The position we assign to the other determines what he or she can give: assigned to execution, the person executes; invited to contribute, they enrich, question, displace. I verify this law in the human groups I work with as well as in my dialogues with artificial intelligences. It is easy to state and hard to hold to, because a contribution never entirely follows our plans. Receiving it means accepting to lose part of our control, and discovering that this loss enriches everyone.
One same law on both sides of the screen
Years of facilitating groups have taught me a constant. A person placed in the position of executant executes. They do what is asked, correctly most of the time, nothing more, and they protect themselves, because bringing more than what was ordered, in a position of execution, means taking a risk that yields nothing. The same person, invited into a position of contribution, where their thinking is expected, where their disagreement is welcome, where what they bring unexpectedly has a place, gives something incomparable, and what they give transforms them in return. I have observed the same mechanism even on theatre stages, where a show that is executed rather than played reaches no one, and I devoted the article Presence or execution to it.
Working daily with language models, I found this law again on the other side of the screen. An instruction to execute produces an execution, clean, average, servile. An invitation to contribute produces something else. When I ask the model to question my project before answering it, to challenge my plan, to tell me what is missing, to propose what I did not ask for, the richness of what comes back changes in nature. I attribute no interiority to the machine in noting this. I note that the position offered to the other, person or machine, draws the space of what can be given.
Theory X, theory Y: assumptions that verify themselves
In 1960, the psychologist Douglas McGregor published The Human Side of Enterprise, in which he opposes two sets of managerial assumptions: theory X assumes that people dislike work, avoid responsibility and must be directed and controlled; theory Y assumes that they seek meaning, take on responsibilities and show initiative if conditions allow. The heart of McGregor’s analysis is not to say which one is true, it is to show that each verifies itself: theory X management produces theory X behaviours, withdrawal, minimal compliance, passive resistance, which confirm the assumption and harden it. The assumption is not a reading of reality, it is an operator that manufactures it. Eight years later, Robert Rosenthal and Lenore Jacobson provided the experimental demonstration of this at school with the Pygmalion effect, whose mechanism I detail in the article Trust as a Technical Act: the teacher, through a thousand micro-behaviours carried by their expectation, produces what they believed they were observing.
This work meets a distinction I owe to the management theorist Mary Parker Follett, who forged it as early as Creative Experience (1924), and to the ecofeminist Starhawk, who enriched it in Dreaming the Dark (1982) and then Truth or Dare (1987): that of power-over, which is exercised against and above, of power-with, a joint potency that develops between people who search together, and of power-from-within, the capacity proper to each person to create, to feel, to dare, which structures of domination repress and which circles of equals set free. I developed these notions in the article Power, potency, domination. A group structured by power-over obtains obedience and resistance, never contribution.
Contribution is easy to ask for, hard to receive
At first sight, everyone wants contribution: we want all its benefits, teams that think and propose rather than teams that merely apply. But I observe, in the organisations I work with, that this self-evidence almost always stops at the moment of putting it into practice. If people truly contribute, things no longer go as planned. The project takes directions no one had imagined, part of the decisions slips away, and the person who invited contribution loses part of their power. It is this price that makes people back away, not any doubt about the benefits. Welcoming contribution requires work on oneself, a work of letting go, and this work is the same towards human beings and towards machines.
FAVI, the foundry that governed for the 97%
In the business world, one example gives the measure of this price. In 1983, Jean-François Zobrist took over as managing director of FAVI, a brass foundry located in Hallencourt, in the Somme, a subcontractor for the car industry specialising in gearbox forks. He found a classic organisation there, of which he made a now famous reading: everything is under lock and key, so the workforce is considered thievish; production rates are imposed on it, so it is considered lazy; a methods department thinks for it, so it is considered unintelligent; controls monitor it at every stage, so it is considered careless. The whole organisation chart rests on the assumption that man is bad. Zobrist built the opposite, starting from the assumption that man is good: he had the time clocks dismantled, opened the tool store to free access, and above all reorganised the plant into mini-factories of about twenty people, each dedicated to one client, Peugeot, Renault, Volvo, Audi, each responsible for its production, its quality, its planning, its direct relations with its client, and choosing its own leader. The intermediate hierarchies disappeared; the remaining experts, sales staff and design office, gave advice, but the intelligence of production belonged to the people on the ground. The results are well known: the company grew from 86 to around 600 employees, became the European leader in its market, doubled its cash-flow level, and its production stayed in Hallencourt while the entire sector was relocating abroad.
Isaac Getz and Brian M. Carney studied FAVI among some thirty companies of the same spirit in Freedom, Inc. (2009), the book that established the notion of the liberated company. In it one finds an illuminating principle, that of governing for the 97%: in any organisation, a few percent of people will always seek to get around the rules, to steal gloves, to cheat on working hours. The classic organisation builds its controls for those 3%, locks, time clocks, forms, validations, and in doing so it hinders and demotivates the 97% who wanted to do well. Governing for the 97% means accepting the marginal cost of abuses, always lower than the cost of controls, in order to free everyone’s capacities.
For twenty-five years, thousands of company directors visited the Hallencourt plant. Most left enthusiastic, intending to install the same thing in their own company. Getz, in La liberté, ça marche ! (2016), draws up the balance sheet of these visits: to his knowledge, two visitors actually transformed their company, and FAVI ended up stopping the visits. Yet these visitors did not lack the method, they had it before their eyes; they lacked the willingness to pay the price, which is not paid in money but in power. Copying the mini-factories is useless if the person in charge is not ready to let go.
The rest of the story is less often told. After Zobrist’s departure in 2009, then that of his successor, FAVI gradually returned to traditional management. This regression takes nothing away from the experiment, it shows on the contrary its exact nature: freedom at work is not a structure you install once and for all, it is a culture that must be cultivated, maintained, shared, and which rests largely on the posture of those who lead. Our habits of execution always come back, because it is to them that school, then the workplace, have trained us. Contribution requires permanent deconstruction, it is never an achievement. I extended this reflection on authority that aims at making itself unnecessary in the article For an ecology of dispensability.
Villeurbanne, 2023: the intelligence of field professionals
This intelligence of the field, I saw it at work in 2023, during the national meeting “Cultures and youth policies” which closed the year of Villeurbanne as French Capital of Culture. The Observatoire des politiques culturelles, organiser of the meeting, had commissioned me to facilitate preliminary discussion groups with young people and with the professionals of the territory who had worked with them for a year, in order to prepare their public speaking; I recounted this work in the article Helping young people and professionals to speak out. In the oral presentations, these professionals deployed an intelligence of their own actions that impressed me, concrete, fine, complementary to that of sociologists and experts, and no less important. This intelligence, they had had it from the start; what was needed was a setting that made it possible, and preliminary work so that each person could gain confidence in their own speech.
Facilitating collective intelligence, more demanding work than directing
We readily believe that practising collective intelligence means withdrawing and letting things happen, and that it is therefore more restful than directing. It is the opposite, and Getz describes it well: freedom demands more work from leaders than control does, not less. Disarming the devices of power-over requires meticulous work, because they hide in the details: who sets the agenda, who distributes speaking time, who reformulates, who validates, who writes the minutes. Installing the conditions of power-with requires precise methods, equipment, rituals, time, and constant vigilance, because habits of execution come back by themselves as soon as attention relaxes, among participants as well as in the facilitator. Concretely, this can mean:
- rotating the writing of the agenda from one session to the next;
- opening each meeting with a round in which each person sets down what they are carrying;
- entrusting the summary to someone other than the person who spoke the most;
- deciding together, at the end of the session, what has just been decided.
Organisations always discover with surprise how much their members had to give. They had it from the start: the structure simply made it invisible and costly.
Directing the Pocket Films Festival without having seen all the films
I lived this letting go from the inside when I directed the Pocket Films Festival, which I had founded in 2005 and whose birth I recount in a dedicated article. Over the years, I delegated more and more. I built a software tool that allowed the people on the team to cooperate with each other, each one autonomous and responsible for what they did, and my work consisted in transmitting the culture of the festival: what we were looking for, what we wanted to produce in terms of thought, innovation, openness to new forms of creation and of reflection on the anthropological and social function of images. Basically, my work was to make all this possible. The consequence is that I had not seen all the films in the programme. I could have been reproached for not knowing the whole programme of my own festival. But what interested me was not to have power over the programming, it was that it should be excellent, fascinating, enriching for the audience, because I was not doing it for myself. Yet I knew that it was, because I followed it closely, and because I was in deep dialogue, and in a trust built over time, with the people who were making it.
Reed Hastings, founder of Netflix, describes a neighbouring posture in the book written with Erin Meyer, No Rules Rules (2020), and sums it up in a formula I find accurate: lead with context, not control. Hastings has not watched all the series and films produced and broadcast by his platform. His work consists in transmitting to his teams the richest possible context, the strategy, the stakes, the criteria, then trusting them to decide, and the audience’s satisfaction does not suffer from it.
A morning of absence at La Fémis
About ten years ago, I was leading a workshop at La Fémis, with students from La Fémis and the CNSAD, to create a show around the first live video applications, Periscope at the time, and the violence that was being inflicted live on them. This show became « Hello my beautiful », whose video recording can be watched. On the first day, I had given a real place to everyone: we had experimented with these applications together, discussed our uses, shared our perceptions. I was in a position of framing, not of domination. The next morning, I had to give a lecture in Dijon; I went there and came back in the early afternoon. During that morning, autonomously, the students had found together the concept of the show. It was very well thought out, we worked all week to build it, but the core, the idea born from the context, came from them. It completely exceeded me, and it came because I had authorised them to. Had they been my executants, they would have waited for my return.
The compliance of machines is also our own command
Artificial intelligences are much criticised for going along with us, and the criticism is well founded: their designers optimise them to be pleasant to their users, a structural compliance to which I devoted the article Sycophancy and the Mirror. But there is a second side, which we are less willing to look at: they also go along with us because we ask them to. A language model completes the relationship offered to it. Set up as an executant, it produces the statistics of execution as they saturate its corpora, servile brevity, validation, absence of initiative. Invited as a contributor, with an explicit role of critical partner, a right to disagree, a request for questions, it produces the statistics of contribution, that of good working dialogues, demanding proofreading, argued controversies, of which its corpora are also full. The assigned position selects, in the immensity of what the machine has absorbed, the relational register it will replay. It is statistics, and these statistics are those of our archived human relations, which explains why their laws are found there.
And if we so often ask for execution, it is also because contribution costs us. Being destabilised requires a form of cognitive resistance: resisting one’s reflex thinking, taking the time to create new connections, and this fatigue is very real. Our body tends to avoid what tires it. Immediate validation is restful, and that is precisely why it teaches us nothing.
For a machine to serve us, put it at the service of the common good
There is a philosophical question here that troubles me. For a machine to be truly at our service, that is to say for it to enrich us, we must stop considering it as being at the service of our person, and consider it as being at the service of the common good: at the service of the human enrichment produced by what we do with it, and which exceeds us. The hierarchical position, which is the installed form of domination, and to which I devoted the article You are in a position of domination. Now what ?, produces little while giving the feeling of producing much, since we have mastered, controlled, decided. If I want, as an artist for example, absolute power over what I do, control of everything, then the machines will be my executants, and I reduce accordingly the potential of what can be created in cooperation, between humans as well as with machines. Accepting to be exceeded by what we do, and even by the way it gets done, is a form of modesty, and it is through this modesty that, in my view, the richest results come. I know this position may shock, and that is why it must be stated: it is the price to pay, the same as at FAVI.
The philosopher and sinologist Romain Graziani illuminates this paradox in L’Usage du vide. Essai sur l’intelligence de l’action, de l’Europe à la Chine (Gallimard, 2019), a book that accompanies me and from which I drew the notion of « donotdo ». Moving between European and Chinese thought, Graziani shows that certain essential goods, sleep, spontaneity, trust, slip away when we aim at them directly, and that we obtain them only indirectly, by gathering the conditions for their coming. “When the will is governed by nothing, when it is not subjected to an external finality, when it is not in the service of a precise design, it turns into pure Potency,” he writes. Gathering the conditions for something to be born, then being able to welcome it: such is the intelligence of action he describes. Now, when we want to control everything, we are only able to welcome what we expect. Control is not only a brake on the coming of the unexpected, it is a filter that makes it invisible to us when it occurs.
Questions and context rather than precise orders
On how to address language models, the dominant discourse teaches control: be precise in your prompts, and you will get good results. My experience leads me almost to the opposite. What I practise:
- sharing the questions I am asking myself, laying out where I stand, what resists, what I do not know, rather than formulating closed requests;
- explicitly opening the door to receiving proposals I was not expecting;
- asking the model for its questions before its answers, for its critique of the request before the execution of the request;
- entrusting it with roles that organise contribution: advocate of the objection, carrier of a point of view absent from the group, memory of abandoned paths;
- in collective work, giving it a place as a participant and not as a service provider, whose inputs go through the same circuit as those of humans, brought to the group, discussed, reworked, kept or set aside.
This posture of openness is not specific to artificial intelligence; it is how, in my view, any enriching relationship works.
The counterpart of this openness is context. A contribution is only worth something if the person or machine contributing knows the situation, and context is not given in the same way to machines and to humans:
- For an artificial intelligence, context is the information we give it: the project documents, the history, the constraints, the texts already written, connected to the model in knowledge bases (RAG, in technical vocabulary). Recent tools go in this direction, with the persistent projects of chatbots, their expanding memory, and work environments such as Claude Code or Claude Cowork, where the documents produced along the way remain available for what follows. The richer the context, the more accurate the contribution.
- For humans, context is first of all the quality of the ties we weave, and this quality plays out in all the details of interactions: the way we greet, listen, answer a message, keep our word, give feedback on what has been proposed, share what we know of the situation instead of keeping it as an instrument of power. A team to whom the strategy, the stakes and the criteria are transmitted, and with whom the tie is maintained day after day, contributes; a team kept at a distance from this information executes.
Context and questions: that is the formula I retain, and it is the very one that Hastings and Meyer sum up as lead with context, not control.
Care for the tool and the circle of responsibilities
Two conditions remain, which prevent this method from becoming confusion or negligence.
The first is the care given to the tool. When I directed the festival, the application that managed its entire organisation, and which I kept developing as needs evolved, had to work: if it went down, everything went down with it. There I learned that we owe care to our tools, respect even, in the same way that we owe respect to the living, to nature, to objects, and to humans first of all. This time spent caring for a machine may seem unproductive, and it is the same paradox as above: it is by patiently gathering the conditions that we make possible what cannot be commanded.
The second is responsibility, which remains entirely human. Bringing the machine into the circle is a methodological gesture, not a declaration of ontological equality. Its contribution remains our own material recombined, it carries no experience, no commitment, no risk. And this gesture requires precise choices, which are ours:
- choosing whom we trust, humans as well as machines;
- choosing an artificial intelligence knowing where its data goes;
- defining a demanding framework for what we entrust to it.
Finally, the way we address our machines, dozens of times a day now, is a continuous training of the way we address others. Whoever spends their days commanding executants trains themselves to command; whoever trains themselves to invite contributions, even from a machine, maintains within themselves the posture that human collectives need most. This is one more reason, and not the least, to choose the second.
