Lesson 01: The Agent Is a Worker, Not a Talker

Lesson objectives:

• Tell the difference between a chatbot question and an agent task.
• Name the two things an agent needs before it can act: a clear goal and a way to check its work.
• Rewrite one chat-style request into a task-style request.

Prerequisites: None | Previous << README | Next 02 Write the ask >>

You already chat with AI. So why does handing it a real task still feel risky?

Most people start with AI the same way: they type a question and get an answer. You ask, "What is a good budget laptop?" and the AI answers in a paragraph. That is a chatbot. The AI stays passive. It responds to what you said, but it does not go off and do something for you. An agent is different. An agent takes a goal, chooses steps, uses tools, and returns a result. It can open a file, search the web, or run a small program for you¹.

That difference is not magic. It changes what you do. When you chat, your job is to keep asking until you understand. When you delegate to an agent, your job is to say what "done" looks like, then check that it got there.

Explanation

Think of a chatbot as a helpful librarian. You ask, it answers. You ask again, it answers again. An agent is more like a temp worker you hand a folder to. The temp worker can leave the desk, look things up, write a draft, and come back. If the instructions are vague, the worker will guess. If there is no way to check the result, you will only find the mistake later.

Agents do best when the task has three things: a clear goal, a bounded scope (a clear limit on what the agent should not do), and a way to verify the output¹. A clear goal means one sentence that says what finished looks like. A bounded scope means what the agent should not do. A verification method means you have a checklist to compare the result against.

The leap from chat to delegation is small but specific. You stop asking "What do you think about this topic?" and start saying "Read the attached notes, pull out the action items, and put them in a table with owner and deadline." The first is a conversation. The second is a job.

The loop works like this: the user does not just receive an answer; the user receives an answer and checks it against the acceptance criteria (the checklist that defines "done") that were built into the original ask.

Worked example (follow along)

Sarah has a long email thread with a contractor. She wants to hand it to her agent and get a clean summary. Here are two ways she could phrase it.

Chat style: "What is going on in this email thread?"

This gives the AI no finish line. It could answer with a paragraph, a timeline, or a list of complaints. Sarah cannot tell if it is good or bad.

Agent style: "Read the email thread below. List every promise the contractor made, the date they made it, and whether it has been kept. If a promise has no date, write 'no date.'"

Now the agent has a clear goal (extract promises), a bounded scope (only promises, not opinions), and a verification method (every row needs a date and a kept/not-kept status). Sarah can check the table against the thread.

Your turn (faded example)

Rewrite this chat-style request into an agent-style task. Fill in the blanks.

Chat: "Tell me about my bank statement."

Agent: "Read the attached bank statement. List every transaction over $____, and for each one write the date, the merchant, and the amount. Do not include transfers between my own accounts. Put the results in a ____-column format."

Answer: The blanks are 100 (or any threshold the user picks) and three.

Summary + what's next

An agent is not a better chatbot. It is a worker that needs a goal and a way to be checked. The skill is learning to stop asking questions and start delegating tasks. In the next lesson, you will turn a vague wish into a clear ask with four parts: task, requirements, boundaries, and acceptance criteria.