Chapter 2 — AI Agents

What is an AI agent?

If you’re at this part of the book, it’s because you’ve already chatted with an AI, like ChatGPT, Claude, or Gemini.

You noticed that this interaction is like a game of ping-pong.

You send a message, the AI analyzes, thinks, and returns an answer to you.

You write something else after the answer, and it gives you another response.

It could be a “Good morning!”, a request to write a text, a question to clarify, a question about a medication, a conversation about a problem…

But it’s always that two-way exchange of messages, one message here, one message there.

Until now, we’re talking about your interaction with the AI.

It’s basically a back-and-forth conversation.

This emerged in late 2022 with the launch of ChatGPT.

At that time, even before that launch, I had already stepped away from my focus on digital marketing.

I was 100% focused and dedicated to artificial intelligence, with my own laboratory, doing my research.

A few months after ChatGPT, in March 2023, AutoGPT emerged.

And with it, the first AI Agent.

The week AutoGPT came out, I was already running it on my machine at home.

I was incredulous at what I saw happen before my eyes.

With AutoGPT, I was no longer exchanging ideas with an AI that thinks.

I was watching the AI that thinks exchange ideas with itself, multiple times, to accomplish a task I gave it.

So, for example, if before I was discussing some topics with the AI, sending a question, receiving the answer, returning the question, now it was different.

I had asked it to do research and write several articles for me.

And it kept doing that.

It searched and found an article, reflected on the article, had ideas about the article…

So it was interacting with itself, it wasn’t depending on me to send the ping-pong ball back.

I sent a ping-pong ball over and it played ping-pong with itself.

And that brought a lot of information back to me.

And it even stopped being just information and became extremely complex texts, things it didn’t have the capacity to do before.

Before AutoGPT, for example, I couldn’t say “ChatGPT, write an article for me.”

The article would come out poorly.

It would give a simplified version (a technical issue of attention window that I’ll tell you about in chapter 3) and bring a simplified article.

With AutoGPT, the story is different now.

I’d ask the AI Agent to write an article and it would start the article.

It would go there, improve the article, review the article, refine the article, finalize the article…

All of this without sending the article back to me.

And of course, it took longer, but it came back with the finished article.

It took longer because it was, at each interaction, sending the article back to itself and evaluating it itself.

As if it were doing my job of asking it to work on it again.

So, if before I had to receive an article draft and ask it to improve the introduction, review the conclusion, adjust the development…

Now it was doing that by itself.

So, basically, in simple terms, the AI Agent is an AI like ChatGPT, but you don’t need to keep hitting the ping-pong ball with it.

You send a more complex task and it goes on its own until it solves it.

Let me give you some analogies to better understand:

It’s like having a highly capable personal assistant. While a conversational AI needs you to guide it at each step, an AI Agent is like an assistant who, once you delegate a task, goes and executes it autonomously, making decisions and solving problems on their own.
It’s like turning on the autopilot of an airplane. Interacting with a conversational AI is like flying manually, where you need to constantly adjust your course. Using an AI Agent is like turning on autopilot: you set the destination and the plane flies on its own, adjusting its course as needed.
It’s like having a chef in the kitchen. A conversational AI is like a recipe - it guides you step by step, but you need to execute each step. An AI Agent is like a chef: you say what dish you want and he goes to the kitchen, chooses the ingredients, prepares, cooks and serves you the finished meal.
It’s like having a private driver. Interacting with a conversational AI is like using a GPS: it gives you directions, but you still need to drive the car. Using an AI Agent is like having a driver: you tell him where you want to go and he takes you there, dealing with traffic and choosing the best route.

→ Next: 2.2 What is the difference between AI (LLM) and AI agent