[00:02]
AI agents
[00:07]
AI. Since the majority of videos on a
[00:10]
YouTube about AI either too technical or
[00:13]
too basic, I decided to record this one
[00:16]
for all of our subscribers who either
[00:19]
have just a little bit or no technical
[00:21]
background at all. In this video, I will
[00:24]
explain what AI, AI workflows and AI
[00:28]
agents are. But don't worry, I will take
[00:31]
it step by step and explain it in a
[00:34]
simple words and we'll utilize our
[00:35]
educational strategy we use in our boot
[00:38]
camp starting with the things that you
[00:40]
are already familiar with just like chat
[00:43]
GBT rock or cloud apps. Then we're going
[00:46]
to move forward to AI workflows and
[00:48]
finally to the most popular hype words
[00:51]
these days such as AI agents. And I
[00:54]
promise those intimidating words such as
[00:57]
rag, react or agents are going to become
[01:00]
much simpler once we break them down
[01:03]
using real life examples. But before we
[01:06]
get started, let me quickly remind you
[01:08]
who we are and then we will kick it off.
[01:10]
My name is Sergio Kchenko. I'm a
[01:11]
software QA engineer, lead manager, and
[01:13]
a senior engine manager of ASD that in
[01:15]
the past, but these days I'm helping
[01:18]
people like you to become a QA from
[01:20]
scratch or to improve your existing
[01:22]
skills. Now, let's get
[01:25]
[Music]
[01:29]
started. Level one, large language
[01:32]
models or LLMs. Every single app you
[01:36]
have played with in terms of AI such as
[01:39]
Chat GPT, Google's Gemony, Claude or
[01:42]
Tesla's Oron Musk's Grog, those are just
[01:46]
applications built on a top of LLMs. And
[01:50]
LLMs are simply brains for every single
[01:53]
AI app. Let me explain that in a simple
[01:56]
words. Imagine using Google. You type
[01:58]
something, Google looks it up, you get
[02:01]
an answer. So that's pretty much how
[02:03]
nowadays internet works, right? In a
[02:05]
simple words. Now let's think about
[02:07]
LLMs. You give a prompt or same input.
[02:10]
You pretty much type something. Then it
[02:13]
uses what it learned and it will get you
[02:15]
the response just like in a Google but
[02:18]
instead of the Google's server, we have
[02:20]
LLMs or large language model which uses
[02:24]
a lot of data behind it that was it was
[02:26]
trained on. So far pretty
[02:28]
straightforward, right? input LLM and
[02:31]
your response. Easy peasy. But here's
[02:33]
something to keep in mind. Every single
[02:36]
LLM or large language model was trained,
[02:39]
even though it was trained on a lot of
[02:41]
data, it doesn't know your preferences.
[02:44]
It doesn't know much information about
[02:46]
you. It cannot access your files, your
[02:48]
order history, or realtime info unless
[02:51]
you specifically connected to it. And
[02:54]
most importantly, LLMs are passive,
[02:56]
which means they're simply sitting there
[02:58]
and waiting for you to input some data
[03:01]
to ask them questions. They don't go and
[03:04]
do things on their own. And that's why
[03:07]
we need
[03:09]
workflows. Now, let's imagine familiar
[03:12]
situation. You are hungry and you're
[03:15]
telling your AI or chat GPD, next time
[03:19]
I'm going to be talking about food.
[03:21]
offer me something based on my previous
[03:23]
order history from Uber Eats and it will
[03:26]
definitely fail because it has no idea
[03:29]
what Uber Eats are in terms of your
[03:31]
orders cuz you did not give access to
[03:34]
Uber Eats. But if you will build a
[03:37]
workflow, then you could give access to
[03:41]
chat GPT or the other LLM app. So it
[03:45]
could access your orders previously,
[03:47]
take a look at them, and then give you a
[03:49]
recommendation based on your previous
[03:51]
orders. And you could even specify such
[03:53]
as tabs as check past orders, fix
[03:56]
something with a five plus star rating,
[03:59]
send a suggestion. And that works until
[04:02]
you say this. Can you recommend
[04:04]
something new and then you will fail?
[04:07]
Why? Because the path you programmed
[04:10]
doesn't include look for new options.
[04:12]
Workflows are just like recipes. They
[04:15]
only work the way they were written. And
[04:18]
in technical terms, it's called control
[04:21]
logic. Now, let's say you expand it a
[04:24]
little bit. You add Yelp reviews, Google
[04:27]
Maps, and make it read the
[04:29]
recommendation out loud.
[04:32]
Great. Based on your past preferences, I
[04:34]
found a five-star vegan sushi place
[04:36]
nearby. Want directions? Still just a
[04:39]
workflow. Still not really thinking, but
[04:42]
doing step by step what you told it to
[04:45]
do. Quick myth or technical term
[04:48]
breaker. You've heard about term rag,
[04:50]
which is retrieval, augmented
[04:52]
generation. Don't worry, it's just a
[04:54]
fancy way of saying look something up
[04:57]
before answering. Just the way we do
[05:00]
with the workflows. That's it. It's just
[05:02]
one of the tools inside of the workflow.
[05:05]
And here is actually the main problem of
[05:07]
workflows. If the output isn't right, if
[05:11]
the recommendation doesn't sound tasty,
[05:14]
you will have to go back, tweak the
[05:17]
setup, and run it again. You are still
[05:19]
doing thinking. You're still the one in
[05:22]
control. And that's where AI agents come
[05:25]
[Music]
[05:27]
in. Let's stay with the same example.
[05:30]
Right now you're the one making
[05:32]
decision, what tools to connect, what
[05:34]
steps to follow, when to make changes.
[05:37]
But what if we replace you a human
[05:40]
decision maker with actual AI itself?
[05:43]
Then that's the time when workflow,
[05:46]
fancy workflow becomes even fancier AI
[05:49]
agent. And an agent can do three things
[05:53]
that workflow cannot. Number one, reason
[05:57]
or reasoning. It doesn't simply check
[06:00]
the past orders. It thinks, "Oh, it's
[06:02]
actually cold outside today. Maybe he
[06:04]
wants something warm." Or, "Oh, he did
[06:08]
order sushi 20 times within the last 10
[06:10]
days. Probably we should offer him
[06:12]
something else." It reasons through
[06:15]
context just like you would. Number two,
[06:19]
act. Now, AI says, "Let me check if
[06:22]
there are any restaurants nearby." Then
[06:24]
it will also check if they're actually
[06:27]
open now. And finally, it will probably
[06:29]
read some reviews before making
[06:31]
decision. And it does that all on its
[06:34]
own without you giving exact stepbystep
[06:38]
instruction just like you did with the
[06:40]
workflow. It figured it out on its own.
[06:43]
Let's say the agent suggests a
[06:45]
restaurant, but then realizes the
[06:48]
reviews are meh. Maybe the curry was too
[06:51]
spicy. It doesn't send you
[06:53]
recommendation and simply call it a day.
[06:56]
What it does, it goes back, finds a
[06:59]
better option and only then sends you
[07:02]
final and improved result. That's called
[07:05]
iteration. And it can actually do
[07:06]
something cooler. It can critique its
[07:09]
own answer with another AI model before
[07:12]
even sending it to you. That's what
[07:14]
makes it different from regular
[07:16]
automation or workflow. It can actually
[07:19]
think. Second nerdy word of the day,
[07:22]
react. You might have heard this word
[07:24]
but probably got scared because it's
[07:26]
something related to AI that you haven't
[07:28]
heard of. But let's take it apart. It's
[07:30]
very simple. Re stands for reason and
[07:34]
act stands for acting just like the
[07:37]
second difference with the workflow. And
[07:40]
that's why React is one of the world's
[07:42]
most popular frameworks for building AI
[07:46]
agents. And that's what pretty much it
[07:48]
does, right? aside of iterating. By the
[07:51]
way, I've been mentoring people for 8
[07:53]
years. And if you're interested to learn
[07:54]
how to build AI workflow or AI agent or
[07:57]
how to become a software QA engineer
[07:59]
completely from scratch, you can follow
[08:01]
the links right below this video. Let's
[08:05]
continue. Let's take a look at the real
[08:07]
life examples. And we have a lot of
[08:09]
different ones, but I'm going to pick
[08:11]
one of the most useful ones for our
[08:13]
community. Detecting wildfires. I'm
[08:15]
going to click right here. And by the
[08:16]
way, we're using landing AI website. And
[08:20]
I'm going to click on one of the
[08:22]
pictures to see how exactly it will
[08:26]
detect the wildfire because we cannot
[08:28]
have people all over the world
[08:31]
constantly monitoring all of the forest
[08:33]
cameras simultaneously. But we can have
[08:35]
AI and you can clearly see it right here
[08:38]
that there is a fire. As the human you
[08:40]
can see that but AI can also see that
[08:43]
and you can even specify the threshold
[08:45]
of how confident it should be. If we
[08:47]
minimize the threshold it will specify
[08:50]
more you will find more fires or more
[08:53]
things that look like fire. If we
[08:56]
increase it then it will be more
[08:58]
challenging. Let's say that way it will
[09:00]
be more sure before it say hey there is
[09:04]
a wildfire. So if you if you go all the
[09:06]
way to 100% it will say that there is
[09:08]
none. But if you at least get it to 85
[09:11]
or 90, you will still show you, hey,
[09:13]
that there is a fire and the government
[09:16]
or any companies that run this AI can
[09:19]
simply specify this threshold. Now,
[09:22]
quick recap. Level one, LLMs. It's just
[09:25]
like human brains. You ask it a
[09:28]
question, it gets you an answer, just
[09:30]
like the thing that we have right here.
[09:32]
Level two, AI workflows. You design a
[09:35]
process and AI follows it step by step.
[09:38]
And level three, AI agents. You give it
[09:42]
a goal and AI will figure out how to get
[09:45]
there on its own. If this video helped
[09:48]
you to make sense out of AI, AI
[09:51]
workflows and AI agents, and now you're
[09:53]
thinking possibly to build one. Leave a
[09:56]
comment right below this video and let
[09:57]
me know what would you like me to show
[09:59]
you how to build on the next video.
[10:01]
either send a message create AI
[10:03]
workflows or create AI agent and I will
[10:07]
do it on the next video. Now, thank you
[10:09]
for watching. Get some workout, get some
[10:11]
rest, drink a lot of water and I'll see
[10:12]
you next time.