---
title: 'AI Agents, AI Workflow - Explained'
source: 'https://youtube.com/watch?v=wpxvQi7mpno'
video_id: 'wpxvQi7mpno'
date: 2026-06-14
duration_sec: 620
---

# AI Agents, AI Workflow - Explained

> Source: [AI Agents, AI Workflow - Explained](https://youtube.com/watch?v=wpxvQi7mpno)

## Summary

This video explains AI, AI workflows, and AI agents in simple terms for non-technical viewers. It starts with familiar concepts like ChatGPT, then builds up to workflows and agents, breaking down jargon like RAG and ReAct with real-life examples.

### Key Points

- **LLMs as Brains** [01:29] — Large Language Models (LLMs) are the brains behind AI apps like ChatGPT, Gemini, and Claude. They take input and generate responses based on training data, but they are passive and don't know personal preferences or real-time info unless connected.
- **Workflows as Recipes** [03:09] — Workflows are step-by-step processes (control logic) that automate tasks, like checking past orders and suggesting food. They follow a fixed path and fail if the situation deviates from the programmed steps.
- **RAG Explained** [04:48] — RAG (Retrieval-Augmented Generation) is just a fancy term for looking something up before answering, a common tool within workflows.
- **AI Agents: Reasoning, Acting, Iterating** [05:27] — AI agents go beyond workflows by reasoning (e.g., considering weather), acting (checking restaurants and reviews), and iterating (critiquing and improving results). They use the ReAct framework (Reason + Act).
- **Real-Life Example: Wildfire Detection** [08:07] — Landing AI's wildfire detection uses AI to monitor forest cameras, with adjustable confidence thresholds to balance sensitivity and accuracy.

### Conclusion

AI agents represent a leap from passive LLMs and rigid workflows by adding reasoning, autonomous action, and self-improvement, making them capable of handling complex goals without step-by-step instructions.

## Transcript

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