestone in specific

164:33

and I can see here that it says

164:34

here's the milestone one to-do list

164:36

create product directory, build index HTML, it can do X

164:39

Y, Z

164:39

and it gives us all the steps in

164:40

specific. Because

164:42

we right now are acting as a supervisor

164:44

for the actual developer that's building the thing

164:45

And now I can say

164:46

build milestone one using the to-do list

164:48

I'm not going to press enter because I

164:49

don't want to build anything

164:50

but I can tell you that once it finishes milestone one

164:52

you can then go and say

164:53

okay, can you not build a to-do list for milestone two

164:55

and milestone three, and milestone four, until you're finished

164:58

And yes, you've taken a bit more time

164:59

Time and to structure things

165:01

but

165:02

you've made the output ten times better

165:04

just because you've been more structured

165:05

Remember that Cloud Code is only as good

165:07

as your input and the prompts that you

165:08

give it

165:09

This is a developer

165:10

but a developer is good at coding

165:11

but it might not be the best to

165:12

understand the context behind the code and what

165:14

it is that you actually want

165:15

which is why we want to be as

165:16

structured and explicit as possible when building these

165:18

agenda workflows or apps. All right

165:20

so now we get to the execution and

165:21

performance hacks. The first one being permission mode

165:24

settings. So down here when we're talking to Cloud Code

165:26

we can actually change the mode or the

165:27

settings

165:28

of the permissions that we're giving Cloud Code

165:29

You can either press and change

165:30

or you can simply go shift and tab

165:32

and it will change automatically

165:33

So the main ones is bypass permissions

165:35

and then we have the plan mode

165:37

Bypass permissions is used when you're making a

165:39

very simple task. This is what we call the YOLO mode

165:41

which is telling Cloud Code

165:42

hey, you do everything, don't ask me any questions

165:44

don't plan, just like go at it

165:45

This is great for something very

165:46

very simple. What's plan mode?

165:47

It lets you collaborate with Cloud Code to

165:49

explore solutions and create detailed structures without wasting

165:52

tokens of going back and forth

165:54

What we usually say is a minute of

165:55

planning saves you 10 minutes of building

165:57

The next hack is using sub-agents Parallel processing

166:00

So what this means in a very

166:02

non-techie way for beginners is that

166:04

when you talk to Cloud Code here

166:05

you're talking to the main agent

166:06

What you can also do

166:07

is talk to the main agent and then

166:08

have other agents working in parallel as all these ones

166:12

which means that you can do

166:13

four times the work at the same amount

166:14

of time. All right

166:15

so here I have my partnerships email inbox

166:17

Let's say I want to process

166:18

all of these emails together

166:20

Of course, I only have about 84 right now

166:22

but

166:22

just

166:23

let's say I had a thousand or two

166:24

thousand or whatever it is

166:25

because most people do have that

166:26

And then instead of having one agent trying

166:28

to process all of them

166:29

you can split the work between agents

166:31

And so what you could do

166:32

is you can split the tasks across these

166:33

four agents. So you can say to this agent

166:36

hey, can you label emails from one to 200

166:39

this one from 200 to 400

166:41

and then 400 to 500

166:42

and so on, right? And so when you split the task

166:44

they all work simultaneously at the same time

166:46

and it'll be so much faster

166:48

Here's my client ID and client secret

166:49

I want to create a email classifier system

166:52

where it goes through my Gmail

166:53

and it basically classifies the emails on whether they're

166:55

it's, it's a full, full caps

166:57

so editor request or software partnership

166:59

So

166:59

I want to Give you my kind of

166:59

decline secret so you can use that and

167:01

then authorize my Gmail

167:03

like ask me to authorize my Gmail

167:04

I'll let you in and then you can

167:05

we can start the task

167:06

All right

167:06

and these four agents just finished running

167:07

I can see here that they've all been

167:09

labeled either a software partnership

167:10

other, or, editor request as well

167:13

Now the example I showed you

167:13

because it's 80 emails

167:15

it wouldn't be realistic

167:15

but if you have thousands and thousands of emails

167:17

then it would work really well because you

167:19

have these agents running at the same time

167:21

doing the exact same thing

167:22

And all you have to do

167:23

is make sure that you're there

167:24

to be able to conduct them and

167:26

answer any questions if you need to answer

167:27

All right

167:27

the next task is going to be the

167:28

control and optimization hacks

167:30

The first one is mastering context management

167:32

So think of Cloud's brain like a notebook

167:34

with only two hundred thousand pages

167:35

right? Every time you chat with Cloud

167:37

it writes down your message

167:38

his response, all the files it reads

167:39

every tool he uses. As the notebook fills up

167:42

Cloud gets dumber. It gets more stupid

167:44

right? Because it's trying to remember

167:45

too much at once from a conversation and

167:47

files. And so if we keep having a

167:48

chat with this agent right here

167:50

it becomes very sluggish because you've used up

167:52

fifty thousand plus tokens just on conversation history

167:55

The solution to this is using the smart reset

167:58

right? The smart reset is when you Slash dot compact

168:02

right? And this compresses your conversation into a summary

168:05

but it keeps the most important stuff

168:06

right? And you use this when your chat

168:08

is getting way too long

168:08

but you want to keep working on the

168:09

same project. And it's compacting the conversation that

168:11

we had here. Of course

168:13

most of this is just

168:14

Cloud Code just working on its own

168:15

But let's say you were to have a conversation

168:17

it would extract all the summary points of this

168:19

and then it would give us the summary

168:21

which it will then use to be able

168:22

to go to the next steps

168:23

As you can see here

168:23

it says

168:24

the session has been continued from a previous

168:25

conversation that ran out of context

168:27

The summary below covers the earlier portion of

168:29

the conversation

168:30

And so this right here

168:31

summarizes the conversation that we had to then

168:33

be able to then keep going with the

168:34

conversation. And once you keep talking to it

168:36

now you have used about 180,000 tokens free again

168:39

but Cloud

168:40

still remembers what you were working on

168:42

And the next command is actually something that

168:43

you've seen me used over time and time again

168:45

which is d slash clear conversation

168:47

Because when I want to have a new

168:48

conversation with Cloud Code

168:49

instead of having to spin up a new agent

168:51

I want to have the same conversation here

168:52

but the conversation will be cleared and you

168:54

can start from zero

168:55

And one more thing you can do

168:56

within the context management is check your context

168:58

usage. So if If you're worried about how

169:00

much context you're using within a conversation

169:02

you can simply press slash context here

169:05

So I actually have to say

169:06

hey

169:06

can you give me the context usage slash

169:07

context? And here it says

169:08

hey, here's your current context usage

169:10

Model is cloud opus 4.6

169:11

Tokens is 55.5 thousand out of 200 thousand

169:14

We're using about 28%. Again

169:16

200 thousand is the maximum of tokens

169:18

that you can use within one conversation

169:20

And it tells us the system prompt

169:21

the system tools, the memory files, the skills, the messages, the free space

169:24

basically giving us an overview of how much

169:26

we can use

169:27

more than what we're using right now

169:28

This context hack is important because it allows

169:30

you to be able to understand exactly how

169:32

much more you have left

169:33

how much runway you have left

169:34

and also how much you've used so far

169:35

And the last hack here is

169:36

using the pixel agent extension

169:38

So there are some of you here

169:39

including myself when I was starting out

169:41

that look at this and think that is

169:42

very intimidating and it looks very

169:43

very complex

169:44

And there is actually a way

169:46

to view

169:47

your agentic workflows or your cloud workspace

169:49

in a much more visual way

169:50

We can go here to extensions

169:52

We can look for pixel agents

169:54

And then you'll choose this one here

169:57

You can press install

169:58

and now this will be installed in your

169:59

workspace. Then we're able to view our workspace

170:01

in this manner right here

170:02

It's sort of like different emojis or Lego

170:04

blocks that show us all the different agents working simultaneously

170:07

which is actually really

170:08

really cool. And the way to find it

170:09

is to go to command shift P

170:10

I can look for pixel agent show panel

170:13

There we go. And on the right hand side

170:14

we can see here that we have a

170:15

visual representation of our workspace

170:17

And if I press new agent on the bottom

170:18

you can now see that this person just appeared

170:20

and he is

170:20

idle because we haven't given him any tasks

170:22

to do. And I can press more and

170:23

more agents

170:24

to add more and more people here

170:26

And these are the different

170:27

agents that we can use to be able

170:29

to run up and spin up different things

170:31

So maybe the first agent will be

170:32

hey

170:32

can you research the web on the top

170:34

AI tools out there in 2026?

170:36

And then the second agent will be

170:38

hey

170:38

can you research the best workflow automation tool

170:40

past 10 years? Hey

170:41

can you do extensive research on human psychology

170:43

and tell me the top 10 trends that

170:45

you see over time

170:46

right?

170:47

And now, using agent needs approval

170:49

so I can see exactly when something needs

170:50

approval. I can go back to the agent itself

170:52

right

170:52

and just give them approval on what they

170:53

need. But all in all

170:54

You're not able to visually see

170:56

your workspace and the agents working at the same time

170:58

all simultaneously through a very

171:00

very visual representation. I do love this because

171:02

it also makes it fun for me to

171:03

work with Cloud Code

171:04

and I can see different things

171:06

And also, in addition to that

171:07

I can

171:08

go to the layout and I can start

171:09

moving things around. I can put the plant here

171:11

or I can put the boxes here

171:12

or I can put the desk here

171:14

or the computer and so on

171:15

in case you're bored and you want to

171:16

change things around for no reason

171:18

But you have some

171:19

you have some options

171:20

you can also choose the floor

171:21

and you can basically create your own game

171:22

and then work with Cloud Code at the

171:23

same time

171:26

Quick one before we get back to the video

171:28

if you're looking to work with me

171:29

want one

171:30

two people to start and scale your own

171:32

AI automation agency in a 90-day one-on-one mentorship program

171:35

then check out the first thing down below

171:36

With that being said

171:37

let's get into it. All right

171:38

so you've now got

171:39

eight proven hacks to build better agentive workflows

171:41

But everything we built so far lives in your machine

171:45

lives in your computer

171:46

which is why in the next lesson

171:47

I'm going to show you the top four

171:48

ways

171:49

that you can deploy your agentive workflows

171:51

onto the cloud. So they run 24-7 without

171:54

you having to touch a single thing

171:55

Let's dive in. All right

171:56

so first of all

171:57

what's deployment? Because I know it's a word that everybody says

172:00

but only a few people actually know what

172:01

it means. But essentially

172:02

when you're building an agent in Cloud Code

172:05

so let's say you want to build an

172:06

agent that can run every single morning at

172:08

3 a.m. When you're testing

172:09

it's hosted on your computer

172:10

That means you tell it when to run

172:12

You press the button

172:13

you extract it

172:14

to run the actual workflow

172:15

to give you the desired output

172:17

The problem with this method is that

172:18

if you close your laptop

172:19

it stops, and if the Wi-Fi goes out

172:21

it stops

172:22

It's sort of like a plan that only

172:23

lives in your desk

172:24

and it can only survive If you water

172:26

the plant. And if you want the workflow

172:27

to run every 3 a.m

172:28

in the morning

172:29

then you would have to be

172:30

in your computer turned on with Wi-Fi to

172:32

be able to make it work

172:33

or else it just wouldn't run

172:34

And so in very easy terms

172:35

deployment means that we're taking this exact agent

172:37

that we built in Cloud Code

172:38

and we're hosting it

172:39

we're putting it in someone else's computer

172:40

And this is great because the computer is always on

172:43

it's always connected to the internet

172:44

but most importantly, it runs 24-7

172:46

So you can actually run the workflow at

172:48

3 a.m. every single day without you having

172:50

to be in your own computer

172:51

Now getting more in-depth into it

172:53

a computer here is called a server

172:55

So this right here is a server

172:56

And servers

172:56

live in big warehouses called data centers

172:59

So data centers right here

173:00

they basically store all these computers

173:04

right? Each one is a computer right here

173:05

And a computer hosts your workflow

173:07

and it's always on

173:08

Okay, that's essentially what it gets

173:09

Now

173:09

the good news about deploying it to these

173:11

servers is that you don't have to buy a server

173:13

and you don't have to know exactly how it works

173:15

because what they do is they rent server time

173:17

so you pay for the time that the

173:18

workflow is in the actual server

173:20

and they handle all the complicated stuff for

173:21

you. So all you have to do is

173:22

build it on Cloud Code

173:23

you have to take the code

173:25

and then upload it in the app

173:26

and then it runs it for you

173:28

and these are the companies that we're going

173:29

to be talking about today

173:30

So here's an agenda workflow that I built

173:32

in Cloud Code. Now

173:32

I'm not going to show you the full

173:33

build step by step

173:34

because if you want the full video

173:35

you can check it out up here

173:36

but essentially

173:37

it's a strategy system

173:39

that runs every single week on a Monday

173:41

and it gives me a full presentation

173:43

basically comparing my analytics with my competitors on YouTube

173:46

and then it gives me content ideas

173:47

So I'm going to show you the output first

173:48

so you can see exactly what it is

173:50

that we need

173:52

I can put slash start

173:53

So I just ran the actual workflow

173:55

All right, so the workflow just finished running

173:56

It gave me the weekly report right here

173:58

that I can copy

173:59

and then I can paste

174:00

in my browser right here

174:02

And this is a presentation that we get every single

174:04

week. So I get analytics report

174:06

which tells me my stats on my channel

174:09

my top performing video right here

174:11

my benchmark comparisons against other competitors in my space with views

174:14

engagement

174:15

and so on. Key insights from my channel

174:16

that I can see

174:17

Then I get five ideas

174:18

So one, two, three, four

174:19

five that I can use

174:21

And then finally I get next steps as

174:22

well. So now we get to the part

174:23

where we deploy the actual workflow

174:25

So there are four main ways to deploy

174:27

the agentic workflow. The first way

174:28

is using modal

174:30

The second way is using railway

174:31

Then we have render

174:32

and then we have replant

174:33

So we're going to start with modal right here

174:35

because it is one that a lot of

174:36

people in the cloud community are already using

174:38

and for good reason

174:39

because it was literally built for AI workflows

174:41

So first thing we want to do is

174:42

press get started. You can then sign up

174:44

through Google

174:44

right here. You can choose your account

174:47

Push continue

174:48

You can do personal

174:49

and then you can get started

174:50

And one important thing here is that we

174:52

get five dollars of free credits

174:54

but you can also have thirty dollars of

174:55

free credits if you add your card details

174:57

So once you add the payment method

174:58

you should be able to see that you

174:59

have now

175:00

thirty dollars of free credits right here

175:02

Now you do get charged fifty cents for the actual verification

175:05

but

175:06

it's worth it because then you get

175:07

six times the amount of money

175:09

that you can use on your own account

175:10

So the first up here is going to

175:11

cloud code and saying

175:12

hey, I want to host the

175:14

agentic workflow to run every morning

175:18

on a Monday at nine AM

175:20

London time

175:21

and I want to host it in modal

175:23

And all I have to do is go back here

175:25

and I can run this

175:26

so I can copy this

175:27

and then simply paste it here

175:29

And I can press go

175:30

We could go down different ways

175:31

but this is the easiest way

175:33

to just stall Cloud Code to figure it

175:34

out

175:35

before it starts adding it to the model

175:37

deployment. So after a few minutes

175:38

you will get Cloud

175:39

which will make a plan

175:40

and it will go step by step

175:41

as to how it needs to go through

175:42

deploying the workflow

175:44

All you have to do is press yes and auto accept

175:45

and now it's going to start executing on

175:47

the actual plan. All right

175:48

so the first thing it does is that

175:49

it creates tasks that show us step by

175:50

step what it has to do

175:52

before starting the plan of the deployment in model

175:55

All right

175:55

and here's the plan that Cloud has to

175:56

deploy the system on model

175:59

We have step one until step For

176:02

All we have to do is press yes

176:04

not to accept

176:04

And you can also read through this if you want

176:06

but of course everything is in code

176:08

so you won't really understand exactly what it

176:09

is. But one important part is

176:10

as you can see

176:11

we just made a file called modal app.py

176:14

which is going to be the Python file

176:16

Not that you should know exactly what it says here

176:18

but it'll be the Python file that it

176:19

uses to then deploy the actual

176:20

workflow on modal. And we also see that

176:22

it made tasks to then deploy to modal itself

176:24

All right, so once everything is done

176:25

you should be able to see that it

176:26

finished all the to-dos

176:27

and it says test runs

176:28

results all passed. And now you can press

176:30

this link right here

176:31

which will take you to the actual app

176:32

in the place as well

176:34

And if you want to see all the apps

176:35

all the different workflows that you have

176:36

you can simply just go to apps

176:38

and then press on each one in specific

176:39

Now here we have the overview

176:41

which is the app function

176:42

so the thing that we actually need to

176:43

do

176:44

Then we have the deployment history

176:45

So let's say you want to change something

176:47

in the actual agenda workflow

176:48

you don't actually go here and change it

176:50

you go back to cloud code

176:51

you go right here

176:52

you then tell it to change things

176:53

and then it automatically updates

176:55

and it makes a new version of the

176:56

actual agenda workflow. And that's what we can see v1

176:58

v2, v1.3, v1.4, and so on

177:01

And you Get to see all the different deployment histories here

177:03

so you can always come back to this

177:05

and maybe change to the previous version

177:06

Then we have app logs

177:07

and by the way

177:08

replace the word app with workflow

177:10

so workflow logs

177:11

So you can see all the different logs

177:13

all the different runs that we had before

177:14

And then we have run weekly YouTube report

177:17

which is the

177:18

action that the actual workflow needs to do

177:20

We get to see all the function calls

177:21

and then up here we get to see

177:23

that it is scheduled for 9 AM and only on Mondays

177:26

London time. And the next one will be February 16th

177:28

as today is Wednesday

177:30

at 1 PM. Now to show you how

177:31

it does look like when it runs

177:33

we can press run now

177:34

and what this will do

177:35

is it will schedule an immediate run

177:36

and we should be able to see

177:38

the run show up somewhere here

177:39

As you can see

177:40

it just showed up

177:41

so acute started, startup, and running

177:43

So now it means that it's executing the

177:44

actual workflow. And just so you know

177:46

the workflow finishes by sending myself an email

177:48

with the actual report every single week

177:50

So we should be able to get an

177:51

email if it succeeded. All right

177:53

as you can see here that it took

177:54

56 seconds for the presentation to actually run

177:57

and I can see the status is success

177:59

We can see the time here

178:00

and if I go to log

178:01

I Can see all the steps that it

178:02

took to be able to make the actual presentation

178:05

Now, to show you that it did actually make the presentation

178:06

I can go to my email

178:08

and I can see that the latest email

178:09

we got is weekly YouTube analytics report

178:11

and I get an email saying

178:13

hey, this is a report

178:14

I can press view presentation

178:15

and this will take me to the actual

178:16

report with all the different

178:18

ideas and so on. As you can see

178:19

it took us less than two minutes to set everything up

178:21

and a good thing is that you are

178:22

now able to run it every five days

178:24

so it will run every single Monday

178:25

but you can also test it the same

178:27

way that you're in

178:28

a workflow platform like Anytan or make.com

178:31

You can simply just manually press this

178:33

and see if it works in real time

178:35

Now, there is a chance that when you do run this

178:36

it doesn't work, which is why you can go inside here

178:39

and it will show you the message as error

178:41

so there's something wrong with it

178:42

and you can simply copy this

178:44

so copy the whole thing

178:46

bring it back to Cloud Code

178:47

and say, hey, I just ran this on modal

178:49

it didn't work, it gave me this error

178:50

and you can paste it here

178:51

okay? Now, of course, this is success

178:53

so I'm not going to run this

178:54

but if you have an error

178:55

that's what it will show you

178:56

You copy this, you paste it into the Cloud Code

178:58

and you let Cloud Code

178:59

debug and self-heal its own workflow

179:01

The pricing

179:02

the way that modal works is that we

179:04

are actually able to pay per second

179:06

of the workflow running. Now

179:08

there are three plans

179:09

The first one is starter

179:10

team

179:10

and enterprise. The good thing about modal is

179:12

that it is very easy to set up

179:13

Plus, because we get $30 of free credits

179:16

we're pretty much able to run this every

179:17

single month for completely free

179:19

unless you have some astronomical

179:21

automation and you have like 50 agents that

179:22

are all running at the same time

179:24

But you're most likely not

179:25

You get three workspaces

179:26

you get 100 containers

179:27

and 10 GPU concurrency

179:29

Now

179:29

what this means basically is the size of

179:30

the automation to the space that you have

179:32

So I definitely recommend starting with the plan right here

179:34

and just know that it is free because

179:36

you get $30 of free credits

179:37

So who is modal for?

179:38

Now

179:39

modal is great for beginners who want the

179:41

easiest path

179:42

to deploy and maintain a workflow built in

179:44

Cloud Code

179:45

It's also best for people who have a

179:46

workflow that is Python-based

179:48

which it probably is

179:49

since you're using Cloud Code

179:50

It is also best if you have a

179:51

workflow that is not running 24/7

179:54

but it runs in bursts

179:55

So maybe it starts the automation

179:57

when it gets triggered by something

179:58

so it can be every single day at 9 AM

180:00

it gets triggered. And also it's Great

180:01

if you want to pay only for what

180:03

you use. So the second way that you

180:04

can host your workflow is using a software

180:06

called Railway. Now Railway is a slightly more traditional hosting platform

180:09

but it is still incredibly simple to use

180:11

Think of it as a very friendly modern

180:13

version of what developers used to use

180:15

when deploying workflows. All we have to do

180:16

is press deploy

180:18

And we can see here that we have

180:19

the option of GitHub

180:20

database, template, Docker image, function, bucket

180:23

and empty projects as well

180:24

So we're going to be using GitHub

180:25

Now GitHub is a software that we now

180:27

have to make an account as well

180:28

which is a software where we basically upload the code there

180:31

and it's going to be the middleman between

180:32

Cloud Code and Railway itself

180:35

So the first step is going here

180:36

and going to github.com

180:39

Actually, here's GitHub

180:40

and think of this as your Google Drive

180:42

folder

180:42

So we're going to host all the different

180:43

files in the Google Drive folders

180:45

and then we're going to connect the Google

180:46

Drive folders to Railway

180:47

so it can look inside the code

180:48

and then be able to host it there

180:50

So first things first is you want to

180:51

sign up

180:52

You want to continue with Google

180:55

Choose your email

180:56

Press continue

180:57

Create an account

180:58

And once you've made an account

180:59

you will get on a page that looks

181:00

like this. So the first step that we

181:02

want to do is you want to create a repository

181:04

because we're making the Cloud Code agent

181:06

we're hosting it on GitHub

181:07

and we're deploying it on Railway

181:09

We can create a repository

181:10

we can name this agentic workflow

181:13

we can do public

181:14

everything else is fine

181:15

we can create the repo

181:17

and then you will get on a page

181:18

that looks like this

181:19

All we want to do here is you

181:20

want to copy this URL

181:21

bring it back to Cloud Code

181:23

and say, hey

181:24

can you

181:26

upload all the code to this GitHub repo?

181:29

And then you want to paste the link

181:30

Okay, so we're saying, hey

181:31

take all the code

181:32

of this and connect it to GitHub repo

181:35

so that we can then use Railway here

181:37

so that we're able to see the code

181:38

in GitHub

181:39

through Railway. Because there's no way of connecting

181:41

it directly to Cloud Code

181:42

that's why we're using this

181:43

and I can press go

181:44

Alright

181:44

so here we have the plan to push

181:45

code to the new GitHub repository

181:47

We can press yes

181:48

auto-accept

181:49

Alright

181:49

so now it's telling me that I need

181:50

to get something called an SSH key

181:53

Now, if you've heard of API keys or passwords

181:55

then this is exactly what it's asking us

181:57

So you can press add key to account

181:59

submit answer

182:00

So what you want to do here is

182:01

you want to go to github.com slash settings slash keys

182:04

and you will see the

182:05

SSH and GBG keys

182:08

All you have to do is press

182:09

new SSH key

182:10

You can name this MacBook

182:12

Authentication key is fine

182:14

And then make sure to ask Cloud Code

182:15

to give you the actual key that you

182:16

can paste

182:17

Once you have the key

182:18

you can copy it

182:19

you can paste it here

182:19

and then you can add an SSH key

182:21

and that's pretty much all you have to

182:22

do here. I can go back and say

182:23

okay, I just finished and added it there

182:25

Keep going

182:26

And it says that all my code has

182:27

been pushed to my agent to workflow

182:29

so I can press this link right here

182:31

Which will take me to my repo

182:33

which is this one right here

182:33

Or if you're stuck on here

182:34

you can simply go to three dots

182:36

you can go to home

182:37

and you'll be able to see that you

182:37

have the agenda workflow here

182:39

that you can then see

182:40

So all the files that we have here

182:41

are all the files that we have here

182:43

And now I can go back to railway

182:44

I can press GitHub repo

182:45

I can log in with GitHub

182:47

I can authorize the railway app

182:50

And here you can press I agree

182:52

I will not deploy any

182:52

you can press this button right here

182:54

And you want to press configure GitHub app

182:56

All repositories, press install unauthorized

182:58

And then you want to press new

183:00

GitHub repo

183:01

And now you'll be able to see the

183:02

Mikata partnerships one

183:04

Press this right here. It's deploying the repo

183:06

in Railway. As you can see now

183:07

there was an error

183:08

So I can press this right here

183:09

It's error creating build plan with Railpack

183:11

So what I would do here by instinct

183:13

is go back to Cloud Code and say

183:15

hey

183:15

I've tried to upload it to GitHub

183:18

and I got everything working

183:19

I got all the

183:20

all the files. And then I tried putting it

183:21

everything into Railway, so connecting Railway to GitHub

183:24

And it said error creating build plan

183:26

with Railpack

183:27

So I don't know what to do here

183:28

It's going to ask me

183:28

what do you want Railway to do with

183:29

this project? This will determine how we set

183:31

it up

183:31

We can run it on a schedule

183:34

So we can just

183:35

yes and auto accept

183:36

the plan that he uses to then fix

183:38

the railway. All right

183:39

so just fix this so I can go

183:40

back here and I can see that this

183:41

is now building. So it just fixed this

183:43

thing for me by actually modifying the GitHub repo

183:46

which automatically updates the railway app

183:48

And if I go here

183:49

I can see that I have building status

183:51

As you can see here

183:51

the deployment was successful. And now we have the agentic workflow

183:54

which is online

183:54

which will now be triggered automatically

183:56

and is uploaded and deployed

183:58

If I go back here

183:59

I can see that now I have

184:00

the actual app in production

184:02

Now bear in mind that this

184:03

is also running on a cron timeline

184:05

which means that we are running it every

184:06

Monday at 9am UTC. Up here you have the architecture

184:09

which is where you can actually edit the

184:11

agentic workflow. Then we have this cabra here

184:13

observe the environment

184:14

So you can monitor different projects and look

184:16

at the metrics

184:17

Then we have log

184:18

which is very similar to the one that

184:19

we were talking about before

184:20

which is essentially all the logs that we

184:21

have here

184:22

And whenever something goes wrong

184:23

whenever there's an error

184:23

you can always go here and copy this

184:25

and then paste it into Cloud Code

184:26

and it will fix it automatically

184:28

And then we also have settings here as well

184:29

All right, so in terms of pricing

184:30

we have the $0 per month plan

184:32

which actually is an A zero dollar a month plan

184:35

because we start with a 30-day free trial

184:36

with five dollars credits

184:38

and then it's one dollar per month

184:39

Now, if you're confused about vCPU

184:41

what that means is basically how fast the

184:42

computer can think

184:44

So

184:44

one vCPU means that you only get one

184:46

brain for working on your tasks

184:48

whilst 0.5 gigabytes RAM

184:50

is how much the computer can remember all

184:52

at once. Basically like a shorter memory or

184:53

the space that it has to contain everything

184:55

And I would recommend this plan to start out

184:57

but as you start to build more and more agents

184:59

you ideally want to go to the hobby plan

185:01

just because there isn't enough RAM and CPU

185:03

but also there's rate limits

185:04

on the things that you can actually build

185:06

So for one agent

185:07

it's okay. For multiple

185:08

I'll recommend going to the hobby

185:10

but you also get a $20 a month

185:11

plan as well. Now

185:11

when it comes to railway itself

185:13

it's best if your workflow needs to run

185:14

continuously. So maybe it needs a bot that

185:16

runs and checks the workflow 24-7

185:18

railway is good for that

185:19

It's also good if you want the database support built in

185:22

in the software

185:23

and it's also good if you like the

185:24

idea of putting everything in GitHub

185:26

and then automatically deploying it on railway

185:28

Now, in terms of debugging

185:29

so when something goes wrong

185:30

the way that it tactically works so you understand is that

185:33

so let's say there's an We go to Cloud Code

185:35

we say, hey, there's an error

185:36

Cloud Code fixes the code

185:38

which then

185:38

goes to GitHub and updates the code with

185:40

the new updated version

185:41

And then

185:42

it goes to Railway and redeploys the actual

185:44

agentic workflow. All right

185:45

so the next platform is called Render

185:47

Render is often described as the modern Heroku

185:50

And if that means nothing to you

185:51

then just think of it as a really reliable

185:53

no surprise hosting platform. So all we have

185:55

to do here is you want to get

185:56

started

185:57

You want to go to Google

185:58

use your account, continue

186:00

and you will get on the page that

186:01

looks like this. You want to press on background workers

186:03

as this is the agentic workflow builder

186:06

And then you want to connect this to

186:07

GitHub. So this is very

186:08

very similar to Railway in the sense that

186:09

we use GitHub

186:10

as a middleman between Cloud Code and the

186:12

deployment

186:13

We press GitHub here

186:14

We can authorize this

186:16

so authorize render. And when we go back

186:17

we can install. And right here

186:19

I can see that now I can choose

186:20

my repo

186:21

To choose this here

186:23

I can name this here

186:23

Python three, leave this all the same

186:25

And now I have to choose between one

186:26

of the plans. Because render does not have a free plan

186:29

you either start with a seven dollar a

186:30

month plan or you don't start at all

186:32

So render, in contrast to the other ones

186:33

does not have a free plan

186:35

so you have to pay for the seven

186:36

dollar a month plan

186:37

if you want to use this software right

186:38

here. We can go here

186:39

we can deploy background worker

186:41

and now this will take you to this right here

186:43

where it's now going to GitHub

186:44

it's extracting all the code

186:45

so it's taking all the code

186:46

and it's building the actual workflow

186:48

and hosting it and deploying it inside

186:50

render right here. All right

186:51

and after a few minutes

186:52

you can see that it is now successful

186:53

It says the service is live

186:55

and we're now able to use this workflow automatically

186:57

and now this is hosted easily

186:58

on render right here. If I go to logs

187:00

I'll be able to see all the logs

187:01

of the actual workflow

187:02

and metrics here is where you can track

187:03

the memory and CPU that the workflow is using over time

187:06

and you also have other things as well

187:07

here

187:08

And now the workflow is going to be

187:09

triggered at nine AM every single Monday every week

187:12

as that is will be added to GitHub

187:14

The process of debugging is the exact same

187:16

as the one that we had on railway

187:18

You basically go to cloud code

187:19

you fix the code

187:20

which then updates the code

187:21

GitHub

187:22

and it finally redeploys on Render right here

187:24

So who is Render for?

187:26

Render is great if you want predictable flat

187:28

rate pricing for your workflows

187:29

It's also good if you need background workers for your agents

187:31

so basically sub-agents for your main actual agent

187:34

to work. It's a platform that has been

187:35

around for a while

187:36

and it has been battle tested

187:37

So if it's important to you to have

187:38

a platform that has a good reputation

187:40

and has been for a while

187:41

then Render is your choice

187:42

And also it's great if you value automatic

187:44

rollbacks when something breaks. All right

187:46

and the last platform here that we're going

187:47

to be using to deploy is called Replit

187:50

Replit is one of the most popular platforms

187:51

as well because it does actually have the agentic feature

187:55

Very

187:55

very good

187:56

So first thing you want to do is

187:56

press create an account

187:57

choose your email, continue, put your name

188:02

First next, personal project, that's fine

188:05

And now we can go straight to pricing

188:06

So we have the free version here

188:08

where you can publish one app

188:09

Then we have the Replit Core

188:10

which is the most recommended one

188:11

is the one that most people use

188:12

where you can publish more apps and you

188:13

have more money as credits that you can

188:15

use. And then finally

188:16

we have the Teams plan as well

188:17

And it's always something that you use

188:19

if you're trying to build something more complex

188:20

and you have more and more users

188:21

in your organization

188:23

So I'm going to go ahead and start

188:24

with the starter

188:25

And here we have the app right here

188:26

I can press import code or design

188:28

I can go to GitHub

188:30

I can connect my GitHub account

188:31

I can press continue to GitHub source control

188:34

I can authorize Replit integration

188:35

You just connect in my GitHub here

188:37

I can choose a gentle workflow

188:39

And I can simply press import from GitHub

188:41

As you can see now

188:42

the workspace is loading

188:43

and it's now importing all the code

188:44

from GitHub to Replit to then be able

188:46

to deploy the agent here

188:47

As you can see now here

188:48

it says status okay

188:49

service is YouTube Strategist

188:51

and it says the app is running successfully

188:52

Let me verify

188:53

a response correctly and then set up the deployment

188:55

All right, so now it's finished

188:56

and it says everything is set up and

188:57

running. Let me finalize the import

188:59

and it basically imported all the different files

189:00

here

189:01

And all we have to do is press published

189:03

and if I go back to the homepage

189:04

I can see that I have the agenda workflow

189:06

which is the one that we have right

189:07

here. If I go to apps

189:08

I can see that I have the agenda workflow

189:10

which is private

189:11

and this will now be able to run

189:12

every 9 AM in the morning

189:14

on a Monday

189:15

Replit is great if you want to code

189:17

and deploy in the exact same place

189:18

It's also best to use if you're building

189:20

a quick prototype or a proof of concept for something

189:22

because it has the agenda feature which allows

189:24

us to vibe code different softwares as well

189:26

And it's also great if you don't want

189:27

to deal with GitHub itself or terminals or

189:29

any setup at all. So those right there

189:31

were the four main ways that you can

189:32

deploy and maintain your agenda workflows that you

189:34

build on Cloud Code. Now

189:35

one important thing is that no matter what platform you choose

189:38

the debugging workflow is the exact same

189:39

So here's the process. Your workflow breaks

189:42

You go to the platforms dashboard

189:43

you open the logs

189:44

because every platform has logs

189:45

you find the error message

189:47

which is usually in red

189:47

you copy the error message

189:49

you go to Cloud Code

189:50

you paste the error message and say

189:51

this is the error from my deployed agent

189:53

fix it

189:54

Cloud Code fixes the code

189:55

you redeploy

189:56

which means that you push to GitHub or

189:58

you run the deploy command

189:59

and you check if it works

190:00

if not, go back to step one

190:02

Which makes it so much easier for you

190:03

as you don't have to go in the code yourself

190:05

you can simply let Cloud Code take care

190:06

of the whole thing. So if you're still

190:07

not sure as to which software you should

190:08

go with when it comes to deploying

190:10

then here is the thought process

190:12

If you're someone who says

190:12

I want the absolute easiest option to deploy

190:15

well, modal is your best option

190:17

If you're someone who says

190:18

my agent needs to run 24-7 like a background process

190:21

then Railway and Render

190:22

will be good for you

190:23

If you're someone who says

190:24

I want predictable monthly bills with no surprises

190:26

then Render is good. If you say

190:27

I want to code and deploy without leaving my browser

190:30

then Replit is the way

190:31

And if you say

190:31

I just want to test something quickly for free

190:33

then you can use modal

190:34

Now we do also have some other options like AWS

190:37

GCP, or Azure

190:39

The only reason why I did not include

190:40

them is because they're more for Price grade level

190:42

deployment, and they're more advanced as well

190:44

And that right there is a complete picture

190:45

because building the agenda workflow was just the

190:47

first half. Deploying is everything that comes after you build it

190:51

and it's arguably the most important step

190:53

as building it is fun

190:54

right? But how do you actually give it

190:55

to the client?

190:56

How do you deploy it so it actually

190:57

works without you being there all the time?

191:00

And a good thing is that to deploy

191:01

an agenda workflow with all this code

191:02

you don't have to be a developer

191:04

All you have to do really is just

191:04

follow step by step what I did in this video

191:07

and let it run on its own

191:13

Right, congrats on finishing module four

191:15

At this point, you have everything

191:17

I mean, the foundations, the customizations, the prompting, the agenda workflows

191:21

and the deployment itself. So at this point

191:23

you're not just someone who uses Cloud Code

191:25

but you're someone who understands how it actually works

191:27

but also how to build with it

191:29

Now it's my turn to prove it to you

191:30

because in the final module

191:32

you're going to watch me build

191:33

real products live from scratch

191:35

using everything that we covered so far

191:36

There's no theory

191:37

just building, making it really

191:39

really practical for you

191:40

Let's dive in. All right

191:41

so when it comes to my YouTube content pipeline

191:43

there are four different stages

191:44

The first one is the ideation

191:46

which contains three different types of

191:48

we call them skills or automations that are

191:50

built inside of Cloud Code

191:51

The first one here is called the comment scraper

191:53

which is where the automation itself will go

191:55

to

191:55

X amount of videos posted

191:57

by X amount of people in X amount of time

192:00

and then extract all the comments and then

192:02

give me ideas for content

192:02

Twitter monitoring is where it's going to the

192:04

actual Twitter profile. It's extracting all the tweets

192:06

and gives me ideas

192:06

And Outlier videos is the videos that have

192:08

gone well for other channels

192:09

Why have they gone well?

192:10

How can we apply this same idea

192:12

to my channel? Then we have packaging

192:14

which is title one thumbnail

192:15

So for example, this video right here

192:16

we have a title

192:17

there's a picture before

192:18

that made you want to press on the actual video

192:20

and that's the packaging

192:21

Then we have the video itself

192:22

which is the hook

192:23

So the first 30 seconds of this video

192:25

that was a hook

192:26

that was scripted. There's a script

192:27

which is what I say in throughout the video

192:29

Of course, a video like this

192:30

there's no script

192:31

I go off the rip

192:32

right? But something that's more like a head talking video

192:35

where I go through a concept

192:37

right? Where it's me standing up

192:38

Well, that's scripted, so we have a script

192:40

And then we have visual diagrams

192:42

So this right here

192:43

for example, is built by an automation

192:45

I didn't actually build it myself

192:46

That's diagrams

192:47

And then lastly, we have post video

192:48

which is looking at the performance of my videos

192:50

but also views to clicks

192:51

Views to clicks

192:52

is looking at the amount of people who

192:54

clicked to watch my video going through my program

192:57

versus the amount of views

192:58

right? Because obviously on the back of my videos

193:01

I want people to work with me one-to-one

193:02

I want to help them

193:03

I want to help them scale their agencies

193:05

So I want to check

193:05

which videos do people like the most that

193:07

make them want to

193:08

kind of want to work with me one-to-one

193:09

That's what it is. So with that said

193:11

Here is my cloud code workspace

193:12

It's called YouTube workspace

193:13

It's consisted of different folders

193:15

that we have. The first one is dot cloud

193:17

then we have context

193:18

executions, outputs, references, scripts, dot env

193:21

and then the main one here

193:22

which is cloud.md. Now

193:23

I actually separated the

193:25

folders into four different categories

193:27

The first one is commands

193:28

then we have skills

193:29

then we have context

193:30

and then we have the cloud.md

193:31

I'll explain what these are as well

193:33

but not a main priority

193:34

like they are main folders that we use

193:36

okay? So the first one here is called

193:37

commands. So commands are reusable slash commands

193:40

like slash competitor analysis or slash create plan

193:43

or slash something else

193:45

that allow you to trigger

193:46

to start any workflow just by calling it using a slash

193:50

right? And when I say workflow

193:51

I mean it triggers a skill

193:53

So if I want to run the Excalidraw diagram

193:55

for example, right?

193:56

Then I would do slash Excalidraw diagram

193:59

and that would call that specific skill instantly

194:01

Now, if you're not sure what skills are

194:03

first I recommend you watch this video up

194:04

here

194:05

but in case you are aware

194:06

skills would be modular instructions

194:08

that we give cloud to

194:10

Explain

194:11

or to give it an overview

194:12

of how to do a specific task

194:14

step by step. So for example

194:16

Excalidraw is a platform like Miro

194:17

where you build kind of visual diagrams

194:19

like I showed you before

194:20

Well, inside the skill

194:21

there are step-by-step instructions as to how to

194:23

do that specific thing. And so all in all

194:25

our cloud is going to be our kind of company

194:28

right?

194:29

And inside the company

194:30

there's different employees

194:31

which are responsible for different things

194:33

And those employees are skills

194:35

Then we have context

194:36

so everything the cloud needs to know about

194:38

my business. So first of all

194:40

who I am, my strategy when it comes to the content

194:43

my data, and my content guidelines

194:44

Because for example

194:45

let's say we're using the skill to script my videos

194:48

right? I'm making a script

194:49

When I want to make my script

194:50

I also want cloud to use my YouTube

194:52

content guidelines and business info to check

194:55

hey, this is the script

194:56

this is the guidelines

194:58

are we sure we're following the right guidelines?

194:59

That's all it is

195:00

And then finally, we have the cloud.md

195:02

which is a master file

195:03

here that ties the whole workspace together

195:05

telling cloud how all the pieces work in one system

195:08

Okay, so it's all just called

195:09

hey, we have the folder here

195:10

which is called Context

195:11

the folder here, which is called skills

195:13

and it gives it an overview of the

195:14

whole organization. The reason that is

195:16

is because it will look at the cloud.md

195:18

every single time before it does anything

195:20

That way it has context as to

195:22

your workspace and how it all ties together

195:23

Now, first things first is the ideation

195:26

So ideation consists of the comment analysis report

195:29

It consists of the competitor analysis

195:31

which is the outlier videos

195:32

the videos that have done really well for

195:34

other channels. And then we have Twitter cloud

195:35

monitoring. I'm going to run it first

195:37

and then I'll walk you through how it's

195:38

set up as well

195:39

So dot comment analysis report

195:42

and then we can do bypass permissions

195:44

because I don't want to tell it

195:45

yes, approve, approve

195:45

approve. I just want Cloud Code to do its own thing

195:47

because it knows everything already

195:49

As you can see here

195:49

it says that it fetched 1,369 comments across

195:52

five different channels

195:54

Now it's reading through all the comments

195:55

It's looking at the YouTube content guideline

195:57

very important. All right

195:58

so now the skill finished running

195:59

it said it found 1,369 comments

196:01

It made the PDF that I can see

196:03

and it gave me the top five content

196:05

opportunities that I can do

196:06

So the PDF looks like this

196:07

where we have my channel

196:08

top four content ideas requested by your audience

196:10

By the way

196:11

this is the comments that you guys add

196:12

to my videos. And then what I also

196:14

did is I looked at Nate Hurk

196:15

looked at the top four content ideas from his audience

196:17

looked at Liam Audley

196:19

looked at Nick Sarive

196:20

and then Jack Roberts as well

196:22

right? And we have

196:23

the ideas as well. And then he gave

196:25

me the cross-channel themes

196:26

So what are the common requests

196:28

or ideas that people want across all channels?

196:30

For example

196:31

how do I actually deliver AI work to

196:32

clients?

196:33

This came from my channel

196:34

Liam and Nate, or token management, cost optimization for cloud code

196:38

Nate, Jack, and Liam, and so on, right?

196:40

Plus

196:41

it also gave me content opportunities

196:43

specific to me. So how to deliver AI products to clients

196:46

complete guide. This is the biggest gap in

196:47

the market. Nobody's making comprehensive content about the delivery side

196:50

account setup, hosting, handover documentation, maintenance agreements

196:54

access management

196:55

You've done this for real clients

196:56

make the definitive video

196:58

right? And we go step by step

196:59

Then we have the stats at the end with subscribers

197:01

videos, comments, and average comment per video

197:04

Guys

197:04

I believe

197:05

you should be commenting more on my videos

197:07

so we can get the numbers up

197:08

But hey, comparison is the thief of joy

197:10

Am I right?

197:11

anyways, we have the full PDF

197:13

which I can use now

197:14

as my way of generating ideas for content

197:17

which are really, really good

197:18

And inside the skill

197:19

we have the skill.md

197:20

which is the full instruction manual for the skill

197:23

which goes through step one

197:24

fetch comments

197:25

Step two, analyze the report

197:26

Step three, generate PDF, and that's it

197:29

And it gives me the PDF itself

197:30

Then we also have scripts

197:31

which are the different Python files

197:33

the code that basically allows the automation to

197:35

run. Think of this

197:36

like if you know N10

197:37

if you've heard of N10 or make.com

197:39

think of this script

197:40

each script is like a workflow

197:41

right? And so we are running workflows

197:43

The first one is fetching comments

197:45

So we're running the workflow to fetch comments

197:46

to extract comments from videos

197:48

Then we run the generate PDF

197:49

as another workflow to

197:51

basically make a PDF that we can use

197:53

And then we have references

197:54

which is the report format

197:55

is giving the cloud code a

197:57

a way to understand

197:59

what the format is of the report itself

198:01

right? So we have raw stats

198:02

we have cross-channel themes

198:04

which is how we make the report look

198:05

like this. How does it know the structure?

198:06

And how is it consistent with the structure?

198:08

Well, it has instructions for that

198:09

All right, so now that we're done with that

198:10

We can go to the outlier videos

198:12

So here I can press competitor analysis

198:16

there we go

198:17

And then I can do bypass permissions

198:19

so I can let it run

198:20

So the first thing it does

198:21

is that it looks at the CSV file

198:23

which is called competitors.csv. And this is a

198:25

CSV file that I have down here

198:27

which is a list of 43 channels

198:31

including mine

198:32

which

198:32

are going to be the channels that we

198:34

use to extract the past 10 videos that

198:36

they've done. And then we can start doing

198:37

the math

198:38

as to

198:38

what videos have done really well for them

198:40

compared to the other videos that they've done

198:41

All right

198:42

now it's saying that it just finished and

198:43

exported everything to the Google Sheet

198:45

I'll show you exactly what that means

198:47

We found four outlier videos across 43 channels

198:49

a lean week

198:50

all scored 10 out of 15

198:52

Cool. And then we have key patterns

198:53

most actionable for you. But if I go

198:55

here to my views to clicks Google Sheet

198:57

where I host everything

198:58

we can see that we have the channel name

199:00

subscribers, I think it's a bit blurry

199:02

but it's fine. Video URL

199:03

the amount of views

199:04

the upload date, the duration of the video

199:06

the median views

199:07

so the

199:08

average views that they got

199:09

over the past 10 videos

199:10

And then we have the all-or-multiple

199:12

So the all-or-multiple is a mathematical equation that we use

199:16

which is how many views this video get

199:18

over

199:19

the average views of their channel to see

199:20

if this exceeded that

199:21

right? And then it also gives me a

199:22

relevance score. So for example

199:24

this video over here

199:25

which was about MCP is dead

199:26

and nobody saw it coming

199:29

yeah

199:29

gave a zero relevance score because

199:31

it's not aligned with my brand or audience

199:33

That's what I mean by cloud code using

199:35

my brand guidelines and my

199:37

the person that I'm targeting

199:38

what they care about

199:39

and what that looks like

199:40

And what we have relevance scores are three

199:42

three

199:42

and two. So not only does this save

199:43

me hours and hours and hours of time

199:45

of me having to go through the videos

199:47

having to look at the views

199:48

having to average it out

199:49

having to make the equation

199:50

but it also gave me the relevance that

199:52

I could use for my channel

199:54

And so here I could do

199:55

is that to A?

199:56

I can look at these two are the

199:57

top ones

199:58

Let me look at the ideas

199:59

how to use cloud code work

200:00

how to get unlimited website clients

200:01

And so I could do a video based on that

200:03

in my kind of way

200:04

And this skill right here is consisted of

200:05

references. So this is the scoring and output

200:08

which is a kind of document that goes

200:09

through how to score Outlier video

200:11

what does that look like?

200:12

And then we have

200:12

again, the scripts, which are the different workflows

200:15

and then we have the skill.md

200:16

which is the step-by-step instructions

200:18

that we go through

200:19

to be able to actually make this work

200:21

And then we also have google.token.json

200:23

which is the way that we are able

200:24

to access the YouTube API

200:25

but also my Google Sheet account

200:27

Then last one we have is called Twitter

200:29

Cloud Monitor

200:30

So Twitter Cloud Monitor is a workflow

200:32

that goes to configuration profiles

200:35

Yeah

200:35

These profiles in Twitter

200:37

and it might look a bit kind of weird

200:39

but by the way

200:40

Cloud wrote this name

200:41

and it looks at these channels

200:43

not channels, these Twitter profiles, right?

200:45

And what it does is that it goes through each profile

200:47

it extracts the tweets from the past 24 hours

200:50

right? And would have this workflow running every single day

200:52

and it gives me the new updates on

200:54

call code. Because honestly

200:55

I can't keep them up myself

200:56

and I'm sure you cannot too

200:58

So that's exactly how a lot of YouTubers

200:59

do get their ideas

201:00

is through Twitter

201:01

right? So why not make an automation that

201:02

can automate that process

201:04

of you having to find the tweets yourself?

201:06

So I can go here and put slash Twitter monitor

201:09

Cloud monitor. I can do bypass permission

201:11

Right, it just finished running

201:12

It said it scanned through 63 tweets

201:14

The relevance score above three is just two

201:16

and then we have the key findings

201:18

and it also made the PDF

201:19

which was sent to my email

201:20

If I go to my email

201:21

I can see that I have a new

201:22

PDF

201:24

I can open it up

201:25

and it can give me now

201:26

The new features that we have here

201:28

and then we have the content opportunities that we can use

201:31

right? Now

201:32

are all of them going to be great ideas?

201:34

No, right? But it filters them out

201:36

Because not every day Cloud Code releases something

201:38

maybe every two days or every three days

201:40

So it looks at the tweets

201:41

it checks the relevance

201:42

and then it gives me the report as

201:43

well. So I basically keep myself updated without

201:45

having to go to these 12 different profiles

201:47

every single day and check what they've posted

201:49

That's how you get the YouTubers making videos

201:51

about the new slash command or the new

201:53

slash body

201:53

or the new slash executions or the new slash whatever

201:56

right?

201:57

It's because they look at these exact profiles

201:59

that I walked through. Now that we're done with the ideation

202:01

let's go to the next one

202:02

which is packaging. So packaging is consisted of

202:04

thumbnails and titles. So let's go through thumbnails

202:06

first. So the first one is thumbnail creation

202:08

So the way that this skill works is

202:09

that we take a video

202:10

so let's say we have a video like this

202:13

let's copy the video

202:14

copy the link, and then we go back to Cloud Code

202:18

and I can put slash thumbnail creation

202:22

There you go. And then what we do

202:23

is

202:24

I took the link of the video with

202:25

Alex Hermosi. It will look at my face

202:26

right, that we have a picture of

202:28

which is going to be here

202:29

reference, it's right here. That's my face

202:32

And then it will take that face

202:33

it will replace Alex Hermosi's face in that

202:35

thumbnail with my face

202:36

and then give me back the thumbnail

202:37

basically. What source would you like to use

202:39

as thumbnail? I'm going to put

202:40

the link here

202:42

of that video that I copied

202:43

What it's going to do now is it's

202:44

going to go inside this video

202:45

it's going to download the thumbnail

202:46

and then it's going to go to my picture

202:48

swap the face

202:49

and then give me back the image

202:51

And this is using a server called file.ai

202:53

because

202:54

that's where we're going to use the

202:56

nano banana model to be able to make

202:58

this happen. All right

202:58

so it just finished the thumbnail

202:59

I'm going to say

203:00

can you open up the image?

203:03

And to be honest with you

203:05

it didn't do the best job

203:06

right?

203:10

And replace my face with his face

203:12

Obviously, that's what he looks like

203:13

I don't have a beard

203:14

as you can see

203:15

But it copied the words

203:16

it made the words good

203:17

It made these massive traps that he has

203:20

and it put my face there

203:21

So ideally, now we can go and say

203:23

hey

203:23

can you remove the beard from my face?

203:26

But it swapped the face initially

203:28

And by the way

203:28

this is all prompting. So

203:30

one thing I could say is like

203:31

hey, color code, it actually added a beard

203:33

Can you remove the beard?

203:34

Hey, color code

203:35

can you

203:35

remove the beard from

203:37

my face in that

203:38

thumbnail and give me back the thumbnail?

203:39

And also

203:40

replace the body with like a black shirt

203:41

as well. Keep everything else the same

203:42

All right, it just changed it

203:43

and here we are

203:44

much better thumbnail here. It had no beard

203:46

and it put me in a better place

203:47

as well. So that's the thumbnail creation

203:48

which I think is nuts

203:49

because you can keep iterating it

203:51

you can keep saying

203:51

hey, can you change the text?

203:53

Can you change this? Can you change that?

203:55

But essentially, we're taking inspiration from other channels

203:57

and then using it for our own

204:00

Now

204:00

I don't recommend copying the exact same as someone else's thumbnail

204:03

but you can always copy

204:05

the way of thinking or taking inspiration

204:07

on the idea of the package

204:08

Right? Now with that said

204:09

we go to the next one

204:10

which is the title generation

204:12

Let's say I was going to do the

204:13

title generation

204:15

I'm going to wait for it to

204:16

ask me any questions. So now it's saying

204:18

what's the video content you'd like to titles

204:19

for?

204:20

Any specific angle or hook you want?

204:22

How many titles? You can do 20 titles

204:23

that's fine. So the video is about how

204:25

I'm using Cloud Code to automate my YouTube

204:28

channel that I've grown to 51,000 subscribers

204:31

about 2 million views

204:32

in the past 7 months

204:34

That's what the video is about

204:35

I'm going to go to bypass permissions

204:37

just to make sure that it does everything

204:38

As you can see here

204:39

it's looking at the title frameworks

204:40

because we have a list of title frameworks

204:41

that work really well. So 300

204:43

that it's using

204:45

It's then looking at the YouTube content guidelines

204:47

which is my MD document

204:49

which goes through my guidelines and my

204:51

kind of avatar that I'm looking for

204:52

And then it's going to give me 20

204:53

different titles

204:54

So how I automated my entire YouTube channel with Cloud Code

204:57

I used Cloud Code to grow to 51

204:59

subscribers. Here's how

205:00

How Cloud Code runs my YouTube channel

205:02

51K subs in 7 months

205:04

I let AI run my YouTube channel 2 million views later

205:06

and it does a lot more

205:08

Which I think are really

205:09

really, really

205:09

really good. And then it gives me the

205:10

top titles based on curiosity

205:12

specificity, but also emotion. All right

205:14

with that said, we're done with packaging

205:16

Now we can go to the video itself

205:17

All right, so for the video itself

205:19

we have script writing

205:20

which is the hooks plus the script of

205:22

the video. And then we have

205:23

two different Excalidraw diagrams and visuals that I'm

205:25

going to walk you. So the first one

205:26

is script

205:28

So script writing, let's go here

205:30

Let's wait for it to ask me any

205:31

questions about the video. Target length

205:33

any specific key points to cover

205:35

hook style and preferences

205:36

and CTA goal. I'm going to start writing

205:39

There we go

205:39

So with all this said

205:40

I can press go

205:41

and then I can do bypass permissions

205:44

let it do its own thing

205:45

And while this is running

205:46

I can show you the actual references

205:48

which we have here

205:49

So here we have the two different videos that I've done

205:52

so head talking videos. So previous examples of

205:54

my scripts that have worked really well

205:56

Then the intro framework. So

205:58

what's a good hook?

205:59

What does that look like?

206:00

So we give it references

206:01

Script writing framework. So this is a

206:03

YouTube video which talked about

206:05

the best storytelling

206:06

or the best ways to write scripts

206:08

Then we have this skilled.md

206:09

which again is a step-by-step instructions

206:12

as to what it needs to do

206:13

So step one, gather information

206:15

Step two, load context. Step three

206:17

build the audience avatar. Step four

206:19

write the hook. Step five

206:20

write the body. And then step six

206:22

write the mid-roll CTA. And step seven

206:25

write the closing CTA

206:26

Step eight, add production notes

206:28

Step nine, quality check

206:30

And then we have the output

206:31

which is insane for AI to do all of this

206:32

because previously we have to spend so much time on everything

206:35

but now it does it in seconds

206:37

There we go

206:37

So now we have the script

206:39

I can go here and I can see

206:40

that this is the full script

206:41

What I typically do

206:42

because this looks very confusing

206:44

is I copy this and I paste it

206:45

in Notion

206:46

Like this

206:47

And now we have the hook

206:48

It took me five months to get my

206:49

first AI agency client

206:50

five months of building systems that nobody asked for

206:52

sending messages that nobody replied to

206:54

and honestly

206:55

questioning whether the whole thing

206:56

was even going to work

206:57

And then in 14 months

206:58

I scaled the same agency to six figures

207:00

working with over 50 businesses

207:02

and teaching over 22,000 people in the process

207:04

all starting with zero business or technical knowledge

207:06

I can read the whole thing

207:06

obviously I'm not going to

207:08

in case you were wondering

207:08

But we have section one

207:09

which gives me the script itself

207:11

And I'm reading this

207:12

it's very

207:13

very good. And you can see it also

207:14

gives the editor a

207:16

kind of

207:16

a reference of what

207:17

edits would need to be here

207:18

Text on the screen

207:20

B roll, pause, section two, section three, closing CTA, and that's it

207:25

Now bear in mind that

207:26

you can use a general AI to make

207:28

your script. There's a difference between a general

207:29

slop AI script

207:30

and actually making a good script

207:32

And that's typically using references

207:34

using what's a good script look like

207:36

using human language, using good transitions

207:38

And so that's what we feed a lot

207:40

of information to clock code to tell it

207:41

hey

207:42

here's a video that worked really well

207:44

in the past

207:45

Here's a framework of how To make good scripts

207:47

here's the idea

207:48

now give it back to us

207:49

And this also looks at the context

207:51

So the personal info

207:52

the strategy, the content guidelines, like I mentioned before

207:54

All right, so now that we've made the script

207:56

there are times in the video

207:57

where we need supporting visuals

207:59

That would be a diagram

208:00

that would be some sort of

208:02

yeah

208:02

a diagram or an illustration of a concept

208:04

So for example, here in the Cloud Code Skills video

208:06

we have diagrams that look like these

208:08

we have this, all of these, where skills are stored

208:11

this, and it keeps on going

208:13

right? More and more diagrams

208:14

And so here I can say

208:15

hey

208:16

can you make an Excel draw a diagram

208:17

based on a video that I'm doing about

208:19

sub-agents? So make a diagram as to like what sub-agents are

208:23

why they're important, how they're different from agent teams

208:27

and that's it. Make them really good

208:28

sophisticated, and give it back to me

208:30

I'm going to go and bypass permissions

208:31

There you go

208:32

So now what it's going to do

208:34

is it's going to read the skill first

208:36

So let me read the Excel draw diagram

208:37

first. As you can see

208:39

Cloud Skills Excel draw diagram

208:41

Excel draw diagram dot nd

208:42

is looking at the instructions

208:43

and now it's going to give us back the code

208:45

which is essentially The diagram itself

208:47

All right

208:47

so it just finished making the diagram

208:48

I can go here. I can copy this

208:50

So command A

208:51

and then bring it back and paste it here

208:53

God damn, look at this

208:55

Beautiful. So sub-agents, architecture, and purpose

208:58

What are sub-agents? Why they matter?

209:00

And sub-agents versus agent teams

209:01

I mean, guys, come on

209:02

This is insane for it to be made

209:04

in a few minutes. It made some errors here

209:06

but nonetheless, this is crazy

209:08

And you can instantly make a video about

209:10

sub-agents as soon as you've made the diagram

209:12

Now, this saves a lot

209:13

a lot of time because sometimes when I'm explaining a concept

209:16

I want to show a diagram

209:17

I want to show some visual representation so you can hear

209:21

but also see at the same time

209:22

because you retain more by doing both

209:23

So that's where I use this

209:24

Now, let's say

209:25

can we also add some visuals

209:27

using the Excalidraw visual skill?

209:30

Yeah, add some visuals there as well

209:31

There you go. So now it's going to

209:32

take the diagram. It's going to then add visuals

209:34

and then we'll see the difference as well

209:35

All right

209:36

so it just finished making the three different

209:37

images

209:38

So I can go to the Excalidraw

209:39

and I can drag them across from here

209:41

to here

209:42

And here we have the three different visuals

209:45

So you see how this is a diagram?

209:47

But this are images

209:49

That's exactly what the Excalidraw visual skill does

209:53

Which one's better?

209:54

It kind of depends on you

209:55

Sometimes I like to have a mixture of both

209:57

but nonetheless

209:57

they save me a ton of time

209:59

whenever I have to make a video

210:01

and make diagrams for that video

210:02

All right, so now that we're done with the video assets

210:04

let's go to the last one

210:05

which is post video data

210:07

So here on Google Sheets

210:08

I have a list called the views to clicks

210:10

which is out of this many views

210:12

I get this many clicks

210:13

and this is a CTR

210:14

And then we have the type of video

210:15

Now the reason we do this is because

210:17

we want to see out of this many views

210:19

how many people click to see the video

210:22

of my one-to-one program to then be able

210:23

to work with me

210:24

right? And we do this because

210:25

typically the more value that you provide

210:27

the higher the CTR is

210:29

okay? So in this case

210:30

if I have more clicks

210:32

and I have more data as to what

210:33

videos people like or people want

210:35

then I can provide a shit ton of

210:36

value to you guys and then be able

210:38

to help you guys even further

210:39

right? So now you might be asking yourself

210:41

how do you do this?

210:42

So manually

210:43

the way that this works is that if

210:44

I go to one of my videos

210:45

you can press a link

210:46

Now this link, you can see Bitly

210:47

Bitly is a software that pretty much all the YouTubers

210:50

that I know kind of use

210:51

to be able to track how many people

210:53

click to the next thing

210:54

And then obviously you click to

210:56

kind of watch the video of like the full one-to-one program

210:59

how I help you to be able to

211:00

start and scale your own AI agency

211:01

right? And so here

211:03

I see how many people clicked out of

211:04

this many views

211:05

So out of this many views

211:07

how many people press this button?

211:08

And it's very tedious because I have to go to Bitly

211:11

I have to see the views

211:11

and it takes me a lot of time

211:13

especially if I have to do this weekly

211:15

for these many videos

211:16

And the videos just keep on adding and adding and adding

211:18

right?

211:18

And so what we do here

211:20

is I have a automation that I can do

211:22

I can put slash views to clicks analysis

211:26

And what this will do

211:27

is that it will look at my videos

211:28

so all my videos in my channel

211:30

since a specific date where I started tracking

211:32

So now it's running the automation

211:33

and now it said that it updated the

211:35

sheet successfully. So if I go back

211:37

I can see that this is now all in different places

211:39

and I can delete these two columns

211:41

I don't know why they're here

211:42

there you go. And I can do sort Z to A

211:44

and I can see the CTR here

211:46

So this is the videos where it had

211:47

a higher CTR rate

211:48

right? There's actually BTC had talking and BTC

211:50

had talking. And I can start to see

211:52

hey, these many videos on the bottom

211:54

they don't do too well in terms of the CTR

211:55

but these videos do

211:57

And so

211:58

let me make more of those videos and

211:59

less

212:00

of these videos. And that's one of the

212:01

ways where we strategize what content idea works

212:04

better and what content ideas don't work better

212:06

based on my audience

212:07

my offer, and who I'm targeting

212:11

Now let's take it one step further

212:12

because in the final lesson

212:14

I'm going to show you

212:15

how you can build and deploy

212:17

$10,000 websites using Cloud Code

212:19

Let's dive in

212:20

So to give you like a full recap

212:21

a high level of what we did

212:22

we made an image that looks like this

212:24

of this Ferrari

212:25

F1. Then we made a video that looks like this

212:27

where we have the components of the car

212:29

breaking up. To then make a website that looks like this

212:31

where we have Born to Perform

212:32

Seven Decades of Fury

212:34

Forza Marinello

212:35

Then we go down

212:36

you can see the car breaking up and going back in

212:39

which is super, super, super sick

212:41

And by the way

212:41

this right here is the video

212:42

So we have the video incorporated in the

212:43

websites. You can also see here Prancing

212:47

which is the, the horse of the Ferrari

212:49

Because now we can see maximum speed

212:51

we can see zero to 100 km per hour

212:53

top speed here. We also have the final page

212:55

which is experience a legend

212:56

reserve now

212:57

So reserve the spot for the

212:58

for the actual car. Okay

212:59

so that's a full website that we're going to build

213:00

and we're going to go through everything step

213:01

by step. Now there's a lot of moving

213:02

components for this exact websites

213:04

because there's different places where we get data

213:06

make the image, make the video

213:07

which is why we're going to break it

213:08

all down into a four-step framework

213:10

The four-step framework is called plan

213:12

execute, review, and deploy

213:15

All right, so the first one here is plan

213:16

Okay

213:16

so here's a product that I use to

213:18

be able to create the website that I

213:19

showed you. As you can see here

213:20

I have the video

213:21

of the Ferrari that we used inside the website

213:23

And essentially, to be able to start everything

213:26

we need two different things

213:27

We need something called the cloud.md file

213:29

and we need the second thing

213:30

which is two different skills

213:31

So first thing is the cloud.md file

213:33

Now

213:33

what is it? This right here represents the

213:34

definition quite in a

213:36

in a very quick way

213:37

which is, are you tired to repeat yourself?

213:39

Tell cloud to remember what you've told it

213:40

using the cloud.md file

213:42

So essentially

213:42

it's a way for us to give cloud

213:44

memory of the project

213:45

Because every single time that we're talking to cloud

213:47

it needs to have context as what it

213:49

is that we're doing. So this right here describes the project

213:51

which is a landing page for Ferrari

213:53

Then we have the file structure

213:54

So when it makes the landing page

213:56

where does it store everything?

213:57

What folders? Where do the files go inside

213:59

the folders?

213:59

And it gives structure to the whole project

214:01

Then we have more things specific to the actual landing page

214:03

the design

214:04

So these are the different designs that we

214:06

use. Animations and external dependencies

214:09

coding conventions and constraints

214:10

Now, a lot of the things that are here

214:12

I don't fully understand

214:13

right? Because I'm not super technical

214:14

But Use AI to craft the Cloud.MD file

214:17

which makes it easier for us to be

214:18

able to do it. So once you have the Cloud.MD file

214:19

which gives Cloud Code the memory

214:21

then now we have the two different skills

214:24

So

214:24

skills are basically ways for us to

214:26

give Cloud Code

214:27

a series of instructions to make something better

214:30

So for example, we need to make a landing page

214:32

and we need to make a video

214:33

that fits well inside the landing page

214:35

So the first skill that we need is

214:36

called front-end design. Now a skill

214:38

again, is just a list of instructions

214:40

So this skill right here tells Cloud Code

214:42

hey, here's how to create a nice looking UI

214:45

so user interface, basically a landing page

214:47

which

214:47

increases the output

214:49

for the landing page that we're going to

214:50

make. And so every single time we mix a landing page

214:51

it looks at the skill

214:52

it looks at all the components

214:54

how it needs to think about design

214:55

the front-end guidelines, and so on

214:57

And so it will use the skill

214:58

when making the landing page

215:00

Now your question might be

215:01

how do I even get this skill?

215:02

If you go to slash manage plugins

215:05

You'll be able to see this skill right here

215:07

called front-end design

215:08

You can press install

215:10

and you have three options

215:11

Either you install it available for all your projects

215:13

which means that you don't have to install

215:15

it again for any other project

215:16

or install it for this project in specific

215:19

But bear in mind that if you do that

215:20

every single time that you start a new project

215:22

you will not be able to see the skill

215:23

and you'll have to install it again

215:24

Once you've installed it

215:25

you should be able to see

215:26

either here or here. Make sure you refresh your cloud code

215:29

and you'll have the skill.md

215:30

which is the instructions itself

215:32

The second skill is called the video-to-website skill

215:34

Now

215:34

this right here is not something you'll be

215:36

able to find in the plugins

215:38

so

215:38

you can find it in the resources that

215:39

are left below. But essentially

215:40

this

215:41

gives cloud code

215:42

instructions as to how it should take the

215:44

video and turn that into

215:46

something that you can use within a website

215:49

And it's very, very long

215:49

as you can see here

215:50

List of instructions. So once you have those two things

215:52

we are good to go

215:53

to the next steps. Before we do that

215:54

I'm in a new project

215:55

with the two different skills

215:57

and a cloud.md to keep it very

215:58

very simple

215:58

as you can follow along with me when

216:00

I do this. Now that we're done with planning

216:01

we can go to step two

216:02

which is execute. Okay

216:03

so when it comes to execute

216:05

We have to ask ourselves

216:06

what do we need

216:06

to be able to make this landing page

216:08

happen?

216:08

So I firstly need an image

216:10

which then turns into a video

216:11

and then also one more thing that I

216:12

want to use is a reference landing page

216:14

so that we can feed it all into Cloud Code

216:16

and it uses the reference landing page

216:18

it uses the video

216:19

and it uses the skills to be able

216:20

to craft us the beautiful landing page

216:22

that we're going to use

216:23

Now, to be able to make the image and the video

216:24

we can use a software called Kube API

216:27

which is one of the best softwares that I've used

216:29

because it gives us access to all the video models

216:30

image models, in the fastest way

216:32

and it pretty much has all the options

216:33

that you need. So first thing is to make an account

216:35

and then you can go to market to

216:36

be able to add your API credits

216:38

because this is not free

216:39

It costs credits to run

216:40

these images and videos

216:42

So first thing is to make an image

216:43

and I'm going to be using the Nano Banana 2 model

216:45

which is a new model that Google Gemini made

216:47

because it's fast and it's cost efficient

216:49

And down here

216:50

we can write the prompt that we need

216:52

So I'm going to be copying a prompt

216:52

that I already made. There we go

216:54

Which says that I need a high-resolution studio-grade

216:56

image of a Formula One racing car

216:58

positioned against a completely plain black background

217:01

The scene should be a minimal and isolated

217:02

with no shadows, reflections, hands, or

217:05

Objects visible, the car should appear sharply detailed and professionally lit

217:08

as if photographed for a premium automotive studio

217:11

shoot. And then we can remove this

217:13

we can go down to 4K

217:15

aspect ratio, we can do auto

217:17

automatic is fine, and I can press run

217:20

Bear in mind that if you're wondering how

217:21

much does this cost

217:21

well, you paid $5 for 1,000 credits

217:24

so it costs you 18 to be able to do it

217:26

which means that you're paying about $0.09

217:29

to make the image. All right

217:30

so it just finished making the image

217:31

as you can see here

217:32

very, very clean. Now we can download this

217:34

and we can go to the next step

217:35

which is making the video

217:36

So I can go to AI Video API

217:38

I can go to Kling 3.0

217:39

There's also other image or video models that you can use

217:42

but this one is by far

217:44

the best that I've seen

217:45

I can take this out

217:46

and I can drag the image across

217:49

Boom

217:50

So this right here is the input image

217:51

I can take out the sound

217:53

We're not gonna add any end frame

217:54

So let's say you're making a

217:56

a website about a cup

217:57

right? Just some sort of cup

217:58

You would have a cup with no water

217:59

and then a cup with water

218:00

That's the start frame

218:01

and then

218:02

end frame

218:02

So you can basically see the animation of

218:04

the water going inside the cup

218:06

But in this case

218:06

we don't need it. And now when we

218:07

get to the prompt

218:08

we can go to ChatGPT

218:10

I can paste this prompt right here

218:11

which says that it's an AI

218:13

video prompt engineer. You can stop this video

218:14

and read it

218:15

But I'm gonna

218:16

then

218:17

give it the image

218:18

of the Formula One racing car as an input

218:20

and then I'm gonna press run

218:21

and now it's gonna give me the full

218:22

prompt that I'll need

218:23

to be able to make the video

218:25

Now, one thing, I do like this prompt

218:27

I just think it's a bit long

218:28

so I can say

218:29

can you shorten this

218:31

a little bit and make it more concise?

218:35

But sometimes when you provide too much context

218:37

or when the prompt is too long for an AI

218:39

it does a worse job

218:40

Okay, cool, that's fine. So I can go back here

218:42

we have the Kling 3.0

218:44

and I can paste this prompt

218:45

and now the duration will be five seconds

218:47

that's fine, and I want to make sure that the mode

218:50

so the actual resolution is higher

218:52

so I can press pro

218:52

because the resolution is basically how clean or

218:55

how nice does the video look

218:56

We can press run now

218:57

as you can see this is taking 135 credits

219:00

which is about 0.625 dollars

219:03

that we're spending on the video

219:04

Alright, so just finish making the video

219:05

let's see the video here

219:09

It's even better than the one I had before

219:11

Cool, so now we can download this

219:13

save this to your desktop

219:14

I can go back to Cloud Code

219:16

I can then give it the video as the input

219:19

okay?

219:20

And I can rename this Ferrari video

219:25

There we go. Now that we have the video

219:26

there's one more thing that we have to do

219:28

which is getting a reference landing page

219:30

Because yes, we can make a landing page

219:31

but if we give it stricter guidelines

219:33

it does a better job

219:34

So if you go to dribble.com

219:35

you can find the Charger product landing page right here

219:38

which is a very

219:39

very nice landing page that we can use

219:40

as reference. So you can go here

219:41

you can save video as reference

219:45

I can press save

219:46

and then I can drag it across the

219:48

cloud code as well

219:49

So now we also have the video

219:50

of the landing page as a reference

219:52

Finally, I can go to cloud code

219:54

make sure this right here is in plan mode

219:55

and then you can paste this here

219:57

and you will say

219:58

I just dropped in a video

219:59

which is the Ferrari

220:02

Where is Ferrari video?

220:04

Right go

220:05

Create a one product landing page for this

220:06

product. The site must feel modern

220:07

premium

220:08

and highly professional

220:09

with smooth animations and polished UI design throughout

220:12

All text should be clean

220:13

minimal

220:13

and easy to read

220:14

using high contrast typography on a pure black background

220:17

dark mode

220:17

The layout and visual styling should be seamlessly

220:20

blend with the Ferrari image

220:21

making the entire page appear continuous and fluid

220:23

with the background. And then I will be able to say

220:25

I also added a video

220:27

which is a reference video that I want

220:29

you to look at in terms of the

220:30

actual design of the landing page itself

220:32

And now I can press go

220:33

All right, so as Cloud Code is actually making the plan

220:35

you can see that it's reading the skills

220:37

So skill two, skill three

220:39

when making the plan itself

220:41

That's what we use in skills

220:42

All right, so I just finished making a plan

220:43

Context, framing extraction, file structure, design tokens, page architecture, and you're looking at this

220:48

you're like, what the hell even is this?

220:49

The same here, so don't worry

220:51

JavaScript initialization order

220:53

I'm wondering why they have to make such

220:54

complex words for all of these

220:55

man. So

220:56

let's go and press yes and auto accept

220:58

And then we can turn this to

221:00

bypass permissions

221:01

because I don't want to be approving every

221:02

single thing. Bear in mind that

221:04

Pass permissions is a bit more dangerous because

221:06

it takes over and does everything

221:07

but I usually use it when I'm actually

221:09

looking at cloud code myself

221:10

I never

221:11

go to sleep and then let cloud code

221:12

do its thing and then come back

221:13

right? All right

221:14

so I just finished making version one of the landing page

221:16

and bear in mind that on the left-hand side

221:18

you will be able to see new folders

221:19

that were created. And if I go to

221:21

the Cloud.md file that we had before

221:22

we had CSS folder

221:24

we had JS folder

221:25

and we had frames folder

221:26

And here you can see CSS folder

221:28

frames folder, and JS folder

221:29

So that's what we mean by file structure

221:31

Where are the files going to be stored?

221:32

CSS is going to be all the different styles

221:34

that we're going to use

221:35

so all the colors

221:36

and so on

221:36

I'm only saying this because a lot of

221:37

you guys are beginners

221:38

and are getting into this and don't really

221:40

know what all this code stuff is

221:41

but on a fundamental level

221:42

that's essentially what it is

221:43

Putting files into right folders

221:45

and each one doing a specific thing

221:47

You don't have to know what the code

221:48

actually is. So I can copy the link

221:50

and I can paste it in my browser

221:55

That's nice. Born to perform

221:57

so we can see we have the

221:58

the little

221:58

page before we get in the website

221:59

We have this here

222:01

the animation, the car being broken into different pieces

222:05

and then coming back in

222:07

Perform follows function

222:08

Okay

222:10

The speed, maximum power

222:13

Just sick

222:14

So a bit of writing

222:14

giving us some context

222:16

and that's it. To be fair

222:17

it did actually a very good job for

222:18

it to be like a first time

222:20

kind of thing. All right

222:21

so now we can go to the next step

222:22

which is called

222:23

review

222:23

Now reviewing it means that we have to

222:25

look at the website and say

222:26

okay

222:27

do we want any changes

222:28

with the way that it's laid out?

222:29

I can already see a few changes that

222:30

I want. So for example

222:31

I want something happening in the background here

222:34

So some sort of lines or

222:35

or dots. I have a vision of it

222:36

but I have to see

222:37

Then right down here

222:39

I would want the car to be positioned

222:40

on the right-hand side

222:41

and then the writing to come only from the left-hand side

222:44

not the right-hand side. Because when it goes over the car

222:45

it doesn't look nice

222:46

So we can take a change by change

222:48

The first change is having something within the

222:50

background here. So I can go to 21st.dev

222:52

which is a software that we use

222:54

to pull bits and pieces of different website animations

222:56

motion graphics, and so on

222:57

that we can use for our website

222:59

So

222:59

I said I wanted something in the background

223:01

So what we could do

223:01

we can look here

223:02

shaders

223:03

We can try this one

223:04

Background paper shaders

223:06

We can copy the prompt

223:08

And I can go back to Cloud Code

223:09

So we can see we're using 54% of

223:11

the context. And by the way

223:12

the more context you use

223:13

the worse the quality gets

223:15

So I can put clear conversation

223:17

Go to backups permissions and I can say

223:18

okay

223:19

so I'm looking at the website and

223:21

I want you to implement

223:22

the following thing at the start

223:24

of the website. And I can paste the

223:25

prompt

223:26

So essentially 21st Dev

223:27

to be able to get this

223:28

we have the prompt that allows Cloud Code

223:30

to understand exactly what it needs to do

223:32

and how it needs to update the code

223:33

And as you can see here

223:34

it's a full, full thing

223:35

I can press

223:37

Bye, fresh permissions. You can see here

223:38

Because honestly, I just wanted to update

223:40

and that's it. It just finished implementing the change

223:42

so if I go here to the website

223:44

I can refresh. And you can see now in the background

223:46

it's very, very subtle, which I really like

223:48

but if you zoom in

223:50

you can see that we have

223:51

different things changing in the background

223:52

which makes it a lot nicer to

223:54

to look at

223:55

visually. Now that's the first thing

223:56

The second thing is putting the car to

223:58

the right-hand side

223:59

and having the text go on the left-hand side

224:01

Okay

224:01

so

224:02

second thing that I want you to change

224:03

is on the website itself

224:05

the car takes up the whole space

224:07

What I want to happen instead

224:08

is I want the car to be on the right-hand side

224:11

so I want it to take like basically

224:13

75% of the whole screen from the right to left

224:17

75%

224:18

And then the other 25% that's left on the left-hand side

224:20

then I want that to be the space

224:22

where text comes through

224:23

because I don't want text to overlay over the car

224:25

if that makes sense. Keep the animation

224:27

keep the video, but make the car smaller

224:29

Yeah

224:29

position it on the right-hand side of the

224:30

screen. I'm going to go on plan mode for this one

224:32

just because I want to make sure that

224:33

it understands what I'm talking about

224:35

And a lot of this is just talking

224:36

to

224:36

Going back and forth as to what it

224:38

is that you want. All right

224:39

so here it made the plan for the change

224:42

the context implementation plan

224:43

25% to the left. Okay

224:45

good. Moving everything to the left

224:47

the car to the right

224:48

everything looks good. But now I can press

224:49

yes and auto accept

224:51

and we can go back to

224:52

bypass permissions to have Cloud Code make the changes

224:54

All right

224:55

so now I just finished making the change

224:56

I can refresh

224:57

and we can see that all the text

224:59

has been on the left-hand side

225:00

And then this also is fine

225:02

I can go down

225:03

and as I scroll through

225:03

we can see that all the text

225:05

is on the left-hand side

225:06

which looks a lot better than us having

225:08

to have the text over the car

225:09

and that covers the whole website step by

225:11

step. Now that we're done with step three

225:12

we can go to the last step

225:13

which is

225:14

deployment. So the way that we deploy

225:15

Cloud Code websites is through a two-step process

225:18

The first thing that we do

225:19

is we push all the Cloud Code into GitHub

225:21

and then from GitHub

225:22

we deploy it to Vercel

225:24

So the first thing we want to do

225:25

is go to github.com

225:28

and want to make an account

225:29

Once you made an account

225:30

you can press new

225:31

we're making a new repository

225:32

and we can name this website Cloud Code

225:36

We can press create repo

225:38

And so basically what's happening is that

225:40

we're going to be using Cloud Code

225:41

and we're going to be giving all the

225:42

code here into GitHub

225:43

and from GitHub it's going to go to

225:45

Vercel to then deploy it there

225:46

All I have to do here is press copy

225:48

and I can go

225:49

I can clear the conversation

225:51

I can say, okay

225:53

now I want to push this to GitHub

225:55

Here's the GitHub link

225:56

repo that I've just made

225:57

I can paste this. And I can go bypass permissions

225:59

because I know it will do a good job

226:00

because it's a pretty fast task

226:02

It should tell us to log in into GitHub

226:03

because I didn't actually give it any credentials

226:05

here. So you can see here

226:06

GitHub needs your identity configured

226:07

Run these two commands in your terminal

226:09

replace with your actual name and email

226:10

I don't know how to do that

226:11

Can you tell me how I should do it

226:12

or what I should give you to make

226:13

that happen? I do this just because

226:15

a lot of you guys don't know how

226:16

to use a terminal

226:16

so that's fine. But in this case

226:18

what I'm going to do is give them my name

226:21

and then the email that I'm using

226:25

All right

226:26

now it's saying that it's all live

226:27

So if I go here

226:28

I can refresh

226:29

and I'll have

226:29

all the different files and folders

226:31

of this product

226:32

They are automatically added here

226:34

And the last step here is to deploy

226:36

it on Vercel

226:38

I can press start deploying

226:39

I can continue with GitHub

226:40

And once you connect to GitHub

226:41

make sure to choose the

226:42

correct product that you're working with

226:43

In this case, for me

226:44

it was website Cloud Code

226:45

I can import, and I can press deploy

226:47

And as you can see here

226:48

well, now we have our product

226:50

so our, website deployed. And now it says

226:52

congratulations

226:53

you've just deployed a new product to Mikela

226:54

Pass projects

226:55

I can press continue to dashboard

226:57

and if I use this link right here

226:58

you can see that I have the website itself

227:00

which is sick

227:02

Boom

227:06

There we go. And this literally took two minutes to deploy

227:09

which is absolutely insane. And now for the

227:11

final thing is answering the question

227:13

what if you want changes?

227:14

What if you're here and you want to

227:15

change to a website? You have to go to Vercel

227:17

you have to go to GitHub

227:18

Well

227:18

the beautiful thing about

227:19

GitHub and Vercel and the whole process

227:21

is that you just have to go to Cloud Code

227:24

you have to tell Cloud Code what you want to change

227:26

and that will automatically update the GitHub

227:29

which will automatically update Vercel

227:31

And that's the whole flow of things

227:32

To make sure that you don't have to

227:33

go to two different platforms

227:34

at the same time and do any extra

227:36

stuff. You can just talk to Cloud Code simply

227:37

and it will automatically do everything for you

227:40

All right, guys

227:40

so congrats on finishing the full Cloud Code Masterclass

227:44

where you went from a complete beginner in Cloud Code

227:46

all the way to building agentic workflows

227:49

deploying them

227:50

and using Cloud Code to build

227:51

actual products from scratch. Now everything

227:53

all the resources and templates of this video

227:56

are going to be stored in the second

227:57

link down below. In my private community

227:59

you can go to the classroom section

228:00

go to the templates vault

228:01

and you'll find everything there

228:02

All right

228:03

so now you've watched how to build With

228:04

clock code. Your next question might be

228:06

how do I build a business with it?

228:07

Which is why I've put together a full

228:09

AI agency masterclass that you can check out on the screen

228:11

which goes through everything that you have to

228:13

know as a complete beginner

228:14

to make money with AI

228:16

With that being said

228:17

I really hope you found value from this video

228:19

and I'll see you in the next one