From Chatbot to Agent

In Lesson 1 I established what a language model does on its own. That leaves the term “AI agent” to explain, and why it’s replacing “chatbot” or “assistant” everywhere right now.

The boundary isn’t a hard line. Real systems sit on a spectrum. A workflow can contain an LLM without being an agent, and a chat-based assistant can use tools without becoming a fully autonomous agent. The useful question isn’t “Is this definitely a chatbot or definitely an agent?” It’s “How much of the workflow does the system decide for itself, what tools can it use, and how much human control remains?”

The chatbot: talk in, talk out

A chatbot takes your question and gives you an answer. Ask it to summarize an article, draft an email, or explain a concept, and it produces text. The interaction is closed. You ask, it answers, the conversation ends there unless you ask again. Nothing happens outside the chat window. If you ask a chatbot to “book me a flight to Chicago,” a pure chatbot can only describe how you might do that. It can’t go do it.

The agent: talk in, action out

An agent is a software system built around a model that can make decisions and take actions toward a goal. It may search the web, run code, read a file, call an API, send an email, or update a spreadsheet. The model generates outputs that can include structured tool calls or other actions, and the surrounding software executes them.

MIT Sloan describes this as AI that can perceive, reason, and act largely on its own, rather than waiting to be asked a single question and answering it. Forrester’s framing is blunter: a working agent behaves less like a chatbot and more like a small piece of infrastructure, something that has to plan, coordinate, and keep running across multiple steps (Forrester, 2026).

A simple industry example makes the distinction concrete: a chatbot tells you how to book a meeting. An agent checks the calendar, finds an open slot, sends the invite, and books the room, without you doing each of those steps by hand (Insentra, 2026).

What “autonomous” actually means here

“Autonomous” gets thrown around loosely. In practice, autonomy means the system can make some decisions and carry out multiple steps toward a goal without a human directing every step. Give it a goal, “find three vendors and compare their pricing,” and it may decide how to break that into steps: search, extract, compare, summarize. How much of that workflow it chooses itself varies from system to system.

That’s a meaningfully different capability than a chatbot answering one question at a time, and it’s also where a lot of new risk shows up. More on that later in the series.

Not every “AI agent” on the market deserves the name

Worth being skeptical about as a novice: the term “agent” is being slapped on a lot of products that don’t really do autonomous multi-step reasoning. One 2026 industry analysis estimated that of the thousands of vendors marketing “AI agents,” only a small fraction, roughly 130, are actually building genuinely agentic systems. The rest are repackaging older automation under a trendier name (MachineLearningMastery, 2026). The industry has a term for it: agent washing.

A decent gut check is to ask how much of the workflow the system decides for itself, rather than simply following a fixed script a human wrote in advance. Fixed scripts triggered by AI, “if this email arrives, draft this reply,” are useful automation, but they are not necessarily agentic in the sense this series is building toward. More agentic behavior means the model has meaningful latitude to choose what to do next based on what it observes.

Carry into Lesson 03

• A chatbot answers. An agent acts, across multiple steps, toward a goal.
• Autonomy means the system decides its own steps, not that a human is absent entirely.
• Watch for “agent washing.” Ask whether a product actually reasons through steps or just runs a fixed script with an AI label on it.

Next lesson, I open up an agent and look at the four pieces every one of them is built from.