A Task, Not a Question: What an Agent Does Differently

▸MODULE 1 / A TASK, NOT A QUESTION

A chat answers, an agent gets it done

You know the chat window. You type a question, you get back some text, you copy out the useful part, and the rest disappears. That is a conversation. It ends with an answer in a window, and afterwards the work is still sitting with you.

An agent works differently. An agent is an AI tool that carries a task through from start to finish instead of just answering a question. It is allowed to take several steps in a row, it is allowed to look things up along the way, and at the end it puts down a result. The difference is not intelligence. The difference is the shape of what you put in.

Picture two notes on your desk. One says "How do you write a good rejection?". The other says "Write a rejection to this enquiry, friendly, three sentences, and put it in my folder as a draft". The first note leads to a conversation. The second leads to a result you can hold in your hand.

The four parts of a task

Every workable task has four parts, and I will name them once now, because they carry the whole course. You will meet them again in every lesson.

Briefing. You say what should come out, for whom, and in what form. Not how it gets done, but what stands there at the end.

Delegation. You genuinely hand the work over, together with the material it needs. Not everything you own, but what the agent has to see.

Checking. You look at the result before you use it. Always. Even when it looks good. Especially then.

Filing. The result gets a place and a name. Otherwise you will have worked hard and still find nothing two weeks later.

Four parts, in that order. When a task goes wrong, one of them is almost always missing.

How you notice that your task is too vague

There are three signs, and you will spot all three immediately once you start watching for them.

The first: you get a list of possibilities instead of a result. When the answer opens with "You could, for example", you asked a question rather than giving a task.

The second: you do not know where the result is. It sits in the window and nowhere else. That is the filing part missing.

The third: you cannot tell whether it is finished. If you catch yourself asking "is that enough now?" while reading, you did not hand over a done criterion, and the agent could not invent one either.

A common early mistake: more text helps. It does not. A long task full of wishes and without a clear result is worse than three precise lines. Length is no substitute for clarity.

Three everyday tasks, side by side

So this does not stay abstract, here are three examples of the kind that could come up in your week.

"What is in this contract?" becomes: "Summarise this contract on half a page, for me as the contractor, covering the three points of term, termination and liability. Put the summary next to the original file."

"Can you help me with my quote texts?" becomes: "Write a quote text from these bullet points for a client in the trades, matter of fact, no marketing language, at most 200 words, as a draft I will revise."

"How do I tidy up my folders?" becomes: "Look at this folder and propose a structure for me. Do not move anything yet."

You can see the pattern. Result, recipient, form, place, and a point at which it stops.

I want to turn a question into a proper work task.

Here is my question:
[paste your question here]

Ask me these four questions one at a time and wait for my answer each time:
1. What result should stand there at the end, in one sentence?
2. Who is the result for, and in what form does that person need it?
3. Where should the result be stored once it is done?
4. How will I know that it is finished?

At the end, write my task out as a short, clear text
that I can use again next time.

The honest takeaway

This course does not show you click paths. It names Claude Desktop, ChatGPT and Copilot only as examples that such interfaces exist, and otherwise describes only the way of working. The reason is plain: interfaces change, menu items move, features get new names. Instructions built on buttons are wrong after a short while, without you noticing. The four parts of a task, by contrast, still hold with the next tool.

And there is a limit. An agent takes the doing off your hands, not the responsibility. What you pass on, you have checked. That one rule runs through all twelve lessons, and in the third lesson it gets serious.

In the next lesson we look at the material: what changes when an agent no longer sees only pasted text, but real files in a real folder.