🐙 Building Your First AI Agent

Okay. Hello. Hello everybody.

Uh I hope that my OBS is working. Is it

okay? Good morning.

Good evening. Where are you?

Hello friends.

made myself. It's 2:00 a.m. Wow. Okay.

That's quite

It's quite I was like, is it quite late

or quite early? I don't know. It's it's

it's it's certainly an interesting time.

CVS, hello. It's been a while. Hello,

Elizabeth. I am from Peru. Thank you so

much for everything you share here.

Well, thank you so much for joining.

Hello everybody.

Oh, it's so nice to see such familiar

names. Oh, I hope you guys are doing

well. More like good night. Okay, good

evening. Okay, good evening. Good

evening. Good morning. Um,

good afternoon. I think somebody's in

the afternoon right now. Good strange

time of night. Anywhere between 12:00

a.m. to 5:00 a.m. Strange times. Hello

my strange time friends. Can you

increase the volume a bit? Yeah, sure. I

can I can do that. Actually, I think I'm

at max gain right now. So, it's like

that's the best I can do. I think

yeah, this is maximum gain already.

I hope you can hear me. I guess I can

move it closer to my mouth. It just

looks like I'm like speaking into a mic

like this. Is that better? Okay.

Hopefully it's better. Um, sound is

good. Okay, cool. All right. Shall we

get started? Let's get started. Instead

of me rambling, let's actually get

started.

I shall share my screen.

Okay.

That's not right.

Why is it showing that screen? Technical

difficulties today.

It should be showing that.

I'm confused. Okay, let me try again.

Okay, so this is the YouTube.

Okay. Oh, why is it so small?

What?

Oh, BS is not my friend today. That's

okay. I will

make it really big.

Okay, I think that more or less you guys

can see that.

Okay, so today we are going to be

covering your building your first AI

agent. I want to ask in the chat, put in

the chat, have you built an AI agent

before? Yes or no question. Yes, you've

built an AI agent before. No, you have

not built an AI agent.

Is this really live? I hope so. I mean I

am doing it live. I'm currently alive.

Don't make me doubt my own existence.

You know at some point somebody was like

Tina are you real? And I was just like

that's a very shocking question. I had

never considered that before. And then I

had to like confirm with myself. I'm

like ah yes I'm pretty sure I'm real.

Yes I am live and yes I am real. Oh

dear.

Okay. Okay. Good mix. Good mix. Okay. A

lot of people have never built a AI

agent. We have a few here.

Typical star of relationship. That's not

right. Everything is terrible. Yes,

exactly.

No. Okay. A lot of people have not.

Okay, great. This is wonderful. So, in

this case, this is going to be a good

start for you guys to to actually start

building your AI agents. Okay, let us

get started doing this.

Okay, so what we're going to be covering

today, so I'm going to cover a framework

for how to approach building AI agents

from scratch. Uh, we're going to be

covering the landscape. So, overview of

popular tools. I'll be covering some no

code tools and code tools as well. Uh it

you can do it with both. So if you have

the misconception that you have to know

how to code in order to build an AI

agent, that is not the case. There's

definitely pros and cons to this and

we'll cover that as well, but it is

possible and a lot of people do it um

who do it with no code as well. Show you

some live demos of a few tools and then

also ending with some practical tips as

you build your first AI agent. So by the

end of the session, the goal is to have

you you'll have a clear road map to

building your first AI agent. Uh

knowledge of some popular tools, so some

choices what you're going to build it

with and some practical insights to get

started. Sounds good. Yes. Wonderful. So

I want to start off with just defining

an AI agent just so everybody is on the

same page here. What is an AI agent?

Wonderful question. An AI agent is an

autonomous system that can perceive its

environment, make decisions and take and

take actions to achieve specific goals.

So we usually refer to AI agents as um

an agent for a specific function like

travel agent, right? Would be um an

agent that would be like booking your

travels and trying to figure out where

your vacation spots are going to be. You

can have like personal assistant agent

that's doing stuff like managing your

calendars, research agent is conducting

research on specific topics. So that's

how we normally refer to an agent. Um,

an agent is autonomous, so it operates

without constant human intervention. It

has tool use. It has access to a variety

of different tools through different

APIs and something called MCP, which

we'll also discuss a little bit later on

as well. Um, it has memory, so it's able

to remember context and pass

interactions with it as well. Don't want

your agents like forgetting everything.

And it has the ability to plan. So it's

able to break down complex task into

different steps. So those are the

difference between an agent and a

chatbot. A chatbot is something that you

can talk with and it's like fun and

chill and you can just like chat with it

about stuff, but it's not able to

actually do anything. An AI agent, on

the other hand, is able to take actions

in the real world by using tools,

maintaining context, and working towards

goals autonomously. So it has the action

part in addition to just being able to

chat with you.

Okay. So, I'm going to show you um this

like the like here is an N10 visual

workflow plus an AI. Um I'm going to

show you like a few demos that yeah, I

think I have like a few demos that I can

show over here. Uh so, NA10 as a visual

workflow is the way that you can do it

with with no code. Um and you're able to

it's a visual builder. You can just put

things together really easily and just

like literally drag and drop. Like it is

literally drag and drop. um completely

no code if you want it to be and I'll

actually show you in just a little bit

what you can build using just NA10 and

you're able to switch out different uh a

like different types of um my god I

can't talk today different types of AI

models as well so you can have it as GP4

powered you can have it as powered by

other stuff as well you can stylize the

output so this is a screenshot of what

it looks like so I'm not going to go

into a full like detailed tour of NA10

itself right now because that's going to

take a while Uh, but you can, yeah, it's

just like a drag and drop interface that

you can check out. Uh, we'll go more

into the tools and stuff like that, but

I just wanted to show you guys what it

is that you can do when you can use no

code to build something. Um, so this is

an example of an AI agent that you can

build really, really easily as well. So,

this is an example of using Streamlit

plus OpenAI agents SDK. So, this is a

code approach.

Excuse me. you can have an agent up and

running literally within 30 lines of

code. So this is like a task generator

agent. By the way, these are all like

agents that are built um in our agents

boot camp. So yeah, in case you were

wondering like what these are for. So in

our agents boot camp, like we start off

with very simple agents like creating an

AI agent for stock management. So this

is like the simplest approach for

building AI agents. And this one is the

code approach uh the what you would be

building within the boot camp. And it it

takes you like within 30 lines of code

to be able to do. And yeah, this is a

code implementation. So this is a task

generator agent. Um you can enter your

goal like I want to open a restaurant

and it'll generate certain tasks like

here are your tasks blah blah blah and

what it does is takes any goal as input

and your AI agent is able to analyze and

plan it return a structured task

breakdown and then just in real time

streaming response really simple this

interface is called streamllet and

openai's agent SDK is the thing that

powers the the task generated agents

like the framework that powers it. Uh

this is done using Python code and

literally 30 lines of code total. So um

however I want to show you guys like

some of the more fancy ones that you can

build as well cuz I think that's like

you know I'm going to show you some some

of the cool ones um that you can build

as well. So this is also built in the AI

agent boot camp. So through the boot

camp, you literally start off with

something super simple like this. Uh and

then you're able to like learn a lot of

things and then at the end of it, you'll

be able to build something a lot more

advanced as well. And this is still

built using NA10. So this is the visual

uh workflow using using NA10. So it's

still completely no code, but you can

see like how much variation there

actually is from a very very simple AI

agent all the way to something much more

complex. So I'm actually going to run

this demo first and then I'll explain a

little bit about the architecture of of

how it works.

Um, okay. So this is the deployed

version. So this is what it looks like

when you're actually building it. Um,

it's it's an AI agent that is able to

have access to a message and is able to

have access to a lot of different

things. And this when you actually

deploy it, uh, this is what it would

look like.

I just wait I just want to make sure

like can you guys actually see my

screen properly.

Can I go to Wait. Yeah. My god. I'm like

super paranoid about OBS right now.

You guys can see my screen

properly right?

Yeah. Okay. All right. So, just a quick

little demo here. So, tell me about your

knowledge about the current job market

in Canada. Um, I would like to get

up-to-date information on this topic.

Send me the research as a report with a

link. So, we'll let that run for a bit.

Um, this is a AI agent that's called

Atlas. And then while it's doing its

thing, I'm going to explain the

architecture um surrounding it. So,

okay, this is what it looks like. Uh,

your AI agent, you have a chat message

that's being received and then your AI

agent is will take that message and then

has the ability of um accessing a lot of

these different tools. So the model

itself is the open AI AI model. It's

sorry it's an open AI chat model has the

model itself and here is the

architecture surrounding it. So you have

your main agent um it has a rag agent um

which is able to access a database. So

this is just PG vector store. Uh by the

way you don't need to worry so much

about like the the terminology here. All

all it really is is just like a database

that's able to search things and then it

can search its knowledge base, retrieve

documentation and vector similarity

search. So when we're asking a question

oh wow somehow I already did that but

yeah when we're asking a question um

surrounding getting information from the

database this is how it's able to do

that. It also has a web search agent uh

where it's able able to use SER API

which a tool that allows it to real time

search the web for information like

current events and news and external

data retrieval and it also has a

communication agent that allows it

access through the Google Docs API. So

it can create documents um professional

formatting and report generation. Uh

over here we also have like error

handling which is a graceful failure

recovery. So if something does happen to

the AI agent, it doesn't just like

completely crash out and die, right?

That's not good user experience. So

there uh I guess I don't know if I can

demo the graceful error handling, but

you know, just got to trust me with that

one. It is there. So it won't just

something bad happens over here. Um

yeah, and then for the rag upload

pipeline, this is where you're able to

have a lot of the documents are being

ingested and all the information that's

being stored. So this is the

architecture of how that information um

like how it is that the chat is being

received and the access that the agent

has to all the different types of tools

that are over here. So you have the main

orchestrator which is this one and then

you have the rag agent um which is an

which is an agent that does the rag

stuff. You have the web search agent

that is the agent that does all the web

search stuff and it has a communication

agent which is able to communicate with

different um which has access to

different

cannot talk different APIs including the

Google Docs API for you to do stuff. So

um this is also an example of a multi-

aent um system. So you can see that in

like the simple agents you just have

singular agent but this is a much more

fancy agent agentic system. So that's

why you actually have a main agent and

you also have multiple agents as it has

access to. So we're splitting things up

together um and then putting it back

together again. So yeah, this is what is

called a hierarchical architecture. Um

because the main agent is able to go and

interact with these different agents and

this is a hierarchy that's being

established.

So I know that was a lot of information.

So actually let's actually see what it

came up with. The question that we asked

it was, "Tell me about your knowledge

about the current job market in Canada.

I would like to get up-to-date

information on this topic. Send me to

research as a report with a link." And

it was able to get the current job

market um in 2023.

Okay. As of October 2023, that is not

correct. The job market in Canada is

showing a general positive um

trend with fluctuations. Key highlights

include employment growth, labor market,

and then you can actually view the

document. So, it instead of actually

retrieving That's interesting. Wow, you

guys saw saw an error happen cuz the

what it should have done is actually

went and web search, but instead what it

has done is actually went into the

database and then

got that information. So, yeah, there

you go. You saw an error that did

happen. So it went instead of going to

the web search agent in order to make

sure it has the right date and then get

that information compile it. It actually

went and got the rag agent um and then

took the information from this vector

store instead to create the

communication and the communication

agent was going to get the Google docs

instead. So that was interesting. So I'm

actually going to do here and this is

yes it is completely live. So I don't

know if it's going to be able to do this

um reach do this for the current

current date.

See if it actually does it.

Waiting. Waiting.

Okay. While it's doing that,

let me see if there's any

questions that people have. Bad

prompting, Tina. That's true. It is bad

prompting. That is true.

This is quite simple.

So this this is not simple. Just by the

way, you can see that this is like

multiple agents that are working

together in order to do something. The

experience that you have that it feels

like it's simple. Okay, you know what?

I'm just going to be like do this for

2025. It's supposed to do web search

then. It is bad prompting from my side.

Okay, whatever. Anyways, that was very

bad prompting. I don't know what's

happening over here, but it is a this is

the agent that is going through rag and

then it's supposed to do communication.

I probably need to prompt it a little

bit better in order to actually get the

results that we want to do over here.

Anyways, I'll go see if I can go back to

this and fix it later. But um I do want

to show you the multi-Aian system. So

that one was a noode implementation

using NA10. Um

while the next one over here is the one

that's also built using during the AI

agents boot camp. And you can see

there's a lot more complex like a lot

more stuff that's happening over here.

Let me actually see that anybody has any

questions for this agent first.

Uh, Ernesto, thank you so much. Thank

you so much for supporting the channel

and supporting me. Really appreciate it.

I took the boot camp. It is awesome.

Thank you so much.

This is Google ADK. No, it's uh that was

NA10 and this one is the one I'm going

to be showing you now is going to be um

OpenAI's agents SDK, not Google ADK.

They're not that different though.

Um,

okay. I'm just looking at people's

comments.

Prove that this Tina is not a bot.

That's true. Yeah, if I were a bot, I

would not have made that mistake. Why

not ask Chat GBT? That's a great

question. If you were to ask Chad GPD,

you could ask Chat GPT about this

information. But say that you're the the

database that you current that you have

in there, right? What if it's

information that you don't want to give

Chad GPT, right? like what if it's like

information that you that chatbt doesn't

have access to and what if you actually

want to take this agent um and you want

to create like a web interface right see

the interface that we see over here this

is like a very simple interface that we

have but what if you wanted to create

this into like a full-on application

interface that has really cool UI that's

able you're able to get people to um

interact with it you can build out so

many other parts of of your your

application then you can't just use

Chachi for it because you would be

limited to Chachip PD's interface. Um,

and there's also like things that you

would wanted to do that maybe Chachd

isn't able to able able to do directly

like if you wanted to use a different um

yeah like say for example if you want to

use a different model and you don't want

to use GPT for you can want to use like

Gemini or cloud you can't do that with

CHP. I mean you can change like to

different play like you can use

different models and just use other

people's chat bots but you're never

going to have that control to be able to

actually build the interface and the

product that you want to build unless

you go and build an agentic system

yourself.

So yes

that is why

um is the rag document repository

persistent where you need to upload each

time it is persistent. So you are

putting all that information into the

vector database and then you're able to

access that information and it is

persistent.

Great refresher. Also talk to boot camp.

Oh hey Dea, nice to see you again.

Um

where's the GitHub foundation link?

Not sure what you're talking about.

Which GitHub foundation

link?

Let's see. Your mic is low. This is I'm

already talking as close as I possibly

do have a very quiet voice in person.

That has always been a thing.

Let's see.

Well, I always struggle getting

information for one sub agent to another

one. Okay,

that is true. So, passing one agent

along to another agent is is actually um

it's like a whole thing like that's why

prompting is really really important as

you can see um in order to actually pass

that along. So, every mo in the multi-

aent system each agent has its own

prompt. So, it actually needs to know

when it's supposed to pass it

information on to the other agent

afterwards. Okay, I'm going to move on

to the next um move on a bit because we

do have a lot that I do want to cover

today.

So, this is the multi- aent system that

is built um using the agents uh OpenAI

agents SDK. So, this is a code

implementation

and this is the architecture here. It's

like a lot. Yes, it is

that that is a lot. So, what I'm going

to do here um this is what it looks

like. the code implementation uh and I'm

going to ask this question like should I

buy Apple stock provide a complete

analysis with price target and while it

is doing that

I will explain briefly the architecture

surrounding it so you can see that this

um this interface is completely custom

as well you can like change these around

if you want

let's hope that this live demo works

okay so this is a lot that is happening

over here. Whoa, what's happening? Okay,

so it's it's actually quite similar. You

still have a user query that's coming in

to your portfolio manager agent and then

this agent is for investment research,

right? And then after the user queries,

your portfolio manager agent has access

to these three other sub aents. The

first one is a fundamental analyst. So,

it's able to do things like financial

analysis, business quality, um, v like

valuation metrics, like fundamental

analysis stuff, and it has access to

tools like the code interpreter, the web

search, whatever. Um, and then you have

the macro analyst. So, it's looking at

the economic environment like things

like interest rate and Fed policies and

things like that, market conditions. And

it also has access to web search and it

has access to the Fred API tool, which

is for economic data specifically. And

it has access to code interpreter as

well. Then we have the quantitative

analyst. So this one is looking at

technical analysis. So includes things

like statistical modeling, price

targets, and it has access to code

interpreter with Wi-Fi as a Yahoo

Finance API. Um and it has access to

these tools as well. So these um what's

interesting here is like when we're

asking this question, we're basically

asking it to analyzing from three

different perspectives. The fundamental

approach, the macro approach, and the

quantitative approach. Then we compile

everything together and synthesize them

all together into a final response and

recommendation. So um all that

information is going to be coming

together and synthesized and you would

be able to have a chat response in the

UI which hopefully we'll be seeing soon

and then also an investment memo uh

which is a markdown file that you can

download and have information around it

as well. So um not super important but

if you are curious um this is what it

looks like the project structure from

the code itself. Um and then the

parallel execution as well. So you're

able this is how you're able to run all

three agents simultaneously. So you're

not having to like wait for one to run

after another. You're all three agents

sub agents are running simultaneously

and then everything is being compiled

together. So four specializations

parallel processing the database is a

Postgress SQL database and a vector

database and the deployment is using

Heroku

and it's still going okay maybe we'll

come back to this and okay wonderful

that's perfect okay comprehensive

analysis of Apple as of October 18th

2025 is it October 18 it is indeed

October 18th good job okay so executive

summary Apple's robust financial health

innovative product development strategic

market positioning blah blah blah all

these things so here's the fund

fundamental analysis information uh

valuation context and here's the macro

analysis so interest rate scenario

analysis here's a quantitative analysis

information here portfolio manager

synthesis so you combine all of the

these things together the recommendation

is to buy with a 12-month target of 280

not financial advice okay not financial

advice don't go buy Apple stock just

because our wonderful little agent here

told us to do so anyways um yeah so

would you like to proceed uh you can

also look at the execution details as

well. So here's the information of

exactly what it is that's being called

and how it is that everything is

interacting with each other. So this is

really interesting to look at if you do

want to do this. So do I want to what is

it? Uh create a memo. Yes. So create

memo as well. So we'll let it do that.

Um

it's doing that. Let me see if anybody

has any questions.

Can you share that document? Oh yeah. Uh

actually, oops. Sorry, I forgot. Should

have said this earlier. Please sign up

for um if you do want resources from

this live stream, like here is the I

should probably like Yes. So yeah,

please do sign up here. Um and we'll

just email you

besides sign up there. Uh yes,

quit telling her she's muted.

Audio is fine for me at this point.

Okay,

you're Thank you, Gregory. I appreciate

it.

Um, is this live? Oh my god, you guys.

How many times are you guys going to ask

me this question? You're you're making

me doubt my own existence. Is this live?

Am I alive? Am I a person? Yes, I'm a

person. Yes, this is live. Yes, I also

happen to be alive. I'm as alive as you

are. How about that?

I see no GitHub on this display. There

is no GitHub on this display. Why would

there be GitHub on this display?

Okay, you guys are trolling me. I'm just

going to stop trolling. Okay, I'm I'm

fragile.

Okay, this investment memo for Apple has

been successfully created. You can

access it here. Okay, I guess I cannot

access it here.

Okay. A

This is not going really well today.

Not doing our best today. Anyways,

generally speaking, it seems like live

demos, you know. Let's just leave it as

that. Live demos. Well, it did pull

things correctly.

Anyways.

Oh, app. Okay. Okay. Got it. Got it. So,

I need to actually it's it's it's in my

like actual folder. Okay. Got it. Got

it. Okay. Anyways, yes. So, it did work.

Um I'm not going to pull that up right

now because I have to flip through a

bunch of different folders right now to

actually go in and and actually get the

get the markdown file. But basically, it

just put all this information into a

memo instead. And then you're able to

have access to it and then do whatever

you want with it, send it to your boss

or something. So, yes.

All right. So, those are my two live

demos. Um, yeah, you can see these are a

lot more complex than just building a

simple agent, but I do want to like show

you guys what you can do. And what's

really cool is like throughout the boot

camp, like you literally go from

building something as simple as a task

gener agent with like 30 lines of code

to building like a full out architecture

like this that you're able to have

multi- aent systems and everything being

put together. Um, and you can do this

both using no code and code. So, gives

you hope. You got this. Okay. So now I'm

going to actually do the step-by-step

framework for how to approach building

an AI agent. So how do we actually do

this? The first one is that you want to

define your use case, right? So the

first question that you want to ask

yourself is like what specific problem

are you trying to solve? What task will

the agent perform? Who will use this

agent? And what's the expected input and

output? This is actually really

important because if you're just like

suddenly randomly building an AI agent

for no reason, you're probably not going

to build a very good AI agent. So that's

why you need to actually start um with

defining like what are you actually

doing it for. For example, we had the

research one um for investments, right?

So that was an example. You're defining

the use case because you want to have a

multifaceted approach for looking at um

an investment decision and you want to

have a lot of control over how the sub

aents are going to be looking like what

kind of information they're going to be

compiling and how they're putting all

the information together. So that would

be the defining of the use case. The

step uh the next step you want to do is

choosing your approach. So the general

categories is that you can go with like

a no code low code approach or you can

have a code based approach. The no code

low code approach is best for quick

prototypes, simple workflows and

non-technical users. Uh the pro is that

it's fast, it's visual, it's easy to

modify. And the cons is that it's less

flexible and potential scaling limits.

It also tends to be a little bit more

expensive. While on the code side, it's

best for more complex logic, customer

integrations, and production systems. Um

the p the pros is that it has full

control. It's scalable. It's

customizable. And the con is that it

does require technical skills. So you do

need to know how to code. Um and it has

a longer development time as time as

well. So you can choose based upon your

skill set and your needs which one it is

that you want to do a no code approach

or a codebased approach.

Um for every single agent you also need

to think about these are the essential

components that every agent needs and we

call this like the building blocks of AI

agents. So the first one is that you

need to have the large language model

itself. So the brain of the engine and

you need to choose uh GPT4 maybe claude

Gemini whatever based upon your needs

like maybe you want things that are

faster you care a lot about pricing then

you want to maybe go with Gemini as

opposed to claude if you think if you

care a lot about like quantitative stuff

or codebased stuff analysis then you

probably want to go with claude GPD4 is

a very good um all-rounder kind of model

to start off with you do need to choose

what you want your large language model

to be then your instructions so you need

to have clear prompts that define find

the agent's role, capabilities, and

constraints so it knows what it's

supposed to do. You have to give it

tools like different functions that the

agent can call, APIs, databases,

calculators, whatever um that it needs

in order to do its job. The memory is

for storing the conversation and

long-term information. So, you need to

like store all that information uh so

it's able to have access to it. So, we

saw the databases that we showed

earlier. Those are examples of um

storage. Then there's guardrails. So,

you have safety measures and output

validation. So, I think these two steps

are the ones that people generally tend

to not pay attention to. But, it's

actually really really important because

you need to make sure that your output

is is valid and you also want to make

sure that it's not doing things that

it's not supposed to be doing. Like your

customer service agent, you don't want

it to be like telling people life

advice, right? So, you need to actually

test all of these things um in order to

do it properly. Then, there's finally

orchestration. So, this is when

everything comes together and you're

actually doing a deployment that is

there. So when you're doing a

deployment, you also need to think about

how it is that everything is being

deployed together. Um and then also like

how to scale it and and all these other

considerations surrounding the actual

usage from your users too. So uh the pro

tip here is that start simple just have

like large language models instructions

and one to two tools like just try to

think about it as simple as you can. you

can add complexity as needed, but you do

want to keep these in mind that you do

want these six different components um

while you're building your AI agent.

Okay, so before I move on to talking

about some of the code and no code

tools, let's see if anybody has any

questions. Uh by the way, if you do feel

like there's a lot that's that's

happening over here, like don't worry

about it. Um just follow these steps

through and just realize that there is a

lot that you can build. Like we have our

boot camp is 28 days long, right? So

there's like it does take significant

amount of time to actually work up to

building something that is much more

advanced over here. So don't worry if it

feels like there's just like a lot

that's going on. Um I'm trying to like

simplify it down to when you're building

your first AI agent to all the way

building a full-on like AI agent. You're

still following the same steps that are

here, but you're just increasing the

amount of complexity. You're increasing

the number of agents if you're doing

multi- aent systems. And there's a lot

more like tools and memory that you can

add to it. But it also it does all

follow the fundamentally like it's the

same approach and the same components

that are being put together as well.

Okay. See if anybody has any questions.

Demo god being I know right like even

prior to this I was actually talking to

I'm just like oh my god I hope the demo

works. And I mean it kind of worked. It

more or less worked.

I don't think I've actually done a live

demo which everything worked perfectly

before. So, you know what? That proves

that I'm real, huh?

All right, I shall drink some water.

How are the guard rail set? A lot of

ways of doing this. Um, OpenAI, for

example, has like a full system where

you can set these guard rails in place.

You can also put in evaluations, the

testing, things like that. That's one

approach of doing it. U there's a lot of

third parties as well that you can use

to set these guardrails. Part of the

guardrail is also putting into the

prompt itself. So testing things and

then putting the guardrails like never

do certain things. You can literally

explicitly tell it like never do certain

things like never swear your customers

for example.

How do you guys feel about my picture of

water? Stay hydrated, my friends. Thank

you Gregory.

Appreciate it.

Uh website says boot camp is sold out.

Okay, so we we will be opening

enrollment on the

21st. So if you haven't already, um if

you are interested, you should

definitely check it out. I'll also put

the link over here if you want to check

it out. Uh, so when we do open it up on

the 21st, we're opening up to the

here. Sign up for weight list here.

Yeah, we we open it up to the weight

list first. And yeah, um, honestly like

thank you all so much for just Yeah,

like last time our it sold out within

less than 30 minutes just on the weight

list alone and then prior to that it was

less than an hour that it sold out uh

only to weight list alone as well. So,

if you are interested in the boot camp

at all, uh I recommend that you do sign

up for the wait list because we open it

to the wait list before we open it to

the public. But we actually haven't

managed to open it to the public for the

few times that we've run the boot camp

because it's sold out completely on the

weight list really quickly and thank you

so much for for that. Like we are very

very grateful that is the case. Um but

yes, we'll be opening up on the 21st,

100 spots only. We don't expand the

number of spots because we want to make

sure that everybody gets the attention

that they they need in order they need

and they deserve in order to actually

build everything. And it's 28 days. Uh

there is like a lot that we cover. So

you saw all the way from the very

beginning of the boot camp to the end

you're covering a lot of different

things and building things up in

complexity. So uh we we do keep the

class sizes really small like not really

small but 100 people because that's the

number of people we feel confident that

we can make sure everybody has a really

good experience. So that is why anyways

check it out if you're interested. We'll

be enrolling in a few days.

Uh how much please? It's $997

for the 28 days. Yes. US dollars.

All right. Uh, looks like not having any

questions.

No questions.

Do we get to set up a passive income

stream with this? Oh, I do have another

workshop that I'm doing, but that sign

up has already passed. I don't know if

we'll do it again, but uh, which I do

cover like freelancing in the boot camp.

do cover it like in some of the Q&As's,

but I do have like the full workshop

where I'm doing freelancing on day one

and then content creation using AI,

which is part of generating income

stream by building agents and stuff like

that on day two. So, I think that's

happening end of this month. I forgot

which day. I think it was the 29th

that's happening. But yeah, that

enrollment has passed though. So, maybe

we'll do another one in the future.

How to get the code files? How use MD

files to configure agent context? Well,

the code files is going to be like where

you're deploying it as like that

information like the the code files are

going to be stored there, right? Because

this is the deployed one onto the

website. How use MD file to configure

agent context? Um, I guess you can

you're using a markdown file in order to

feed it stuff like prompts. That's

probably how you would do that. That's

how we do it.

What is the tech stack for this AI

agent? So, it is on the slides uh the AI

agent tech stack. You can actually check

it out. So, this is the full

architecture of it and you can actually

see like what the tech stack is, all the

different tools that are there and

exactly what each agent is doing. So,

yeah, sign up for the workshop as well.

I'm just going to like put the link over

here again. Sign up for the workshop. Uh

you can get we'll send you we'll email

these slides afterwards. All right, let

us move on to popular agent building

tools. So there are we've done like you

know we keep tabs on the tools that are

coming out for AI agents both like the

no code tools and the codebased stuff

with and there are 15 that we gathered

for that we're keeping an eye out on

there's probably more um but the ones

that seem decent there's like 15 and

then the codebased ones right now I'll

say like 13 or so um what I do want to

emphasize here it's also what I say like

over and over again in my videos in the

boot camp everywhere it's don't focus so

much on the actual tools themselves.

Okay? Like why is it that I went like

blah blah blah like I so I talk about

like all these things, right? Like why

am I talking about these frameworks and

stuff like that not just for funsies?

It's because when you understand the

architecture and the structure and the

process for how to build an AI agent

like these essential components, it

doesn't actually matter as much what the

tool it is that you're using is because

you can implement that using a lot of

different types of tools and the tools

themselves are going to keep evolving,

right? There's going to be new tools

that are coming out. the current tools

are going to become outdated where

they're going to keep becoming better

and better. Um, and we don't want to be

like married to a specific type of tool

and then if something changes like

everything falls apart. You don't want

to do that. Like you want to have the

ability to understand how to build an AI

agent from a fundamentals perspective.

So the implementation part is actually

going to be the easy part of building

this this agent and you're able to

translate that into a lot of different

things. That's why you can like build

agents like this investments one, right?

And this is using OpenAI's agents SDK.

But you can also build something like

this which is going to be using NA10,

right? But you can s see the

architecture. They're both agents and

they're both using the same agentic

pattern as well, which is a hierarchal

pattern. Not going to go into much

detail about that, but it's like but

we're able to implement it by using

different tools. So that's why like I

emphasize so much like don't be focusing

so much as on the tools themselves.

Understand the structure of what you're

trying to build. understand the

architecture and the elements that

compose an agent among like that's the

most important thing to actually know

then the actual implementation part is

going to be quite easy and that can

change and vary and as the tools get

better and better themselves.

Okay, so uh with that being said let's

talk about some of the tools. So here

are the no code tools that we keep our

eye out keep tabs on and we're aware of.

Um these are like visual builders for

different types of skill levels. So when

uh as a reminder when to use no code

tools. So you want to do it for rapid

prototyping, simple workflows, business

users without coding experience and

quick MVPs as well. So here's the

popular and accessessible ones. Like

this is the one that we use in the boot

camp. Um I think it's the most I I think

it is the most popular one these days

and it's also very very flexible. So

that's why we like using NA10. Uh you

can also do stuff like self-hosting and

things like that which you can decrease

the cost significantly if you're using

NA10 compared to some of the other ones

but there's also stuff like make there's

zapure agents there's gum loop um this

is specifically more for businesses

there's relevance AI mind studio uh we

also have listed ones that are more

developer friendly and enterprise

platforms enterprise sol sorry excuse me

enterprise uh platforms and solutions as

well so these are going to have more um

built-in safeguards um like guard rails

that are there privacy things because

that's what enterprises and companies

would care about more. Um yeah, so

there's open a Asian kid, there's

Google's Vert.ex AI, Amazon, um

Microsoft Co-Pilot Studio as well, etc.

So, I'm not going to go into all of

these in a lot of detail, but yeah,

after you have access to the slides, um

if you are curious about the tools, I

just wanted to make sure that we have

everything here so you can go and dig

into it and see what what's available

and what makes the most sense for you.

Then we have the codebased options. So

these are SDKs and frameworks for

developers. So when do we use code? Uh

when we want to have things that have

more complex logic, some more custom

integrations, production deployment,

full control over behavior and more

advanced use cases. Of course, the

prerequisite here is that you do need to

know how to code in order to use

codebased solutions. Um here are some of

them. The official provider SDKs like we

use the OpenAI agents SDK for um our

agents boot camp. The reason why we do

this is very popular and also I really

like how open framework they do cover a

lot more than most SDKs do like all the

elements that we see over here it covers

all of these including guardrails and

orchestration. So um it's very fleshed

out. If you use the OpenAI's SDK

ecosystem, you're able to very easily um

do stuff like evaluations, testing, and

have like specific guard rails in place.

You also are able to have a vector store

that has is available too. And doing a

deployment is super simple as well. So

that's why we choose to use the agents

SDK um from OpenAI for the boot camp.

But there's also other SDKs too like the

Amazon Bedrock one. There's Google's

ADK. Somebody mentioned that previously.

Microsoft also has ecosystem like

autogen. Uh this is an open source one.

There's semantic kernel open source SDK

as well. And then here are some popular

open source frameworks that you can

check out too like lang chain, langraph,

curi, autog etc etc. So again not going

to go into too much detail about all of

these here but if you are interested I

just wanted to make sure that they are

all on this slide so that you can check

it out

when you want to and see which tool it

is that you want to be using.

Okay. So here is the no code versus code

decision matrix. So if you're like not

sure what it is that you should do,

you're like should I go no code? Should

I um go with a code approach instead? So

uh we I put this here because in the

boot camp that's also like often the the

f one of the first questions that people

ask us is like should I go with a no

code approach or should I go with the

code approach? Because we do everything

that we teach in the boot camp you can

do it with a no code approach and a code

approach and we show you how to do both,

right? Um and this is kind of like a

summary of how it is that you can choose

which approach you should be doing it

for. So the no code approach um the

learning curve is going to be a lot

lower. So it's going to have a visual

interface and it's also going to be a

lot faster for you to develop. Um of

course on the side of the code it does

higher learning curve because you need

to know how to code. It's going to be

slower so days and weeks as opposed to

hours um to days. customization though

the downside of a no code tool is that

it is more limited to platform features

themselves and the scalability is also

limited to platform limits while for

codebased one this is where it really

shines you're able to have unlimited

flexibility and it's going to be highly

scalable as well on the cost side uh

platform subscription fees are here and

then for the code one so this is a

little bit deceptive so I do want to

clarify this if you just like build

something by yourself using no code it's

probably going to cost you more um

because of the platform subscription

fees while you do it with purely code is

going to cost you less. However, if

you're going to scale it up and actually

build everything up and develop it and

build it into something much better,

then it is going to cost you more on the

code approach because then you're going

to have to start paying for things like

infrastructure um and a lot of like

these other components there as well.

But if you do want to scale, it is worth

this amount of money and it's actually

just like not very possible to do that

if you're going from a pure no code

approach. if you want to scale in like

thousands of people, right? And then

having like full-on infrastructure and

stuff like that. So, in terms of

maintenance, um, no code is super

simple. The platform handles all the

updates. You don't have to deal with

that. And in terms of integrations as

well, there's pre-built connectors, so

you don't need to do that yourself. Um,

with the code one, maintenance is a

little bit more effort because you have

to manage everything yourself, but the

integrations are actually pretty easy as

well. So, putting custom integrations

in, it's really not that hard,

especially with something called MCP,

uh, is really not that hard to do these

days either. So as a summary, no code is

best for MVP, simple workflows, and

business users, while for code, it's

best for production apps, complex logic,

and custom needs. The recommendation, if

you literally have no idea, it's like,

oh my god, I don't know what to do.

Should I go to code with no code? If you

have to ask yourself this question,

start with the no code first, right?

Because if you're like completely fully

comfortable with code, you would just

jump to the code approach. So if you're

like oh I I'm like kind of know how to

code kind of not not sure not really

sure about this then our recommendation

would just is always like just start

with the no code approach to validate

your idea quickly and then you can move

to code when you need more control and

hit platform limitations. So don't

overthink this one. All right let's see

if there's any questions. Uh put into

the chat uh I'm assuming you're at this

live stream because you want to build

your first agent. Do you think you will

go with a code where no code approach?

Can we get the slides? Yes, you can get

the slides. Whoops.

Uh, yeah, you can get the slides here.

I shall link it again. You can Okay, I'm

going to pin it. You can get slides

here. So,

sign up here for slides.

Oops, I pressed the thing. I should not

have done that. Okay, let me do this

of her boot camp weight list

is going to be at only octopus.com. And

so you guys can have all the links just

there.

Okay,

great.

I will pin this message. Okay.

Okay.

Code, code, code. No code. No code for

me. Code. Okay. Good mixture. Good

mixture. No code. No code. Code. Code.

Code. No code. Code. Of course.

as a developer there. No code. No code.

Okay. A very good mix. Yeah. And that's

totally cool, you know. Totally cool.

Code because no code is boring. That is

not a I don't think no code is boring.

Do you need to know how to code to

participate in agent boot camp? No, you

do not need to know how to code to

participate in the Asian boot camp. Uh

our boot camps, yeah, like we have a

code approach and a no code approach for

every single part. So, you can do both

if you want. Some people like to do

that. Uh but you can also choose to do

the no code one or the code one and

you're able to participate fully.

All right, so let's talk about MCP uh

very briefly. I actually just did a

video on this, so you guys you can check

it out if you want more detail about

this, but I do think it's quite

important um because MCP is a universal

standard that lets agents connect to any

tour data source seamlessly. This

becomes like much more um important when

you're doing a code implementation.

Although it you can also build MCPC ser

MCP servers and stuff like that using no

code as well. Um so why does it matter?

So the MCP really is something that made

agents much more powerful because uh

prior to MCP it was actually quite

difficult to provide different tools and

memory to your agent. Say like you want

to give it like a weather tool then you

need to connect it to the API for the

weather tool it's like has a very

specific set sequence that you need to

do in order to connect it and maybe you

wanted to have like a calendar access

then you have to go a completely

different set of sequences and connect

to that API and you wanted to have like

a database then you got to go through

that process again so it's like very it

was very difficult because you needed to

um have like this custom code to

interact with all these different

external tools but with MCP what it

basically did is that it became a

universal standard so when you're

connecting um a calendar tool with a

weather tool and another database, it

all becomes very simple because it's the

same way that you're able to access it.

It's very similar to like a USB stick.

Like prior to having a USBC or USBA, um

it was very difficult to connect

hardware with each other because

everything had its own like special

little plugs, right? But after the

universal standard came out, then using

like USBC, USBA for example, then it

would became a lot easier because you

didn't need to have like a bunch of

different ports. you can just use the

same ports to connect things together

and because of this there was a lot more

innovation that happened. Um yeah, so

that's why I did want to cover MCP

because I think it is like a very core

part of um why it is that agents are

becoming more and more powerful as well.

So, I'm not going to go into too much

detail here uh because there's like a

lot like we have literally like an

entire week, an entire assignment in the

boot camp where we covered how to like

build MCPS and stuff like that. But, um

I think it is important for you guys to

be aware of this. What's interesting is

like people actually build MCPs. Um some

people do it for free, but people

actually build them um and actually like

sell them as well because they are

really really powerful uh if you build

really good ones.

All right. So, another step that you

need to consider when you're building

your AI agent is evaluation. So, this is

the part where you're actually testing

and improving over time. Like for

example, um you know, I saw like an

error that happened over here, right?

Like like oh no, like I I ran into an

error previously. So, this is part of

the testing evaluation process that I

need to do. Um because I need to go back

and like fix the prompt in order to make

it better or maybe fix other parts of

the agent in order to remove as much of

these errors as possible when I'm

testing it. uh and instead of just like

you know doing random prompts and hoping

to catch these errors using evaluations

is a way to systematically do this

approach. Um so you're able to have like

a bunch of different use cases run

evaluations on it and catch a lot of

errors that you might be getting. So why

do we need to do this at all? So what

doesn't get measured doesn't get

improved. This is like a quote I forgot

who said this quote but seemed important

whoever that said this quote and I think

it's a very good quote. So that's why

you need to measure it so you can

actually improve it over time. So how do

we actually do this? The first step is

to create test cases. So you need to

build a spreadsheet with a lot of

different inputs to test different

scenarios. Then you want to run

evaluation. So execute all the test

cases and capture the output of the

agent automatically. Then you want to

analyze and adjust your prompt or retest

to improve the performance over time. So

this is the sequence of things to do in

order to evaluate it. So common metrics

for evaluation include like helpfulness,

like how useful is this response, does

it match, expected to structure,

correctness, is the information

accurate, and custom your specific

business metrics. So these are all like

um things that you need to test for. But

do pay attention to evaluations and

testing. I think it's something that

people don't tend to talk about as much

because it's not as like sexy and cool

as the building of the agent and demoing

of the agent itself. But if you don't

test things early, then you're going to

have problems.

If anybody has any question

any way to run agents 247 remotely for

minimal cost.

If you want it like to literally be

running at all times. Uh you can

probably if you want minimal cost go

with like open source models and

probably like just go with like super

cheap models is how I do it. But I think

the question I would ask you is like

does it really need to run 24/7? Right?

Does it really need to be running 24/7?

I don't know if that's you. In almost

all cases, it does not need to be

running 24/7. There is other things that

you can do like if you're just having a

monitoring system. Your agent itself,

for example, doesn't need to be running

24/7. You can actually have other pieces

of code that is able to evaluate if

something needs to be monitored. And

your agent will only get awoken up and

triggered once something comes in that

needs its attention. The agent Yeah.

Like I would really rethink that

statement like does it really need to be

running 247?

is there an M MCP in there

and where

how different is this from using cloud

or Gemini or GBD5 to build agent. So

those are just different models, right?

Like you it doesn't actually matter that

much like which model it is. You can

switch it out pretty easily. It would

matter in terms of the results that you

get, but the overall infrastructure like

the architecture of the AI agent, you

can switch out the models if you want.

Yeah, it doesn't actually change the the

way that the model sorry, it doesn't

actually change the way that the agent

is built itself.

Okay, cool. So, the final part of it um

is that you need to deploy and monitor

your agent. You build a thing, you're so

happy with it. Wonderful. Amazing. You

want other people to use it, so that's

why you need to deploy it. So, here is a

deployment checklist um that we have on

the side here. So, test early, setup,

monitoring, enable guardrails, deploy

gradually, and continue continuous

improvement. Uh, I know we're a little

bit out of time, so I'm going to like

not go into too much more detail about

this, but this is the deployment

checklist that you definitely need to go

through to make sure that things are

working correctly. So, um, it can be a

simple like deployment can be really

simple if you're using something like

NA10. All you have to do is click it

from inactive to active. Yay, you're

done. Amazing. But it can also be a lot

more complex if you're doing it through

code and you care a lot more about how

it's being deployed and you know how

you're being monitored and things like

that. So it can range from just

literally clicking a button to something

that's much more complex and you want to

have some sort of monitoring system. So

maybe if your thing like breaks down,

you want to know about it.

Okay. So here are some common pitfalls

um that we've seen people do. Um we made

a lot of these mistakes ourselves as

well. So stuff like vague instructions,

your agent doesn't know what to do if

you don't give a very good prompts. Um,

you got to be like very specific about

it. Too many tools. This is a very

common one. You're like, "Oh, I want to

endow my agent with all the tools." This

actually not great because your agent

gets confused because that's too many

tools and it's like, "I don't know what

to do at this point." So you want to

start with two to three essential tools

and only add on more when needed. No

memory management. So if you don't have

any memory management, your context just

gets larger and larger as you continue

having a conversation and getting your

agent to do stuff. So this is going to

end up having problems like token limits

also going to get quite expensive um and

problems like that. So that's why you

want to set token limits and

summarization strategies too. Poor error

handling your agent crashes completely

everything black screens not great. So

uh you want to validate your inputs

catch the exceptions and also implement

things called graceful failures so it's

not like your entire agent just dies.

That would be bad. Cost overruns. Big

one here. So things can get really

expensive actually if you just let it

spiral out of control. Uh for example,

if you're running your agent 24/7,

that's going to be really expensive. So

the solution is that you want to set

spending limit like hard limits. But

there's also a lot of other things that

you can do like caching your results,

having a database that's there, only

running agent when it's necessary. A lot

of ways to reduce the cost and then slow

response times. So the problem is that

maybe your users maybe they love your

tool, right? That's a good thing, but

your users end up waiting too long for

these results. So what you can do is

optimize your problems reducing tool

calls and using streaming using

streaming responses as well. So there

are a lot of solutions for these

pitfalls that you see. Okay. So um we

built this little visualization for so

that when you're building your first AI

agent like go through each of these

steps. So define your use case, choose

your tool, start simple, iterate, expand

and deploy monitor. Whenever you create

an AI agent, doesn't really matter how

complex it is. always go through these

steps in order to make sure that you

have a good AI agent that is functional.

Okay.

Yes. So, oh, I'm on time. I'm two

minutes late, but let me finish my this

last slide. Okay. So, the key takeaway

here is that uh we covered 28 popular

tools. You know, there's a bunch of

tools that are there. Definitely check

them out in more detail if you're

interested. Um, start small with a

single task and a single tool. Like I

said earlier, like don't feel like it's

so overwhelming that there's so much

that's going on over here. It, you know,

the entire boot camp is 28 days, right?

So like obviously I cannot fit 28 days

of things into this entire um

presentation. So it's okay if you feel

like you haven't covered all like you're

not sure about a lot of these things um

that's happening over here. Just just

the only thing I need you to remember is

that there's a step-by-step plan and you

need to follow this plan no matter what

agent it is that you're building. And

then you can add on a complexity as you

keep um improving your agent over time.

Um, how to choose between no code and

code. Choose no code for speed, code for

control. If you don't know how to code,

obviously go with the no code approach.

If you do know how to code very well, go

with the code approach. Test extensively

before deploying or else you're going to

have problems. And then iterate based on

real usage. You want to keep monitoring

and iterating and improving over time.

Yeah. Okay. All right. Let's see if

anybody has any questions. Oh, I need to

drink some water.

Okay.

Can you explain custom integrations? Um,

so integrations that I'm referring to

like custom connectors, uh, it would be

stuff like if we're talking about from

MCP for example, um, when you're

integrating like different tools

together into that is not readily

available. Like what's readily available

is maybe like the calendar API, right?

and you just have that as an MCP server

and then you can just use that and

you're fine with it. But what if you

want to have a tool where a combination

of tools that you don't have direct

access to? Then you need to build your

own custom integration. So you're able

to put together like two or three tools

that are together. You can also add

stuff uh like resources and like prompt

templates and things like that as well.

So you can grab everything and put them

together into a server into what is

called like an MCP server and then you

can have that custom so agents are able

to access it. That's a little bit more

detailed about this. Um but that's what

we mean by custom integrations. So

building out your own combination of

different tools and different usages.

Would this live be available on the

channel? Yes, the live stream will still

be available. Don't worry.

Bigger LMS are overkill for simple task.

That is true. Um that's why like often

times especially when we have multi-

aent systems, you don't actually need to

have like the fanciest model possible.

You want to have the cheapest model that

is the fastest model model in order to

accomplish your task. Like sometimes

instead of using like GPD40, you can go

GPD5 whatever like you can go with

something like

GPD3 right like we actually use GP3.5 a

lot in a lot of our agent systems

especially when you do like B2B clients

and stuff like that like when we build

solutions for B2B clients to reduce the

cost for the client a lot of parts of

that agent we would you like of an

agentic system would be using like super

cheap models and super fast models as

well because it's not necessary to have

something that is really really fancy

for all of your sub agents. Um, yeah,

like the the cost actually does matter a

lot, especially when you have a lot of

users where it's being run a lot. It it

really adds up.

Uh, what's an example of summarization

strategy for memory management? That's a

great question. So, say you have a

conversation with your AI agent and blah

blah blah, like a lot of things that you

guys talked about. Instead of having to

store all of that information in memory,

you can actually get the AI like part of

the agent, it would summarize the

entirety of the conversation prior only

keeping in the important parts and only

storing that part in memory. So that's

how you can reduce the amount of memory

that needs to be managed.

How did you learn all this about agents?

Uh okay. Well, I guess how did I learn?

That's a good question. I learned a lot

of things. I think that's one of them.

So I do have uh I it definitely helps

like I used to work at Meta. I've been

in the field since 2018. So like that

was a big part of it. Uh doing like

computer science as well. So yeah, I've

been like here for a while. Um but

honestly a lot of it is also like we

learn as a team. We have a lot of

clients that we work with. So especially

like B2B clients. You learn a lot when

you're working for different companies

and different industries. Um, and then

you're actually seeing your solutions

being implemented. So, a lot of it is

real life experience because of the

clients that we work with. Yeah, I guess

that answers your question. I also learn

a lot of stuff

online.

Um,

how do we make sure that our agents

understand the business context of the

use case we are trying to implement so

that it can write test case for us? That

is a good question. So you know um the

evaluations which is the test cases

themselves right these are often the

most treasured part of an AI company.

Like that is literally the most treasure

part, not the agents themselves, but the

actual test cases. And that's because

you in order to understand the business

context to write the best test cases to

make sure that your agent is good. That

is actually hard and requires good

domain knowledge for very niche um for

yeah like very very niche um

implementations, very niche domains.

So,

in order to write these test cases, I

don't actually recommend um getting an

agent to write your first few test

cases. I think people should write their

first few test cases

because it's going to be a lot better.

Um like you you can't get your be sure

that your agent understands the business

context for something, right? Like I I

really recommend in your first test

cases to actually go and physically

write them or if you don't know what the

answer is, go figure it out first. And

after the first 10 test cases were stuff

that is more systematic like making sure

like what happens if there's no input uh

what happens if it's like you know

random input like that kind of stuff um

like more technical stuff you can get

your agents be able to write it but the

business context itself um it needs to

be people. It really does need to be

people because if if your agents

understood all the business context that

anybody can build any agent and then we

would all be able to build really hyper

useful agents, right? Like there would

require no thinking, the agents will

just build their own agents. Uh I don't

know, maybe that'll happen one day, but

currently um that is not the case.

Question, how much time do you recommend

weekly for the boot camp? So let's see.

I would say very minimally four hours, 4

to eight hours, uh minimum of four

hours. If you cannot spend four hours,

then I do not recommend this boot camp.

Yes. So, two of them is going to be

spent because we're having like

workshops and Q&A sessions, but you're

also going to have like assignments each

week in which you'll be building your

own AI agents. So, throughout the uh

boot camp, you'll be building minimum of

four different agents. And there's also

like bonus assignments and stuff for

recovery like MCP um and these other

things as well. Those are not required

but they are there if you want to um do

them. But if you cannot shell out those

four minimum hours, I would say four to

six hours is the minimum. Eight hours is

like really really nice. Um with eight

hours you'll be able to build out the

agents that we that we've showcased. You

can do the code implementation. no code

implement implementation all the bonuses

and you would probably also have time to

actually build your own AI agent and be

able to get feedback from us as well. So

a lot of people who are coming into the

boot camp they do have an agent in mind

that they want to build. So, what tends

to be really useful is that if you can

build you you're learning stuff and

you're building out the agents for for

the boot camp, but you can also build

out your own agents and you can ask us

questions during the Q&As's and also um

like we have a community as well and you

can ask us specific questions for how to

implement your own agent. So, if you can

have eight hours you that's like plenty

of time for you to do all those things

but minimum four.

Is NA10 free or paid? Uh so they do have

a cloud version which is paid as a

subscription. You can host it yourself.

Um but you still need to pay for like

API costs. Yeah. So NA10 is free and

yeah they do have a free like 14-day

trial, but there are other associated

costs with it if you actually want to

use it and build AI agents. It's not

super expensive though, so it's not not

bad.

Yeah, it has Thank you, Joe. Free

self-hosted option. Yep. If you don't

use their cloud, you can host it for

free. All you have to do is pay for the

API cost for like the models themselves

per week. Per week. Four to eight hours

per day per week. No, no, no. Per week.

So, four hours per week for the boot

camp. Not per day. That's that that's a

lot. No, no, no. You don't need us to do

that per day. Is NA10 the same as ball

and lovable? No. It is very different.

Have I missed a lot? When did this

start? Uh, it finished already.

Actually, speaking of which, I'm going

to answer a couple more questions and

then I guess I will leave. Um, hey, I

joined your boot camp. Oh, hello. Thank

you so much for joining. I hope you

enjoyed it. We just finished the um I

can't tell who you are, Aspire, because

I don't know your username. I can't

match it to a name, but if you were you

part of the cohort that just finished.

Um,

if we miss a class due to work, can we

review recorded videos? Yes. Yes. So

everything in the boot camp is recorded.

Um yeah, don't worry about that.

Everything is recorded. You also have

access to everything after the boot camp

as well, like lifetime access to it. So

what people do is because say we have

the code implementation, the no code

implementation, and a lot of the bonuses

as well. It's fine if you can't get

through all of it. You can download

everything um and you have access to it

forever. There's also like a LinkedIn

community, a private LinkedIn community

where all the other alumni are there and

people work with each other and they

build stuff together um as well. So

yeah, don't worry

about that. We understand people are

busy.

Let's see

any other questions. I guess I can

switch back to me. Hello, it is me

again.

Um,

any other question?

Yeah, if you have more questions about

like the boot camp specifically as well,

if you sign up for the weight list, like

we'll actually let you know when

enrollment opens and like details about

these things as well. So, um it does

start on the 21st. So, the next

enrollment starts on the 21st at 11:00

a.m. EST in case you are interested.

Uh, okay. Seems like we're about done in

terms of

questions, but thank you all so much for

joining. Really appreciate it. I hope

this was a useful live stream. I think I

covered a lot of stuff. From a scale of

1 to 10, how useful did you find this

live stream? Like, were Yeah, I guess

like how much use did you get from this

live stream? I want a rating. Basically,

rate me. rate me from a scale of 1 to 10

in terms of the information density, how

useful it is. Um, I want to see how good

of a job I did before I sign off.

Yeah. Oh, thank you very much. That's

very sweet. You don't have to It's okay

if you hurt my feelings. That's okay. My

my goal is really like to cover a lot of

these things to increase awareness for

how to build it and also give you a

clear road map in terms of like these

are the steps that you need to do. 10 10

plus. Okay. Okay, 9 plus.

Okay, you guys are so kind. Thank you. I

appreciate it. Six. Okay, could use some

improvement there. Well, thank you for

for that. Appreciate it. I will take

that into account and try to do even

better next time. Uh, all right. Thank

you all so much for joining today. Sign

up for the workshop um so you can get

the slides. We'll send you the slides

and additional resources and stuff like

that on that email mailing list. If

you're interested in joining the boot

camp, it the enrollment goes out for the

uh weight list on October 21st, 11:00

a.m. EST. Uh do join the wait list

because we we sold out like within 30

minutes last time only to the weight

list alone. So it's something that you

know, thank you so much for for doing

that. So if you really are interested in

doing that, make sure you sign up for

the weight list and we'll send you

additional information there as well.

All right, have a wonderful rest of your

day, whether that be evening, morning,

afternoon, strange hour of the day. And

I'll see you guys in the next video or

live stream. See you.

Bye.