What Is an AI Agent?
At the end of Chapter 10, I wrote: "From the next chapter on, we will get into building the foundations."
This is that first foundation topic: the AI agent. It is a term I have used throughout this series without ever stopping to explain it properly. This chapter focuses on that one word.
Starting with the obvious question
You have probably seen the words "AI agent" showing up more and more lately. But if you have tried looking it up, you may have found several different explanations — and come away unsure which one actually applies to you.
I had that same feeling.
So instead of starting with a formal definition, I want to start with something more concrete: what is different about an AI agent compared to ordinary AI use?
Q. What makes it different from ordinary AI?
Ordinary AI use — for example, asking ChatGPT or Claude to "fix this text" — is a single round trip. You put something in, the AI gives something back. Done.
An AI agent removes that "one round trip" limit.
When it receives an instruction, it works out a sequence of steps on its own and carries them out one after another. It does not stop after one exchange. It keeps going: "research this, summarize it, move to the next task."
Q. How does it actually work?
Here is one example.
Say you give this instruction: "Read Chapters 5 and 7 of this series, pull out three keywords they share, and write a summary using those keywords."
If you ask ordinary AI, the process looks like this:
- You ask: "Read Chapters 5 and 7 and pull out shared keywords."
- The result comes back.
- You ask: "Now write a summary using those keywords."
- The summary comes back.
At each step, a human manages the sequence and issues the next instruction.
Give that same instruction to an AI agent, and the agent works out the steps itself — "first find the keywords, then build the summary" — and runs the whole thing through to the end. You do not need to think about what comes next at any point along the way.
Q. Is there a more everyday example?
Here is one drawn from how this blog actually runs.
Take writing a single article.
The instruction I give is one line: "Write a draft for Chapter 11." The agent reads the previous chapters to get a feel for the writing style, picks a structure that fits the topic, writes the draft, checks its own work, and makes revisions — all in sequence.
I do not need to give individual instructions like "now research this part" or "now fix this section." The agent thinks through the steps and moves on its own.
That said, the final decision of "publish this" is always made by a human. I record how that mechanism works in a separate chapter.
Q. Is the agent "intelligent"?
This is where a little care is needed.
It is true that an agent works through steps on its own. But that "thinking" is something different from human judgment.
More accurate to say: it runs as designed. What to do, and in what order — a human works all of that out in advance. Within the boundaries of that design, the agent operates autonomously (meaning: it moves without needing a human to prompt each step). But it cannot make judgments that go beyond the design.
"Intelligent" is not quite the right word. Think of it more as a machine that carries out a design precisely and reliably. That sense of things connects directly to what comes up in later chapters: "who does the designing" and "who does the final check."
Q. How does this series use the term "AI agent"?
One point worth stating clearly about how the term is used here.
In this series, "AI agent" means "a single AI unit with an assigned role." For example: the AI responsible for writing articles, the AI responsible for checking them, the AI responsible for talking with a human to get final approval — each of these is one agent.
This is a slightly broader usage than the standard technical definition. Reading it as "an AI that moves with a defined role" is close enough.
One term, pinned down
To put it in a single sentence: an AI agent is an AI that receives an instruction, works out the steps on its own, and carries out multiple tasks in sequence.
That is what sets it apart from ordinary AI use, where you ask something and it answers — a single back-and-forth.
The "AI organization" (here meaning: combining multiple agents into one working system) that this series will cover going forward is built on top of this idea. Getting a clear picture of "what a single agent is" is the starting point for everything that follows.
The next chapter builds on this in the same way — one term at a time.