Chapter 3 · Chatbots vs Agents

Stop asking questions.
Start giving goals.

A chatbot answers and stops. An agent takes a goal and goes and does the work — the shift the whole field is built on.

The shift

Don't hand it a question. Hand it a goal.

You ask a chatbot something and it replies — reactive, one shot. You give an agent a goal and it goes and does the work: plans, uses tools, checks its own results, and adapts.

The keyword is autonomy.

Barely any hand-holding. That's the moment a language model stops being a tool you operate and starts being a teammate you delegate to.

One goal in

One goal in. A sequence of actions out.

Here's the goal you type once — no follow-ups:

🎯 "Research my top 3 competitors, compare their pricing & features, write a report, and email it to my team."

Four jobs bundled into one sentence. Watch what happens next.

The autonomous run

It just… does all four.

1 · Research

Web search + reads each competitor's site.

2 · Compare

Pulls pricing & features into a structured table.

3 · Write

Drafts the comparison and the positioning.

4 · Email

Sends it to the team via the email tool.

Zero follow-up prompts. A chatbot would have answered step 1 and stopped.

Side by side

Reactive vs. autonomous.

🤖 Chatbot

Reactive · one shot

You ask → LLM → Answer

Responds to exactly what you said, then waits. Nothing happens until you type again.

🚀 Agent

Autonomous · many steps

Goal → LLM + tools → Result

Loops — plan, act, observe — until the goal is actually met.

Same model. It's the loop that changes everything.

On a loop

Think → act → observe. Repeat.

A chatbot is a single round trip. An agent wraps that same model in a loop, so it can take many steps before it's done:

Think

Decide the next move toward the goal.

Act

Call a tool — search, code, email.

Observe

Read the result, then loop again.

It keeps looping until the goal is actually met — that's the engine of Chapter 4.

The test

Five properties make something agentic.

When is a system truly an agent, not just a chatbot with extra steps? Look for five things:

👁️ Perception

Reads files, browses, sees images & API results.

🧠 Reasoning

Thinks a problem through and breaks it into pieces.

🗺️ Planning

Makes multi-step plans, adjusts as things change.

⚡ Action

Actually does things — calls APIs, runs code, sends email.

🔄 Adaptation

Learns from feedback and its own mistakes.

The scorecard

Chatbot 2/5 · Agent 5/5.

A plain chatbot perceives your text and reasons a little. A real agent has all five:

🤖 Chatbot · 2/5 perceives & reasons a little 🚀 Agent · 5/5 owns a goal end to end Perception partly Reasoning partly Planning no Action no Adaptation no

The gap

The missing three are the whole story.

Perception and reasoning aren't enough on their own — plenty of chatbots have them. What turns thinking into doing is the three a chatbot lacks:

🗺️ Planning

Sequences the steps and re-plans when reality shifts.

⚡ Action

Reaches out and changes the world — not just talks about it.

🔄 Adaptation

Notices what went wrong and course-corrects.

A chatbot describes the plan. An agent runs it.

Why action matters

Tools are how it touches the world.

Reasoning happens in the model's head. Tools are what let it act on the outside — search the web, run code, send the email, update a doc.

No tools, no agency.

Strip the tools away and even a brilliant reasoner is back to being a chatbot — it can only talk about the work, never do it.

The keyword

Autonomy.

The capacity to act independently and make its own choices toward a goal. That's the line between a tool you operate and a teammate you delegate to.

Chapter 3 in one breath

A chatbot answers. An agent achieves.

Hand a chatbot a question and it replies. Hand an agent a goal and it perceives, reasons, plans, acts, and adapts — looping until it's done.

Same model underneath; autonomy on top.

Next up

Chapter 4 · The Core Agent Loop.

We've named the five properties and the one word — autonomy. Next we open the engine that actually delivers it: the think → act → observe loop every agent runs on.

Press S for speaker notes · F for fullscreen · to advance.