The return on silence

There is a scene that repeats at the end of every session. A room that has just discovered what it can do with AI, faces still half in disbelief, and then someone who lingers, lets the others leave, and asks me under their breath the one question that matters: can I tell my boss. I see it at every engagement, on one side the person producing in the shadows and waiting to be noticed, on the other a leadership team that sees nothing, measures nothing, and wonders why AI shows no readable result in their house. Each blames the other's silence, and neither is wrong. What holds them there is not a misunderstanding, it is a mechanism, and that mechanism can be priced.

What nobody reports

Your people are not hiding anything from you. They have priced what speaking up costs.

The corridor question, I hear it dozens of times a year, but first it hides a scale worth measuring. In 2025, the University of Melbourne and KPMG surveyed 48,340 people across 47 countries. More than one employee in two admits to hiding their AI use and passing off machine-made work as their own; nearly one in two admits to using it against the rules of their own house.^[1] In France, the Julhiet Sterwen and Ifop barometer, on 1,003 employees of large firms, shows use jumping from 38% in 2024 to 62% in 2026, while only a third say they work under written rules.^[3] Use doubled. Its disclosure did not move.

I do not only read this gap in the surveys, I watch it in the room. Among the three hundred and fifty or so professionals I train each year, the question that comes back most is not a technical one.^[15] It is the question of the cost of disclosure: what will it cost me to say it. It comes from people who have just shown a whole room what they have never shown their own management.

The problem is not ignorance, it is accounting

Let us set aside the lazy explanation, the one invoked in almost every article on the subject: management understands nothing about AI. It is false, and the French numbers say so flatly. In the same barometer, 85% of managers use AI, against 44% of the people they manage.^[3] Managers are not behind on the tool. They are behind on something else: seeing what their teams do with it, and valuing it.

I have told elsewhere how this shift settled from below, with no permission and no rollout plan, in The revolution that never happened, and how the worker now brings their own machine, in BYOAI, when the worker brings the machine. Those two essays lay out the facts. This one raises the next question: why does generative-AI use stay invisible when everyone would gain from its being declared?

The prices of silence

Staying quiet is not a character trait. It is a calculation, and it pays.

Someone who creates value with AI and declares none of it is not betraying some discreet streak in their character. They pay, or avoid paying, a set of costs, and they pick the cheapest combination. Some are as old as working in a group: the fear of judgement, the fate reserved for new ideas, the reflex of hiding the time you save. Two others appeared four years ago. Born with generative AI, they reach wider than one employee's comfort: they touch the very way our organizations, and our schools before them, learned to recognise a skill and to value it. I take them in order, from the oldest to the newest.

The price of disclosure

Say you use AI. You will be taken for lazy.

At every training I open, the same line is waiting for me, delivered as an obvious truth nobody has ever proved: AI makes you dumb and lazy. It is anchored so deep that even the most advanced users hold it. They dread losing their skills, and they already judge the loss in the colleague beside them. All of it before they have grasped that the thing that decides is not the tool but how you use it. As long as that reckoning is missing, the one who uses AI wears the face of the cheat.

That suspicion, research has priced. In May 2025, three researchers at Duke's Fuqua School of Business published four preregistered experiments, on 4,439 participants, in the Proceedings of the National Academy of Sciences. The verdict is clean: whoever admits to working with AI comes across as lazier, less serious, more replaceable than someone using an ordinary tool.^[4] That same year, thirteen experiments gathered in Organizational Behavior and Human Decision Processes land on what their authors call the transparency dilemma: whoever discloses their use is trusted less than whoever hides it.^[5] And in spring 2026, an experiment on 961 knowledge workers put one identical piece of work up for judgement, shown sometimes with the AI mention and sometimes without. Whoever discloses their use loses twenty-four points of likelihood of being recommended for a project inside the evaluator's own organization.^[6]

Twenty-four points is the sticker price of honesty. Nobody pays it twice. And yet, at Val d'Isère, I watched that price collapse in front of me. I trained a whole team there, and the stigma lifted the day the entire team moved into the practice. A few weeks later, someone from an admin department dropped a line I have not forgotten: I feel less stigmatised for using it. He had been hiding, until then, and his level was so advanced that his management had ended up sensing that "it wasn't him doing it". His fear was founded: they had already spotted him, and already stripped him of his output, "it wasn't him". What freed him was not being found out, it was the whole team lining up where he already stood. The same study confirms it: where a company openly celebrates AI use, the penalty disappears, and those who own it even come across as more effective than those who keep quiet.^[6] The price does not follow from the nature of things, it follows from a culture.

The price of what nobody asked for

An organization does not recognise what it never ordered. This is not bad will.

An employee I worked with rebuilt, with AI, the accounting tool he needed, where the company's paid software served him badly. The result was striking, cut to his hand. But his management had no plans to switch, and neither did his colleagues, because for them the paid software did the job. He had solved his own problem, not theirs.

He received nothing, for two reasons that answer each other. His initiative had been framed by no one, and his result, however impressive, did not fit the shared need. With a little method and some listening to his colleagues, he would have outdone their vendor. That is the double wall of the novelty nobody asked for: an organization opens no slot for what it never ordered, and the initiative nobody framed ends up producing something that misses the many.

The mechanism is older than AI. As far back as 2012, three researchers showed in Psychological Science that, under uncertainty, people reject the new while swearing they want it, to the point of no longer being able to recognise it when it turns up.^[7] And the instrument, for its part, is missing: the consultancy WTW went through the 2026 pay filings of the S&P 500, and only 8% tie any part of executive variable pay to an AI measure.^[8] If managers are not paid on their use, nothing comes back down. Back in 1988, Dennis Organ was already defining organizational citizenship behaviour in terms that read today like a prophecy.^[12]

Individual behaviour that is discretionary, not directly or explicitly recognized by the formal reward system, and that in the aggregate promotes the effective functioning of the organization.

The employee automating in silence is waiting for recognition from a system whose very definition excludes what he produces. I showed in Your organization captures nothing from AI that this blindness comes from the architecture of organizations, not from their good faith.

The price of handed-back time

Hand back the time you save. You will be handed more to do.

Handed-back time is nothing new: it is the ratchet effect, known for years, you hide what you gain so as not to be handed more. The two genuinely new prices, getting-it-done and approval, come later. But AI drives this old reflex to a pitch no one has seen. A finance director I met in training had taken back 75% of her working time by running her spreadsheets with a conversational assistant. She kept it to herself, and her employer saw no gain in it. I told her story in AI's real payoff is not productivity. Her move, millions repeat: a 2026 American survey of 1,003 employees finds that 64% deliberately slow down so as not to finish too early, because finishing fast draws more expectations.^[9]

The reflex is old, but AI multiplies it. When a two-hour task settles in ten minutes, the freed time runs in only two directions, up or in place. Anthropic's survey of 81,000 of its users, in April 2026, says which one it takes: among those who name a clear beneficiary of their gain, most name themselves, and only one in ten reports their employer using it to ask for more.^[10]

Handing back time earns you more work. It is a price, and it falls on the one who hands it back.

The price of getting it done

You know how to get it done. Nobody taught you to make anything of it.

An employee was looking for a vendor for a three-dimensional preview of a sports facility. The quotes ran between five and seven thousand euros. He got it from a generative model, free, in two minutes. Here is a real skill, countable, whose market price is already known. And yet, in the room, he said it plainly, but nobody valued him for it. Everyone found it almost normal: after all, it wasn't him who made it, it was the machine. And he himself did not feel entitled to ask for anything.

This price is the most vertiginous, because it plays out first inside the head of the one who produces. In training, most do not feel they own what the machine produced for them. The idea of claiming anything for it does not even cross their mind. It is no longer "I can do it", it is "I can get it done", and nothing in their path prepared them to count that as a skill of their own. A study in Computers in Human Behavior in 2025, on two samples of 246 then 452 people, measures the blur: after a hand from AI, people misjudge their own share in the result, and overrate it widely.^[11] Between what the machine produced and what I brought, the line goes fuzzy, mine included.

This is the mechanism I described in Where there's a way, there's a will, an adaptive preference turned inside out: we stop claiming what we do not feel we own, just as we stop wanting what we believe we cannot get. The way out is not to sell yourself better, it is to change the unit. Not "recognise my talent", but "here is the service you no longer pay for, five to seven thousand euros for one image". The employee who replaces a firm, a freelancer or a piece of software books a saving, not a feat. Another door opens then, the one I push at in The five levels of AI adoption: which meta-skill separates the one who gets anything done from the one who gets the exceptional done.

The price nobody counts

Doing costs ten minutes. Permission costs three weeks. Do the arithmetic for them.

The prices before this one explain well enough why people keep quiet. They do not explain why the best users, the ones with nothing to fear, the settled and the well regarded, ask for nothing either. One last price is missing, and it is not social. It is arithmetic.

For two centuries, making cost time and money, deciding almost nothing. Designing a tool, setting up a line, building software: months, a budget, a team. In that world, standing an approval in front of every act made plain sense, because the procedure protected a heavy investment and cost, itself, only a few meetings. Every corporate scheme we know was set to that ratio. Generative AI cuts the first term to almost nothing and leaves the second whole. Building a useful tool no longer takes months, it takes an evening. Getting the right to use it still takes three weeks, sometimes a quarter.

Permission costs more than the act

A friend of mine, a marketing lead, coded a small module for an internal piece of software in ten minutes, at a colleague's request. Then my friend asked the colleague to say nothing, so it would not come back on him. Through the normal channel it would have meant a spec, a sign-off, a handover to development: six months. Ten minutes against six months, and the man who did the favour asks for silence out of caution. This is not insubordination, it is not even a choice, it is a calculation anyone runs in three seconds. Whoever can settle their problem before the end of the meeting where they should have raised it does not raise it. They settle it.

A procedure that costs more than what it protects is no longer protection. It is a toll.

This price is the only one that belongs to generative AI alone. No technology before it crushed build time this far while leaving decision time intact. The spreadsheet, email, the smartphone at work all demanded months of rollout, which gave the committees time to keep up. Here there is no delay left to fill. And an organization whose decision time stays longer than its members' execution time stops, mechanically, being consulted.

Why the old remedies do not take

Your schemes were built for a world where making took months.

We know serious schemes for surfacing employee initiative, and many have proved their worth: prize suggestion boxes, staff suggestion systems, negotiated gain sharing. But every theory of managing initiative falls flat the moment you have to steer output faster than the time it takes to decide. A file gets filed, a committee convenes, a financial year closes: they add latency to the one activity whose whole advantage rests on the absence of latency. They are not insufficient, they land out of time. I am keeping their examination for a coming essay, "Generative AI and management".

What survives that filter shares one thing, and it all comes down to this: these mechanisms step in after the act, so they slow nothing. Recognising a use once it has happened, sharing a gain once it is there, handing back part of the time already saved. It is, besides, in the air: in April 2026, OpenAI urged companies to pilot a thirty-two-hour week at full pay, under the name efficiency dividend.^[13] Out of that comes the rule that governs all the rest, the one line to keep if you keep just one. We are not moving from an uncontrolled system to a controlled one. We are moving from authorisation to recognition. The organization stops validating acts one by one, which it can no longer take on at the rate they surface. It sets a perimeter once, and it values results afterwards.

Making it pay

Nobody needs goodwill. Everybody needs an honest price.

What follows is not a charter, and above all not a call to transparency. You do not repair an asymmetry with morality. These are ten moves, which I run on engagements and whose effects I still measure, organization by organization, with a diagnostic tool built with my firm KiXiT.^[15] They ask no one to be better, they shift prices, and the order matters, because each move opens the next. One warning first: the dosage depends on what you are facing, the size, the legal status, the share of seasonal contracts, the state of the management line. What unlocks a twelve-person shop is not what unlocks a three-hundred-person company, and nobody picks that dosage without first measuring what circulates and what goes unsaid.

What the organization decides once

Decide a perimeter. You will not have to decide a thousand times.

First move, decide a perimeter rather than acts. It is decided once, and it frames a standard of work, a responsibility for what gets shared, and the acceptable bounds of initiative. As long as the rule is to ask, the rule will be bypassed, the way my friend bypassed the six months of the normal channel.

Second move, sign the no-penalty guarantee before the first question. A verbal promise weighs nothing against a penalty measured at twenty-four points, and employees know that better than their leadership, since they are the ones who pay it.

Third move, pay for the tool without demanding the result. Reimbursing the personal subscription, capped, with no usage audit, settles an entire asymmetry for the price of a pizza a month, and removes a team's most legitimate grievance, financing their employer's means of production out of their own pocket.

Fourth move, write the sharing rule before you know the gain. A rule decided afterward is no longer a contract, it is a favour, and everyone feels it within the second.

What managers do every week

A manager who does not use the tool penalises the one who does. Without meaning to, without knowing it.

Fifth move, practise before you evaluate. A manager who has never handled the tool reproduces, unknowingly, the gaze that penalises; at Val d'Isère, moving the whole team into the practice was enough to lift the stigma.

Sixth move, ask about the method as much as the result. As long as how the work is done does not enter ordinary conversation, it stays invisible, unwritten, untransferable, and so it is never valued. The question runs to five words: how did you get that result?

Seventh move, answer within the week, even to refuse. Silence after a disclosure kills reporting more surely than a refusal.

What the employee should prepare

Stop waiting. Recognition is not earned, it is prepared.

Eighth move, report a repeatable procedure, not a feat. A feat invites judgement, a method invites reuse, and reuse protects its author while it creates value for the organization, because it joins its productive capital.

Ninth move, aim for reuse rather than recognition. A practice picked up by five colleagues is a dated, verifiable, arguable fact, where recognition expected without disclosure cannot be defended.

Tenth move, negotiate at the moment of the gain, not at the annual review, the avoided quote in hand. Leverage lasts while the practice is visible and its withdrawal would show; twelve months later it has faded into the scenery, and the organization has absorbed it for nothing.

We do not pay for what we do not look at. We do not say what it costs to say. Between the two there is no misunderstanding, there is a valuation both parties can accept.

The trap of the individual gain

Keep it all. You keep its limits too.

Now for the uncomfortable part, and I address it to the reader who recognises themselves in the quiet employee, because nobody else will. Before the one who produces a great deal with AI and says nothing, three doors open. The first leads to independence, starting your own company, a real opportunity for a few, but it carries a risk most neither want nor can bear. Often, in fact, the best users suffer from the machine's lack of honest feedback, sink into endless builds, and come back the hard way to reality when they have to sell, alone and against everyone, the product built in one evening. That part, let us say it, does not compress: it rests on relational skills, and no AI will take it on for us. The second leads to the labour market: an analysis of more than a billion job postings across twenty-seven countries puts the wage premium on AI skills at 62% in 2026, up from 57% the year before.^[14] But that premium is earned by changing employer, not by staying: the market pays it through a new offer, and whoever keeps quiet in their post will never see it. The third door is to keep the time and say nothing.

Almost everyone pushes on the third, because it demands nothing, and it is the least profitable of the three. Time saved in silence does not add up, cannot be attested, cannot be handed on, cannot be argued anywhere. It buys comfort, which is not nothing, but comfort is something you get used to, and it does not shield you from its opposite, the boredom at work that carries its own pathologies.

Silence protects. It values nothing.

The market has already begun to decide for the undecided. In the survey on unauthorised use, three professionals in four say they would move to an employer that developed their AI skills better, four in five at the largest companies.^[2] That number does not alarm employees, it alarms leadership: the value they cannot spot, someone else is already buying. And to be fair, the clarity has to be turned back on the one who waits. Expecting recognition for a contribution you never declared is not discretion, it is a contract signed by one side only. Silence is rational, it remains a losing bet over time, and admitting it is not submission.

Jean-Jérôme DANTON