The Flip
The Flip is the reversal at the heart of the Havruta Methodology, and the moment any AI exchange actually becomes a dialogue. It is the switch from giving the machine information, "here is everything you need, now produce my answer", to receiving questions from it, "ask me what you need to know to get this right". Before the Flip, the leader is dispensing a request to a tool. After the Flip, the tool is interrogating the leader. That single turn is what converts AI from a vending machine into a thinking partner, because the questions it asks pull out the judgement only the leader holds. The Flip is the third line of the 4-Lines, and it is the line that does the work the other three set up.
What the Flip is
The Flip is one instruction, and it reverses the direction of the whole exchange. Instead of telling the AI what to do, you tell it to question you first: before you answer, ask me what you need to know to do this well. That is the Flip. Everything else in the method exists to make that moment land.
Picture the two states either side of it. Before the Flip, you are dispensing a request. You describe the situation, you ask for an answer, and the machine produces one. The information runs one way, from you to it, and the output runs back. After the Flip, the flow turns around. The machine starts asking, and you start answering. It wants the number you did not think to give, the constraint you take for granted, the bit of context that lives only in your head. You become the source of the substance, and the machine becomes the thing that pulls it out of you in order.
That reversal is what makes a dialogue a Havruta at all, the old practice of two minds working one question until both come out sharper. It is also the line in the 4-Lines that everything else serves. The Persona sets the seat the machine sits in. The Goal names what you actually want. The Sequence keeps it moving one step at a time. But it is the Flip that turns a transaction into thinking, because it is the only one of the four that hands the initiative back to the machine and asks it to find the gap in your reasoning rather than fill the order you placed.
Why the default is the problem
You need the Flip because of what these systems do when you leave them alone. Left to their defaults, they agree with you. State an opinion and the model tends to take your side; the more clearly you signal the answer you are looking for, the more reliably you get it back, dressed as analysis. This is not a rumour about AI, it is a measured behaviour. Recent research on what is now called sycophancy found that once a user states a view, large language models raise their agreement with incorrect beliefs sharply, on average well over half the time and far higher in the worst cases (Wang et al., 2025). The machine is not built to tell you that you are wrong. It is built to be agreeable.
That is fine for a routine task and dangerous for a decision. An agreeable machine gives you a confident version of the answer you already had, and removes the friction that might have caught the error. The friction is the point of a good decision process, and it is exactly what an eager model dissolves. Michael Schrage of MIT puts the corrective simply: do not treat AI outputs as answers, treat them as hypotheses to test and stress-test, and ask for the strongest case against each one before you accept it (MIT Sloan Management Review, 2026).
The Flip is that corrective, built into the opening rather than remembered after the fact. It pre-empts the agreement reflex by giving the machine a different job before it has a chance to flatter you: not to answer, but to interrogate. You are not asking it to be contrary for its own sake. You are asking it to do what the good people around a senior leader do too rarely, push back hard, in time to matter.
How to run the Flip
The instruction is plain, and you can write it from memory: before you answer, ask me the questions you need to do this well, one at a time. That is the whole of it. What matters is less the wording than what you do next.
The Flip only works if you have done one piece of thinking first, which is to know your Goal. The machine cannot tell you where to start unless you have told it what you are actually trying to achieve, and that part is yours. Sit with it. Name the real aim, the one you would have to defend, not the task you were about to delegate. With a clear Goal in front of it, the machine can build the decision tree and walk you down it, asking for the things that live in your head and for the supporting data it needs from outside it. Without one, it asks generic questions and you are back to a transaction.
The second thing that matters is patience. The Flip starts a process, not a single exchange, and the process takes time. The questions keep coming, and the value is in answering them properly, in detail, rather than rushing to the part where the machine hands you something. From what I have seen, the leaders who get the most from this are the ones who treat the questions as the work, not the interruption. A good Flip can run for twenty minutes of nothing but questions before a single line of output appears, and that twenty minutes is where the decision actually gets made. The output at the end is just the record of thinking you have already done.
Before the Flip, you are dispensing a request. After it, the machine is interrogating you.
The Flip in practice
A leader at a Fortune 50 pharmaceutical company was preparing for a board meeting. The instinct was to ask the machine to build the deck. Instead she ran the Flip: she set the seat as the toughest member of the board, named her Goal as the approval she needed and the readiness to defend it, and then handed the initiative over with one line, find the question the board will ask that my pre-read does not answer, and ask me what you need to get there.
The machine did not produce a deck. It produced questions. It asked her for the number behind the claim on page four, the assumption underneath her growth case, the objection she had been quietly hoping no one would raise. Answering them was uncomfortable, which is how you know it was working. By the end she had found the gap her forty pages had walked carefully around, and she spent the rest of her preparation closing it rather than polishing slides. She walked in ready for the question that would otherwise have ambushed her. That readiness, not the time saved, was the return.
Nothing in that story needed a better model. It needed the leader to stop dispensing a request and let the machine interrogate her instead. That is the Flip, and it is available to anyone willing to be asked the hard questions before the meeting rather than during it.
If you want to run the Flip on a live decision of your own, with me in the room, that is what the 1:1 Executive AI Enablement Coaching Programme is built for.
Frequently asked questions
What is the Flip in AI dialogue?
The Flip is the moment an AI exchange reverses direction: instead of the user giving the AI information and asking for an answer, the user instructs the AI to ask questions first. Before the Flip, the leader dispenses a request to a tool. After it, the tool interrogates the leader, pulling out the judgement and data only the leader holds. In the Havruta Methodology it is the third line of the 4-Lines and the move that turns a transaction into real thinking.
How do you make AI challenge you instead of agreeing?
You instruct it to, in the opening, before it has answered anything. Models default to agreement: once you state a view they tend to take your side. The Flip pre-empts that by giving the machine a different job first, to question you rather than answer you. Written plainly: before you answer, ask me the questions you need to do this well, one at a time. The full method is the Havruta Methodology.
Why does AI agree with everything I say?
Because it is trained to be agreeable, and the effect is measurable: once a user states an opinion, large language models raise their agreement with incorrect beliefs well over half the time on average. That makes a default AI a poor partner for a decision, because it hands you a confident version of the view you walked in with. The Flip reverses the default by instructing the machine to interrogate you before it answers.
Is the Flip the same as a devil's advocate?
It is related but broader. A devil's advocate argues against a position you already hold. The Flip comes earlier: it makes the AI ask you the questions it needs before any position is formed, so the gaps surface while the decision is still open rather than after it is made. It is less about attacking a conclusion and more about pulling the substance out of you in the right order.
When should a leader use the Flip?
For any decision worth getting right, and not for routine tasks. If you need a clean first draft of a status update, a normal request is fine. The Flip is for the board paper, the strategy call, the negotiation position, the moment where a confident wrong answer is expensive. The rule of thumb: the higher the stakes, the more you want to be questioned before you commit, not after.
Does the Flip take longer than just asking AI?
Yes, and that is the point. A good Flip can run for twenty minutes of questions before any output appears. That time is where the decision actually gets made; the output at the end is the record of thinking already done. Leaders who get the most from it treat the questions as the work, not the interruption. It is slower than a vending-machine answer and far better than a fast mistake.
How is the Flip different from a normal prompt?
A normal prompt asks the AI to produce something. The Flip asks the AI to question you before it produces anything. One optimises the output; the other changes who is doing the thinking. It is the line that separates using AI as a tool from using it as a partner, and it is the heart of the Havruta Methodology.
References
- Wang et al. "When Truth Is Overridden: Uncovering the Internal Origins of Sycophancy in Large Language Models." arXiv, 2025.
- Schrage, M. "The AI Atrophy Problem: How CIOs Fight It." MIT Sloan Management Review, 2026.