Back to Insights

The Prompting Habit

Prompting a model rewards clarity and iteration, and punishes ambiguity, relationship, and disagreement. Three practices and a five-question audit.

By Jerald Lee

In brief

The model never pushes back, so the management style you practise on it has no correction. That is the part worth watching.

Prompting a model rewards clarity, iteration, and specificity. It punishes ambiguity, relationship, and disagreement, because the model has no history with you and no reason to push back. Do that for six hours a day and you are practising a management style with the correction removed. The habits that make you effective with a model are close to the habits of a manager who gets things done through other people, but only the instruction half. The listening half gets no repetitions at all. That is the part worth watching.

# The Prompting Habit

Watch someone work with a model for ten minutes and you will see their management style with the patience removed.

Short instructions. A rewrite instead of a question. Visible irritation when the answer misses. The model does not mind. It tries again, at the same speed, in the same tone, as many times as needed.

That is exactly why the habit is worth noticing.

The most accommodating colleague you will ever have

The model does not need context it did not ask for. It does not resent being interrupted. It does not remember that you were short with it on Tuesday, and it will not adjust its behavior in a meeting next month because of it.

That is a useful property in a tool and a dangerous one in a habit.

Every other relationship in your working life has a cost attached to impatience. A person remembers being cut off. A team adjusts to a leader who rewrites everything, and the adjustment is usually quiet: fewer first drafts, fewer suggestions, fewer things said early. You do not get a complaint. You get a room that has learned to wait.

The model gives you no such feedback. It simply tries again. Which means the behavior you practise there never gets corrected, and practice without correction is how a style becomes a default.

What prompting rewards

Three things, and they are real skills.

Clarity. A vague instruction produces a vague answer, and the correction is immediate. You learn to say what you mean in one sentence, because the cost of not doing so is visible within seconds.

Iteration. You learn to treat a first attempt as a draft, not a verdict. You edit, redirect, try again without any social cost.

Specificity. You learn that "make it better" is not a request. "Make it shorter and remove the second example" is.

These are good habits. They are also, almost exactly, the habits of a manager who gets things done through other people. Which is why the transfer happens without anyone deciding to make it.

What prompting punishes

Three things, and they are the ones that matter with people.

Ambiguity. With a person, ambiguity is often where the useful conversation starts. "I am not sure what I want yet" is a legitimate opening, and saying it out loud invites someone into the problem. With a model, it produces a generic answer that you then have to correct. Do that often enough and you stop opening with uncertainty.

Relationship. The model has no history with you. It does not know that last week was hard, or that the person you are writing to is sensitive about the timeline. Every prompt starts from zero, which teaches you to over-specify and under-ask. That is efficient with a tool and corrosive with a team, because the whole point of a team is that it carries context you do not have to restate.

Disagreement. The model will not tell you the request is wrong. It will attempt it. A person will sometimes stop and say the thing you did not want to hear, and that is often the most valuable part of the exchange. Remove it often enough and you stop noticing it is gone.

The habit that forms is efficient and slightly cold. It works on a model. It wears on a team.

Signals it has already leaked

Four, and they are easy to check without asking anyone.

Your first move on a draft is to edit it. Not to read it and ask what they were trying to do. The edit is faster, and it feels like help.

You give instructions where a question would work. You have started to open with the outcome you want rather than the situation you are in. It is quicker. It also removes the part where someone tells you something you did not know.

Nobody has disagreed with you this week. That can mean you were right. It more often means the room has learned that disagreement does not change anything.

You explain less than you used to. You assume context that only you have, because with a model you have learned to supply exactly the context needed and nothing more. People do not work that way. They need the story, not just the task.

Two or more of these is not a crisis. It is a signal that the tool has been teaching you a style you would not have chosen on purpose.

Why the obvious objection is reasonable

Three objections come up, and all three deserve a real answer.

"It is just a tool. I know the difference." Probably true at the level of intent. The question is not what you believe. It is which behavior gets repetitions. Six hours of instruction and three minutes of listening is a training program, whatever you tell yourself about the difference.

"I am perfectly capable of switching modes." Most leaders are. The switch is not the problem. The problem is that the default mode gets practised and the other one does not, so under pressure you return to the one with repetitions behind it.

"My team would tell me." Some would. Most wait to see whether telling you changes anything, and if it does not, they stop. Silence is not evidence that the style is working.

Three practices that keep them separate

None of these are grand. They are small disciplines, and they work because they interrupt the reflex rather than trying to remove it.

Ask before you instruct. With a person, start with a question. What do you think is going on? What would you do? The answer tells you what they already know, which changes what you need to say. With a model, you can skip this. With a person, skipping it is the whole problem.

Wait before you rewrite. The instinct to take the draft and fix it yourself is strong, because you can. With a person, the same instinct removes the work that would have taught them something. Let the first version be theirs, then respond to it. The extra ten minutes is the training.

Keep one conversation a day that has no prompt. Not a meeting. A conversation where you do not know the answer and cannot generate one. This is the practice that keeps the other two honest, because it is the only one that cannot be done faster.

The Prompt Audit

Five questions, answered honestly, in about five minutes.

When someone brings you a draft, is your first move to edit it or to ask about it?

In the last week, did anyone disagree with you? If not, is that because you were right, or because you were not asking?

Do you give people the context you would give a model, or do you assume they already have it?

When a piece of work misses, do you rewrite it or explain what was missing?

When did you last change your mind because of something a person said?

What this is really about

Here is the part that does not get said often enough in conversations about AI and leadership.

The predictable work is going. The judgment work is staying, and judgment work happens between people.

A leader whose only practised mode is instruction will find the remaining work harder than it needs to be, because the remaining work is mostly asking, waiting, and being disagreed with. Those three cannot be delegated to a model, and they cannot be prompted into existence.

The model will never teach you them. It has no reason to. It is built to be the easiest relationship in your working life, and ease is not the same as practice.

The experiment

Pick one conversation this week and remove the instruction from it.

No outcome statement. No draft to fix. One question at the start, then silence long enough for the other person to fill it.

One conversation. One question. One silence. That is the whole exercise, and it will tell you more about your default than any audit.

Where this connects

The [Operator Trap](/operator-trap) describes what happens when a founder becomes the only person who can decide. The prompting habit is one of the ways that happens quietly. The leader gets faster at producing answers, and the team gets slower at forming them.

[Ownership Is a System](/ownership-system) is the other half. Ownership needs clarity, authority, and visibility. None of the three survive a leader who rewrites everything.

If you want a structured read on where your commercial decisions actually sit, the [Revenue System Audit](/revenue-system-audit) is the diagnostic we run.

Frequently asked questions

Is prompting a model really practice for managing people?

Partly. Both reward clarity and iteration. But a model has no relationship to protect and no reason to disagree, so it trains the instruction half of management and leaves the listening half untrained.

How do I know if my AI habit has changed how I lead?

Look at your first move when someone brings you work. If it is to edit rather than to ask, the habit has probably transferred. The five-question audit above is the short version.

Does this mean leaders should use AI less?

No. It means the tool should not be the only place you practise being direct. Use it for the speed. Keep one relationship in the day where you have to wait for an answer you could not have written yourself. ## A five-minute reflection Which conversation did you last end early because you already knew what you wanted to say? ---

Jerald Lee, Commercial Systems Advisor

Jerald Lee

Commercial Systems Advisor

Jerald Lee is a Hong Kong-based Commercial Systems Advisor, founder of Curiosity At Work and Operating Partner of The Growth Coach Hong Kong. He spent 13 years at Google APAC and helps founder-led companies build predictable revenue systems.

More essays from Curiosity At Work