I am not married to AI.
I know because my wife pushes back and asks better questions. When AI gives me a soft, agreeable response that lets me keep doing what I wanted, that is a different kind of company.
That is the whole marriage comparison. The business question is what you want from the thing sitting beside your thinking.
Sometimes you need help carrying out a decision. Sometimes you need someone to notice that the decision is weak. Those are different jobs, and asking for a thinking partner does not tell AI which one you mean.
The plan you already like
Picture a hypothetical department leader changing the team’s weekly schedule. You want to move the preparation work to Thursday, with the next person picking it up Friday. That would leave the earlier part of the week available for other work.
You like this plan. You put it in front of AI and ask, “Can you help me with this?”
Help how?
Make the explanation clearer? Find a concern? Challenge the assumption that Thursday leaves enough time? Look for the ways the handoff could go wrong?
Our invented plan also leaves something unanswered: who passes the finished preparation to the person who needs it Friday. You have written down two days without connecting the work between them.
A polished version of the schedule could leave that gap exactly where it was. A list of ten objections could bury it under nine things that do not matter.
Here’s the thing. You keep asking for a thinking partner without saying when it should challenge the thinking.
The missing instruction is about the job you want its thinking to do.
Change the help with the work
In The Four D’s, Dialing is the stance where you stand beside the work and refine the recipe. The Chef tastes the soup and decides what needs adjusting. You do that for one dish, with its particular needs.
Apply the same discipline here. Look at the work in front of you before choosing how you want AI to help.
I think of four requests: carry it out, advise me, challenge the choice, and look for ways it could fail. They describe work you can ask for. They are not four product settings, and they are not another set of Four D’s.
You might use each request on the same plan at different points. You might need only one. More challenge is not automatically more useful.
Carry it out. Suppose the schedule has been discussed, the missing handoff has been settled, and you want a clear explanation for the team. Ask for the explanation. Give it the agreed arrangement and say that the choice of day is settled.
You still want a material problem flagged. But you do not need an argument about Thursday simply because you once told AI to challenge you. A person who reopens every finished conversation makes work harder. You do not need to recreate that experience in a chat window.
Advise me. Now move earlier, while you are preparing the schedule. You want help developing it and a note about anything that deserves attention. Ask for that combination.
In our example, the unanswered handoff belongs beside the proposed schedule. The response could say that the plan names when each part happens but leaves who passes the work unanswered. That is a useful concern attached to the actual work.
Challenge the choice. Move earlier again. Thursday is your preferred option, and you want to know whether your reason holds. Ask AI to examine the assumption behind it.
“I prefer Thursday because it frees up the earlier days. What would make that a poor choice for the person who needs the work Friday? Use what we know, and tell me what we still need to find out.”
Now you are inviting an examination of your choice. If the receiving person needs the preparation before Thursday, that would matter. Whether they do remains a question until you have the answer.
Look for ways it could fail. Before relying on the plan, you may want a deliberate search for weak points. Ask what could interrupt this particular handoff and why each possibility deserves attention.
Keep possible problems labeled as possible. “The work might be finished but never reach the next person” is worth examining when nobody has that job. “Your team will ignore the schedule” invents a fact about people the example has not supplied.
Same schedule. Different help.
An objection has to earn its place
It is tempting to judge the new response by how much less agreeable it sounds.
Finally. Some pushback.
But a forceful answer can be just as easy to accept without thinking as a flattering one. You have changed the tone of the reassurance: now you feel reassured that somebody is challenging you.
Read the reason.
Does the concern point to something in the plan? Does it explain why that thing matters? Does it distinguish a fact from a question you still need to answer?
In our schedule, the missing handoff is visible. The idea that the team will ignore the plan is an unsupported claim. They do not deserve equal weight because both arrived under a heading called “Risks.”
Right? A longer list does not make the thinking more independent.
You also need to leave room for agreement. Suppose the relevant people have settled the handoff, the receiving person confirms Friday works, and the preparation fits Thursday. AI may agree that the plan makes sense on those facts.
That is an acceptable answer to examine. You do not need to keep asking for a darker possibility until it produces one.
The point is better thinking. Let me tighten that. The point is help you can evaluate, whether it supports the plan, questions it, or identifies something you do not yet know.
Tell it when the choice has changed
There is another way to make a thinking partner exhausting: keep giving it yesterday’s job.
Yesterday you wanted the schedule challenged. Today you have heard the concerns, answered the relevant questions, and chosen Thursday. Now you need help writing the explanation.
Say that.
“The Thursday decision is settled. Help me explain the arrangement clearly. Flag a material inconsistency if you see one, but do not reopen the choice just to offer an alternative.”
You are changing the assignment because the work has moved. The request for challenge served a purpose. It does not have to become the assistant’s permanent personality.
And if a new fact appears that changes the choice, you can open it again. Settled means you have made a decision on the information available. It does not mean nobody may mention something that makes the plan fail.
Yeah. The leader still has to tell the difference.
Keep a useful request close to the work
The Professional Recipe calls the house’s way of working Training. A preference for honest, constructive disagreement can belong there. It is different from a hard stop about what someone may do.
Keep useful wording beside the kind of work it serves, and include it with the next task. For example: “If my reasoning holds, say so. If a weakness matters, name it and explain why. If a fact is missing, keep it a question.” Saving that sentence does not establish how a later answer will behave. Asking for advice does not give permission to act on it.
Your Monday Move
Choose one upcoming job you want AI to help with. Before you ask, finish this sentence: “For this job, I need you to…”
Carry out a settled choice. Advise while helping with the work. Challenge an open decision. Look for ways a particular plan could fail.
Name the help, then read the reasons that come back.
If you want a small check on wording you plan to reuse, try three short cases you understand and may share: one with a known weakness, one with an important unanswered point, and one settled task. Write what useful help would notice before looking at the answers.
An honest concern, a clear question, and straightforward execution can all be useful results. Three objections are not the target. Three responses also cannot promise how the next job will go.
Keep or revise the request based on why the answers helped. Stop using wording that keeps pulling the conversation away from the job.
Does that make sense? You are choosing the help before judging whether you got it.
Ask for the kind of help the decision needs. Judge the reasons that come back.
Original framework. Distilled from client work.
~ source material · Original framework. Distilled from client work.
