Build an AI Agent · Lesson 1Free ~12 min · No code

From chatbot to agent — your first task agent.

Most people use AI like a vending machine: ask, get an answer, paste it somewhere. An agent works differently — you hand it a goal and it does the steps for you. This lesson shows what actually makes something an agent, which tasks to delegate first, and how to set one up for a real job.

The mental model

A chatbot answers. An agent acts. The shift is from "ask and copy" to "delegate and review."

When you chat with AI, you ask a question and paste the answer somewhere yourself. An agent flips that: you give it a goal, it makes a plan, takes the steps, and hands you a finished result. You stop doing each step and become the one checking the work.

The reframeYou're a manager now, not a typist. A good agent task is one you'd hand to a capable assistant with clear instructions. If you can write the instructions, you can build the agent.

Step 01 · Chatbot use, or agent task?

Four things separate an agent from a chat: it's given a goal (not just a question), works in multiple steps, can use tools/info along the way, and returns a finished result, then stops. Tap each one — is it a job for an agent, or just a chat?

Step 02 · Write your agent brief

Pick one repetitive, multi-step task that's low-risk to get slightly wrong, then write it like a brief for an assistant. Pick one option per row and watch the setup prompt assemble:

1 · Goal the end result, plainly

2 · Steps the rough sequence you'd follow

3 · Done looks like exactly what a finished result is

4 · Constraints what it must not do, when to stop

Brief:

0 / 4

Your agent setup promptPick options above and your prompt builds here…

"Show your plan first, then carry it out, then summarize what you did" keeps you in the loop. The agent plans, you approve, it acts, you review — the manager's loop.

Step 03 · Make the first one easy to trust

Your first agent should be narrow and safe. Good first tasks are:

Clear done condition — you'll know instantly if it worked.
Low risk — a mistake costs minutes, not money.
Repetitive — worth automating because you do it often.
Read-mostly — gathering and summarizing, not sending or deleting.

Left vague, agents wander — they over-research, invent steps, or take an action you didn't want. Keep the first one narrow and require it to confirm before anything irreversible. Autonomy is earned, not granted on day one.

Your challenge: delegate one real task

Write the agent brief (goal, steps, done, constraints).
Have the agent show its plan before it acts.
Let it run, then review the finished result.
Note where it wandered — that's what you'll tighten next time.

✓

Lesson complete

You can tell a chatbot from an agent, recognize the four traits of an agent task, write an agent brief, set one up that plans-acts-reports, and keep a human in the loop.

Up next

Lesson 2 · Give your agent tools — connect apps & multi-step work

Continue to Lesson 2 →

Open the Playground Back to Build an AI Agent