Discuss before you produce
This lesson is still being tested: it improves with learners' feedback. If something is unclear, it is the lesson that needs fixing, not you.
By the end of this lesson, you will be able to
- have your request rephrased before the AI gets started;
- ask for two options and choose knowing their pros and cons;
- have your idea critiqued, honestly;
- give your green light (“go”) only when the request is clear.
Why it matters to you
When the AI heads in the wrong direction, you often find out by reading the result: then you have to fix everything. Discussing first costs two or three messages, and often saves you from redoing a whole document. This is the second step of the method: after SAY, DISCUSS.
Before you start
| You need | Your assistant, free account, and your “My requests” note |
| Cost | €0 |
| Duration | 35 minutes |
| Reminder | The project in the example is made up; for yours, nothing sensitive |
Understand it in 3 minutes
Discussing before you produce is three steps, then a green light.
| Step | What you ask | What it gives you |
|---|---|---|
| Rephrase | “Rephrase my request in 2 sentences, and tell me what you’re missing.” | You see what the AI understood, and you correct it right away |
| Offer two options | “Give me 2 ways to do it, with their pros and cons.” | You choose, instead of taking the first idea that comes up |
| Critique | “What could go wrong? Be frank.” | You discover a risk before it costs you time |
| The green light | “Go: now do…” | The AI produces, taking into account everything that was said |
One sentence changes everything: “Don’t write the result until I say go.” The AI can rephrase, ask questions and suggest options; the final result waits for your green light. Without this sentence, it often wants to deliver right away.
For an honest critique, ask a neutral question. You saw it in module 1: the AI tends to agree with you. “My idea is good, right?” calls for a yes. “What could go wrong?” lets it disagree with you.
When should you skip this step? For a simple question (“translate this sentence”, “give me a synonym”), discussing would waste time. The longer or more important the work, the more time the discussion saves.
Do it together
The project in the example is made up: organizing a yard sale on your street.
1Step 1: Get it to rephrase and ask questions
I'd like to organize a yard sale on my street, in the spring.
Before suggesting anything, rephrase my idea in 2 sentences, then ask me the 3 questions that would help you most. One question at a time.
Don't write the result until I say "go".
Read the rephrasing. If a point is wrong or missing, say so. Then answer its questions with made-up answers (for example: about twenty neighbors, a Sunday, no budget).
2Step 2: Two options
Give me 2 ways to organize this yard sale, with the pros and cons of each. Tell me which one you would choose, and why.
Choose one option. You can also mix the two: “I’ll take option 1, with the idea of the posters from option 2.”
3Step 3: Get it critiqued
What could go wrong with the option I chose? Be frank: give me the 3 main risks, and what I should check myself.
If the AI talks about permits, rules or deadlines, its answer is a lead, not proof: for a yard sale, the real answer is with your local authority (your town hall or city office). You will see how to check in lesson 4.
4Step 4: The green light
Go: now make the task list for the chosen option, taking into account my answers and the risks. For each task: what, and when.
Compare with what you would have gotten by directly asking “Make me the task list for a yard sale”: today’s list takes your situation and the risks into account.
Your turn
A real project from your life, small or medium: a meal, a tidy-up, a move, a presentation, an application…
- Write your request with the 4 elements from lesson 1, and add “Don’t write the result until I say go”.
- Go through the three steps: rephrase, two options, critique.
- Give your green light, then read the result.
- Keep in “My requests” the discussion sentence that helped you most.
Check your work
It worked if:
- the AI rephrased your request, and you corrected or confirmed its rephrasing;
- you chose an option knowing why;
- you got a critique with at least one risk, and you decided what to do about it (even if you had already thought of it);
- the final result was not written before your “go”.
Common mistakes:
- answering “yes” to the rephrasing without reading it: this is the moment to correct;
- asking a leading question (“it’s a good idea, right?”): the critique will be weak;
- discussing for twenty messages for a simple request: two or three are enough;
- taking a rule quoted by the AI (permit, deadline, price) for a certainty: it is a lead to check;
- letting the AI produce before your green light: if it does, remind it “I didn’t say go: go back to the rephrasing”.
Keep it and pick it up later
- Your “My requests” note now contains a discussion sentence that works for you.
- Your next important request contains “Don’t write the result until I say go”.
Going further
optional
- You will find this step again with an agent in module 4: Claude Code’s Plan mode prepares a plan, and only changes your files after your approval.
New words
4 words explained
| Word | What it means |
|---|---|
| Rephrase | Say a request again in your own words, to show what you understood |
| Option | One possible way of doing something, with its pros and cons |
| Neutral question | A question that doesn’t suggest the answer: “What could go wrong?” rather than “It’s good, right?” |
| Green light (“go”) | The moment you allow the AI to produce, once the request is clear |
Check your understanding
3 questions, no grade. An explanation appears after each answer.
What's next?
Lesson 3 · Iterate: correct, ask for alternatives, ask for a critique: improve an answer without starting over.
Module 2:
✏️ Do the exercisesMy space: your progress, your exercises and your last lesson.
Follow this lesson with Claude Code
With the course companion installed, open Claude Code in your working folder and type:
/formation-ia:suivre m02-l02 enIt reads the up-to-date lesson on this site and guides you step by step, without changing anything without your consent. Install the companion. Claude Code is not included in the free plan: an eligible Claude subscription, or a Console account billed by usage for the terminal.
Sources and checks
2 sources
Date, version, sources Written on 24/10/2026. The method in this lesson works with any assistant. The tendency of AIs to agree with the person using them is described by an Anthropic study (“Towards understanding sycophancy in language models”, anthropic.com), read on 24/10/2026. Claude Code’s Plan mode is described according to its documentation (code.claude.com/docs/en/permission-modes), read on 24/10/2026.
Sources checked: