Originally written in French. Translated by AI — the meaning has been preserved, not the prose.
There's a bad question people often ask about AI.
Is it going to decide for us?
I understand the question.
But in my daily use, that's not where the interesting point sits.
I don't ask AI to decide in my place.
I ask it to help me mature faster what I'm currently thinking.
That's not the same thing.
Judgment Rarely Arrives All at Once
In intellectual work, we sometimes imagine a decision as a clean moment.
You look at the evidence.
You call it.
You move on.
In reality, many decisions mature through small successive corrections.
An idea looks good in the morning. Then an objection appears. An example is missing. A group of users wasn't taken into account. A source contradicts the initial intuition. A sentence sounds good but hides a weakness. An assumption seemed solid, then you realize it rests mostly on an impression.
Judgment isn't only about choosing.
It's about recognizing what matters.
And that often takes time.
You reread. You walk. You talk it through. You sleep on it. You come back to the text two days later. A detail resurfaces. A question comes up. Something you hadn't seen becomes obvious.
That delay is part of the work.
But it also has a cost.
What AI Speeds Up
AI doesn't remove the maturation time.
It speeds up some of its stages.
It can surface very quickly what you would have eventually seen later:
- a blind spot;
- a contradiction;
- a fragile assumption;
- a vague definition;
- a missing example;
- an obvious objection;
- a missing source;
- an untreated consequence;
- a confusion between two adjacent subjects.
It isn't magic.
It isn't because AI "knows better".
It's because it can quickly test an idea against several frames: the goal of the work, the available sources, the decisions already made, the method rules, the text already written, the questions we tend to forget.
It provides a first resistance.
And that resistance is valuable.
An Objection Is Not a Decision
One distinction still has to be held very clearly.
When AI flags a gap, it doesn't rule.
It offers a point of attention.
Judging comes next, and that's on the human.
A blind spot may be real.
It may also be beside the point.
A contradiction may need to be resolved.
It may also be kept, because it says something true about the problem.
An objection may be important.
It may also come from reasoning that is too general, too cautious, too textbook.
The danger would be to treat every piece of AI feedback as a correction to apply.
That would mean swapping your judgment for compliance.
That's not what I'm after.
I want AI to increase the surface of confrontation.
Not to choose in my place what deserves to survive.
The Right Use: Challenge, Don't Conclude
In my work, the best questions I put to AI aren't always requests for production.
They're often requests for tension.
For example:
- what's missing?
- what moves too fast?
- which distinction isn't clear?
- what objection could a reader raise?
- which assumption stays implicit?
- which passage repeats an idea already covered elsewhere?
- what isn't proven enough?
- which part of the reasoning rests mostly on my intuition?
These questions change the relationship.
AI is no longer there just to write.
It's there to put the thinking to work.
It pushes. It tests. It pushes back. It asks for precision. It signals that the idea may work, but isn't stable enough yet.
It isn't pleasant in the comfortable sense.
But it's useful.
Why Context Changes the Quality of the Feedback
An AI in a chat window can already produce a list of blind spots.
Sometimes that's useful.
But it often stays generic.
It can tell you to define the audience, add an example, define the terms, qualify the claim.
That's good advice.
But it's advice you could give to almost any text.
With a context, the feedback becomes situated.
AI can compare what you've just written with what already exists in the file. It can see that an idea has already been covered in another section. It can spot that a validated decision isn't being respected. It can notice that an important source hasn't been used yet. It can flag that a passage contradicts a nuance you had kept elsewhere.
The feedback no longer comes only from a general competence in writing or analysis.
It comes from the meeting of:
- an intention;
- a body of material;
- a memory;
- a set of rules;
- the actual state of the work.
That's what makes it more useful.
AI Gives You Handles
There's one very concrete thing I appreciate in this way of working.
AI turns a vague impression into a handle on the work.
Before, I could sense a text wasn't good yet without knowing exactly why.
Something was missing.
But what?
An example? A transition? An objection? A definition? A consequence? A sentence that stays too abstract?
AI helps break that unease apart.
It can say: this passage repeats an idea already stated, this section arrives too early, this distinction isn't prepared enough, this promise isn't kept, this conclusion opens a subject the text never covered.
After that, I can decide.
I can fix it.
I can refuse.
I can keep the tension.
But I'm no longer facing a blurry feeling.
I have handles.
Judgment Becomes More Important, Not Less
The more useful AI becomes, the more human judgment matters.
That's counter-intuitive.
You might think AI shrinks the space for judgment, because it produces, analyzes, summarizes, rewrites and criticizes.
I think the opposite.
It increases the number of proposals, objections, variants and possible paths.
So there's more to judge.
You have to know:
- which objection deserves to be handled;
- which suggestion is too generic;
- which simplification betrays the idea;
- which nuance weighs the text down for nothing;
- which contradiction has to stay visible;
- which source is enough;
- which risk is acceptable;
- which trade-off genuinely commits the work.
AI speeds up the circulation of possibilities.
But it doesn't automatically hand you the right choice.
Judgment doesn't disappear.
It moves up a level.
What This Changes for a Product Manager
In product management, this difference is essential.
A PM doesn't only lack documents.
They often lack the time to test the signals against one another.
One piece of customer feedback looks important. Another contradicts it. A technical constraint changes the scope of the problem. A commercial opportunity pushes one way. A product reality pushes the other. An assumption looks appealing, but the evidence is thin.
The risk isn't only having too little information.
The risk is not testing the information enough before turning it into a decision.
In that setting, AI can help you slow down intellectually while speeding up the work.
The formula is odd, but it seems right to me.
It speeds up the gathering, the confrontation, the rewriting, the search for what's missing.
And because it does that fast, it leaves more room for the real slowdown: the one where judgment happens.
Do I believe this?
Does this objection count?
Is this decision robust?
Does this idea deserve to be pursued?
Are we solving the right problem?
Don't Confuse Speed With Delegation
The risk, obviously, is confusing acceleration with delegation.
Because AI is fast, you can be tempted to take its answer for a conclusion.
Because it phrases things well, you can believe the idea is clear.
Because it hands you a structured list, you can believe the problem is covered.
But good phrasing proves nothing.
A list that looks complete doesn't guarantee the right subjects are on it.
A well-written objection isn't necessarily relevant.
The human's role is precisely not to be hypnotized by form.
They have to keep responsibility for meaning, for business context, for the level of evidence, for the final decision.
AI can speed up maturation.
It can make thinking more inspectable.
It can show faster what's missing.
But it doesn't automatically turn an idea into a truth.
A Better Intellectual Solitude
There's one last, more personal aspect.
A large part of intellectual work happens alone.
Even inside a company, even with a team, even with meetings, there's always a moment where you have to think alone: write a note, prepare a decision, put an intuition into words, understand why something bothers you.
AI changes that solitude.
It doesn't remove it.
It makes it less mute.
You can drop off a still-imperfect idea and get friction back immediately. Not automatic validation. Not truth. Friction.
Something answers.
Something questions.
Something forces you to be precise.
It's in that friction that the idea matures.
And that, for me, is where AI becomes genuinely interesting.
Not when it replaces judgment.
When it gives judgment something to work on, sooner.