Unit 7 · Lesson 1
From answering to acting
What turns a chatbot into an agent — and why that changes everything.
Welcome to the Builder Track. You've got the foundations — now we build things that do stuff in the world.
Back in Unit 1, you learned the five verbs. Four of them just make output: classify, predict, generate, retrieve. The fifth one is different. The fifth one is act.
An AI that can act doesn't just tell you the answer. It picks up tools and takes steps. That kind of AI has a name: an agent.
A chatbot hands you words. An agent reaches out and changes something. That's a much bigger deal — and a much bigger responsibility.
One idea: the think–act–observe loop
An agent runs in a loop. It's not one answer — it's a little cycle it repeats until the job is done:
| Step | What the agent does |
|---|---|
| 1. Think | Looks at the goal and decides the next step. |
| 2. Act | Uses a tool — searches the web, sends a message, moves a file. |
| 3. Observe | Looks at what happened. Did it work? |
| 4. Repeat | Uses what it learned to pick the next step. |
Say the goal is "find me three pizza places open now and put them in a list." A chatbot would guess from memory. An agent would actually search, read the results, check the hours, and build the list — looping until it's done.
The power is obvious. So is the catch: an agent that can do things can do the wrong thing — and it does it, not just says it.
Do the thing
For each task, decide: does this need a plain chatbot (just words) or an agent (tools + steps)?
- "Explain how photosynthesis works."
- "Book the cheapest bus ticket to Grandma's for Saturday."
- "Write me a poem about rain."
- "Check every link on my website and tell me which ones are broken."
Quick check. 1 and 3 are pure generate — a chatbot's whole job. 2 and 4 need an agent: they require acting in the world (searching, clicking, buying, checking) and looping until done. Notice that #2 also spends real money — which is exactly the kind of action you'd want to approve before it happens. Hold that thought; it's the whole next lesson.
Why this matters
Agents are where AI stops being a fancy typewriter and starts being a worker. That's exciting and it's the reason to slow down.
- "Answer" and "act" are different risk levels. A wrong answer you can ignore. A wrong action already happened.
- The loop is the unlock. Think, act, observe, repeat — that cycle is what lets an agent handle a real multi-step job instead of one-shot guessing.
Next lesson, we do what you always do in this course: go looking for where it breaks. Because when an agent goes wrong, it goes wrong while doing things — and that's worth seeing clearly.