The one idea
Most people use an AI tool in a single mode: ask a question, get an answer. That mode is real, and it is roughly the least valuable of the ten available.
The modes below differ in who is doing the thinking. In answer mode the tool thinks and you receive. In the two most valuable modes — critic and simulator — you think and the tool provides resistance. That is where the gains are.
Decide what job you are giving it — explain, argue, pretend, check, build — before you type. Different jobs need different instructions.
Match the mode to the stage of work. Early: brainstorm and explain. Middle: transform and analyse. Late: critique, evaluate and simulate. Most people only use the middle.
Modes differ in how much you can verify the output yourself. Generative modes are self-evident. Analytical modes need the source attached. Factual modes need external checking.
The ten modes
| Mode | You are asking it to | Genuinely good for | Cannot do |
|---|---|---|---|
| Explain | Make something understandable | Unfamiliar concepts, jargon, dense text | Know your organisation's version of it |
| Tutor | Teach you, with practice | Building a skill over sessions | Assess you honestly without being told to |
| Brainstorm | Produce many options | Escaping your first three ideas | Know which option is right for you |
| Critique | Attack your work | Finding holes before your reader does | Care about your feelings, unless told to |
| Transform | Change form, not content | Summarise, restructure, translate, reformat | Add information not in the source |
| Analyse | Find patterns in material you supply | Comparing documents, extracting themes | Analyse data it cannot see |
| Simulate | Play a role you rehearse against | Difficult conversations, interviews, objections | Be the actual person |
| Research assistant | Map a topic and point at sources | Vocabulary, structure, where to look | Be the source |
| Build | Make a working thing | Drafts, formulas, scripts, structures | Guarantee it is correct |
| Evaluate | Judge against stated criteria | Consistent first-pass screening | Make the decision |
Three are worth expanding, because they are where the difference between an average and an excellent user shows up.
Critic — the most underused mode
Everyone asks "is this good?" That is a request for approval and you will get one, because agreement is the more plausible continuation of a conversation you started by sharing your work.
The critic mode requires you to remove that escape route:
"Here is my proposal. Do not tell me what is good about it. Give me the four strongest objections a sceptical finance director would raise, ordered by how much damage each does. For each, tell me whether it is fatal or fixable."
The value is not that the critique is brilliant. It is that it is free and arrives before the meeting. You will otherwise hear these objections from a person, in a room, when you cannot address them.
The strongest version is asking it to argue against a position it has just taken. Have it make the case for your plan, then say: "Now you are the person who will lose budget if this goes ahead. Make the strongest case against." Reading both is how you find where your reasoning is thin.
Simulator — the most valuable and least attempted
You can rehearse a conversation you are dreading, as many times as you like, at no cost.
"You are a hiring manager for a business analyst role at a logistics company. Interview me. Ask one question at a time and wait for my answer. Follow up when my answers are vague, the way a real interviewer would. Do not give me feedback until I say stop."
Two instructions do the heavy lifting: one question at a time and wait (otherwise you get a list of questions, which is not practice), and follow up when I am vague (otherwise it accepts your first answer and you rehearse nothing).
This works for salary negotiations, difficult feedback, client complaints, declining a request from someone senior, viva questions. The rehearsal is real even though the counterparty is not — you are practising producing words under mild pressure, which is most of what fails on the day.
What it cannot do is be the actual person. Your manager has a history with you, a mood, and a budget. Simulation gets you fluent, not prescient.
Evaluate — good, with one condition
You can hand it criteria and material and get a consistent first pass. "Score each of these twelve applications against these four criteria, one to five, with a one-line justification for each score."
The condition: the criteria must be yours and stated. Without them you get the average of what such things are usually judged on, applied invisibly, and you cannot say why anything scored what it did.
You are presenting a system migration plan on Thursday. On Tuesday you spend twenty minutes: critic mode for the four strongest objections, then simulator mode with the tool playing the operations head who has been burned by a migration before and will ask what happens if rollback fails.
On Thursday you have already heard the hardest question. You answer it in one sentence instead of discovering, live, that you had not thought about it.
You are learning SQL. Tutor mode: "Teach me joins. Give me one problem at a time against a sample table. Do not show the answer. When I get it wrong, tell me what my query would return and let me work out why."
That last instruction is what makes it teaching rather than demonstration. The default is to show you the answer, which feels productive and teaches little.
Choosing a mode
1 of 5Ask what stage you are at. Understanding something, generating options, doing the work, or checking the work. Each stage has its modes and using the wrong one feels like the tool is being unhelpful.
Try this
For each situation, name the mode and one instruction that would make it work:
- You have written a resignation letter and want to know if the tone is right.
- You need to understand a 40-page vendor contract before a call tomorrow.
- You must tell a teammate their work is not good enough, and you are dreading it.
- You have twelve customer complaints and want to know what they have in common.
Your challenge
Level 3 · IndependentPick a real thing you have to do in the next fortnight that makes you slightly nervous — a conversation, a presentation, an interview, a negotiation.
Use three modes on it, in order: critic on your plan or material, simulator to rehearse the encounter at least three times with the difficulty increasing, then tutor or evaluate to review how you did.
You have succeeded when you can name one specific thing you will do differently because of the rehearsal — a sentence you will not use, a question you were not ready for, an assumption you had not noticed.
If you finish and nothing changed, the simulation was too easy. Tell it to be harder and run it again.
What people usually get wrong
- Using answer mode for everything. The habit is invisible until someone shows you another mode.
- Asking "is this good?" You are requesting approval. Ask for the objections instead.
- Letting the simulator be pleasant. A counterparty who accepts your first answer is not practice. Instruct it to push back.
- Roleplay without a stopping rule. You get a monologue instead of an exchange.
- Treating a critique as a verdict. It generates plausible objections, not informed ones. You decide which are real.
- Evaluating without stated criteria. You get invisible criteria you cannot defend.
- Staying in one mode for a whole conversation. The good sessions switch three or four times.
How someone experienced does it
The pattern experienced users converge on is a loop: generate, then attack, then rehearse. Draft with one mode, break it with the critic, practise defending it with the simulator. Each pass is cheap and the compounding is substantial — most people ship the first draft of their thinking because attacking it used to require finding a willing colleague.
The subtler habit is using the simulator on people who are not adversaries. Before writing anything for a specific audience, they run it: "You are a warehouse supervisor with fifteen years of experience and no patience for head office. Read this notice and tell me what you would think and say." The answer is generic in the details and often exactly right about the reaction — because how groups of people respond to a type of message is a genuinely common pattern, even when the specific person is not.
And they know the mode with the worst return: asking it to make a decision they should own. Not because the answer is bad — it is often reasonable — but because they will have to defend the decision in a room, and "the AI recommended it" is not a defence anyone accepts. Which is the subject of the next lesson.
When not to use this
Simulation is the wrong tool when you need to know what a specific person will do. It models the type, not the individual. Someone who knows your manager will give you better predictions in two minutes.
Critique is the wrong tool when you have not finished thinking. Attack a half-formed idea and you will abandon it for reasons that would not have survived another day of work.
And none of these modes replace talking to someone who has actually done the thing. The simulator does not know what happened last time your company tried this. A colleague does.
Prove it
Take one task you already do with AI in answer mode. Run it again in two other modes — critic and simulator are the ones to try first.
Write three lines: what answer mode gave you, what the other modes gave you, and which you would use again for this task. Some tasks genuinely want answer mode. The point is that you chose it.
Keep learning this
Paste this into any AI assistant. It turns the assistant into a tutor that tests you instead of just answering you.
Act as an experienced practitioner who is good at teaching. I have just learned different modes of working with an AI assistant. Assume I am intelligent but relatively new to this — treat me as intermediate level. Work through this in order, and wait for my reply at each step: 1. Ask me 5 questions that test whether I actually understood different modes of working with an AI assistant. Do not reveal the answers yet. 2. After I answer, tell me which parts I got right, which I got wrong, and which I only half-understand. Explain only what I misunderstood — do not re-teach what I already know. 3. Give me one practical challenge based on something I could genuinely encounter at work or in daily life. Do not solve it for me. 4. Evaluate my solution the way an experienced person would judge it, including what a professional would have done differently. 5. Tell me what to learn next, and why that comes next. 6. Give me trustworthy sources for deeper study — prefer official documentation, primary research or standards bodies over blogs and videos. Rules for you: no buzzwords. No motivational filler. Say "I'm not certain" when you are not certain, and tell me which parts of your answer I should verify myself. Clearly separate facts from your recommendations and your opinions.
Become independent at this
Use this when you want a path from where you are to actually good, with checkpoints you can test yourself against.
I want to become independently capable at using AI as a critic and a rehearsal partner — not permanently dependent on AI, tutorials or step-by-step guides. Design a progression for me with five stages: Beginner, Guided practice, Independent practice, Real-world application, Professional level. For each stage tell me: - what I must know - what I must be able to do without help - the mistakes people make at this stage - one practical challenge - one real project that would prove I reached this stage - one way I can test myself honestly Then tell me the signals that I am ready to move to the next stage, and the signals that I have skipped ahead too early. Keep the theory to the minimum I actually need. Focus on ability I can transfer to situations you and I have not discussed.