A leadership team meets to settle an AI strategy. Three vendors in the running, a license budget to vote, an eighteen-month roadmap. While the meeting drags on, in the offices below, employees open a ChatGPT, Claude or Gemini tab on a free account or one they pay for themselves. The roadmap arrives late on a terrain the leadership has never looked at. And that, exactly, is the subject.

BYOAI, *Bring Your Own AI*, is not just another keyword in the management vocabulary. It is a historical signal that organizations are missing because they file it under the wrong category. Shadow AI, they think. Administrative workaround, like the personal smartphone brought into the office fifteen years ago. A slightly more modern flavor of peripheral disorder. The diagnosis is wrong. And it is wrong in a way that matters, because it makes an anomaly out of what is in fact a turning point. This essay sets the diagnosis straight. BYOAI is not a faster Shadow IT. It is the inversion of a vector that has held for two hundred years, the moment when, for the first time, it is no longer the organization that brings the machine to the worker, but the worker who brings the machine to the organization. ## The line, the machine, the engineer

We've seen this story before. We're seeing it now in reverse.

Industrialization has a history we know, because it played out in factories that could be filmed, on assembly lines you could photograph. Let's run through it in order, because the structure is what will illuminate what's happening today in the offices. *First act, the artisan becomes the worker.* Before the line, there was the workshop. The blacksmith who picked his own iron, the carpenter who drew the cabinet before cutting it, the seamstress who fitted the cut to the body. The gesture had meaning, because the person who performed it had thought the object, seen where it was going, chosen the material. The work mobilized the intelligence of the body as much as that of the mind, and it gave back to whoever made it a visible pride, because the finished object carried a signature. Then comes the line. The worker enters the cadence, executes a gesture broken down into fractions of a second, twenty times a minute, a thousand times a day. Georges Friedmann, the sociologist whose foundational work appeared in 1956, gave that moment a name still in use: *the work in pieces*^[1]. The meaning of the gesture has left the gesture itself. The worker bolts on a door, but he hasn't designed the door, doesn't know where it's going, won't decide when it's pulled from the market. Skill lives in the line, not in the worker. This passage from craft to line is not a neutral improvement of working conditions. It is the invention of a modern form of alienation, whose psychological, physical and social consequences will be documented for the rest of the century. *Second act, the machine replaces the gesture.* Robots take over from the welder, the assembler, the inspector. Tesla's Shanghai plant runs its assembly line at close to 95% automation^[2]. The worker leaves the line, or stays as a supervisor. Harry Braverman, in 1974, theorized this movement as a progressive *deskilling* of manual labor, where expertise migrates from the body of the worker to the software of the machine^[3]. What was know-how becomes machine code. The worker doesn't just lose a job, he loses cognitive grip on what he was doing. *Third act, the work splits in two.* At the top, engineers are hired to design and run the machines. At the bottom, low-paid hands are hired to clean the sensors, unjam the robotic arms, watch for breakdowns. The human hasn't been replaced, the human has been split in two: thinking moves up to the engineers, maintenance moves down to deskilled operators who serve the machine rather than the other way around. Cognitive power isn't redistributed: it is captured higher in the pyramid, by technical functions the organization chooses, hires, pays, retains. The gesture has left the worker, intelligence has left the workshop, control stays with management. The structure of industrial power comes out of the automation episode intact. ### Technological power has always been the organization's affair. Never the worker's. This logic has left the factories and settled in the offices. Marie-Anne Dujarier, a sociologist of work at the CNAM, documented in 2015 what she calls *disembodied management*^[4]. She describes a category of managers she names *planneurs*, who would now make up around 40% of French middle management. These managers don't direct concrete work. They design abstract devices, at distance, that tell others what to do and how. Dashboards, key performance indicators, standardized processes, lean methods, integrated information systems. The office has been brought down to the cadence. And look at what you produce there. The same form as yesterday. The same service contract, three variables apart. The same job posting declined into five segments. The same meeting minutes, whose format has been frozen since the last quality audit. The same LinkedIn post, formatted with hooks, bullet points and hashtags, because that's the format that performs. You don't write anything unique anymore. You fill in. You decline. You vary. You have industrialized yourself, all by yourself, without anyone needing to put you on the line. And no one reads what you produce, because everyone produces the same thing, at the same cadence, and singularity has dissolved in the variants. White-collar work has been industrialized in turn, by the same mechanisms that industrialized the line, with the same consequence: the meaning of the gesture leaves the gesture, skill migrates from the individual to the procedure, and the body follows. Burn-out. Brown-out. Disengagement. Occupational illness as a clinical category whose growth is documented in every workplace health survey of the last thirty years. The sociology of work has kept the record for a long time. The white-collar sector no longer reads its reports because it is too busy filling them in. Four acts, then, over two centuries. Industrialization, automation, recomposition by the engineer, industrialization of the white-collar sector. A single rule holds from start to finish: **technological power belongs to the organization, which distributes it to workers as it sees fit**. This is where the fifth act arrives. And this is where the rule breaks. Because the fifth act isn't delivered by the organization. It is brought by the employee. The white-collar robot, this LLM that the worker calls up from a personal phone or a browser tab, was not bought by the employer, not deployed by IT, not vetted by compliance, not budgeted by finance. It came from below. And the worker who brings it doesn't leave the line through layoffs. He leaves it through appropriation. He crosses, in a single move, the border between executor and engineer, without being hired, without a degree, without a mandate. He becomes the engineer of his own line. This is the inversion that had never happened in two hundred years of industrial history. ## The flip, what the numbers say and don't say

Three years, and the balance of power has turned. But not the way you think.

The data converges, and it is recent. Microsoft and LinkedIn published their annual *Work Trend Index* in May 2024, surveying 31,000 respondents across 31 countries. The central finding fits in one number: 78% of AI users at work bring their own tools, on their own accounts, outside any perimeter approved by their employer^[5]. That share rises to 80% in companies with fewer than five hundred employees. Beyond the number, the report names the phenomenon explicitly, *BYOAI*, and describes it as a turning point that leaders are largely blind to: 79% of executives say the urgency is real, 60% admit they have no plan. A different study, run by economists Aaron Chatterji, Tom Cunningham and David Deming for the NBER in September 2025, on 1.1 million ChatGPT messages, opens another angle^[6]. Roughly 70% of conversations are non-work-related. Personal use is the antechamber of professional use, and the direction of the transfer is plain: people discover the tool at home, then bring it to the office. This interdependence is not a technological novelty, it is what the information systems literature has called, since Aurélie Leclercq-Vandelannoitte's work, the *reversed IT adoption logic*^[7]. The precedent is worth naming, because it lets us measure what is changing this time. Google. In the early 2000s, the search engine arrived from below. People discovered it at home, to look up a definition, then check a recipe, find a movie theater, and help their kids with homework. And they brought it to the office, to replace the corporate intranet search that never returned anything useful. IT departments accepted, reluctantly. Then Google became invisible infrastructure. The same mechanism of transfer, twenty-five years apart. But what gets transferred has changed. Google gave us access to *information*. Generative AI gives us access to *production*. We are no longer bringing back a search skill, we are bringing back a productive capability that changes what we can produce on our own. This is not another step in the same story. It is a qualitative break. A third source confirms the behavioral resilience of the phenomenon. A Software AG survey conducted in 2024 on 6,000 knowledge workers in Europe and the United States shows that nearly half of employees: 48%, would keep using their personal AI tools even if their employer banned them outright^[8]. The same proportion freely shares sensitive information with unsanctioned tools. The balance of power is not ideological. It is practical. Usefulness wins over compliance. The threat of sanction is weaker than the perceived gain.

78%

the share of AI users at work who bring their own tools, outside the perimeter validated by their employer. Microsoft & LinkedIn, *Work Trend Index*, May 2024, on 31,000 respondents in 31 countries.

That's the raw picture. Now, the nuance that saves this essay from caricature. We are not at 100% of employees fully automating their work. The shift is gradual, in motion, with a leading edge and a long tail. *Developers are first in line*, because code is, of all the artifacts of office work, the one LLMs produce with the least friction. The *Stack Overflow Developer Survey* tracks this progression year over year: 76% of professional developers used or planned to use AI tools in 2024, 84% in 2025, of which 51% daily^[9]. A controlled study by Microsoft and Princeton, led by Zheyuan Cui's team, measured productivity gains of +12% to +22% in pull requests per week across nearly 2,000 engineers^[10]. Other professions are following at variable speeds, each one crossing the threshold at its own pace. Accountants, lawyers, communications people, salespeople, HR, analysts, each one is moving through the cycle of the meta-skill at its own friction and its own triggers. It is a deep movement, not a tidal wave. The stages, thresholds and bifurcations of that progression are described elsewhere: [The five levels of AI adoption](/en/pensees/five-levels-ai-adoption) holds the frame, that is not the subject here. The subject here is the *direction* of the movement, not its scale. And the direction is unambiguous: the white-collar robot arrives from below, profession by profession, at varying speeds, but in a single direction. That is what the numbers say. And that is what organizations don't read, because they look in the statistics for a wave of collective productivity when they are watching a silent redistribution of technological power. That redistribution becomes visible the moment you know where to look. Here is one case, worth keeping in mind throughout the essay because it lights up every step. A friend, in a large French media group. His leadership has given him the official tools they consider necessary. He uses them. But he also has others, his own, that he pays for on a personal account. He has built the small pieces of software he needed to run his projects. He has run, in a single night, an agent that produced three thousand six hundred articles covering the results of the latest municipal elections in every major French city. He has automated the production of every one of his executive committee presentations. And when a colleague comes to him with a business problem, he develops the tool that solves it himself, because going through the dedicated service of his group would take six months just to validate the spec of what he produces in an hour. He does, on his own, sometimes, the work of dozens or even hundreds of people. In silence. Without bragging, because the group's administrative layer, in his own words, "doesn't get any of it" anyway. We will come back to him, because his singular story says what the numbers do not say on their own: the *nature* of the inversion. ## BYOAI is not BYOD, and it is not Shadow IT

We've seen people bring in their phones. We've never seen them bring in their augmented brains.

The distinction matters, because it conditions how the phenomenon is read. An organization that files BYOAI in the same drawer as BYOD will not make the right calls, will not ask the right questions, will not see the right risks. Let's take the three terms in order. **BYOD** (*Bring Your Own Device*), which appeared in the early 2010s, refers to the practice of employees bringing in their phones, tablets, sometimes their own laptops for professional use. Aurélie Leclercq-Vandelannoitte analyzed it as the first manifestation of a broader phenomenon she names the *reversed IT adoption logic*^[7]. Where IT used to choose, buy and impose the tool, the employee now arrives with their own. But the break stops at the hardware. The nature of the work asked of the employee doesn't change: read your email, check your CRM, join a video call. The organization accepts because it saves on infrastructure and gains extended availability. It is, fundamentally, an *administrative delegation*. The tool migrates from fixed capital to the employee, cognitive power stays with the organization. **Shadow IT**, theorized by Mario Silic and Andrea Back in 2014, by Steffi Haag and Andreas Eckhardt in 2017, refers to the use by employees of software tools not sanctioned by IT^[11]. Dropbox instead of the internal server, WhatsApp instead of Teams, a shared spreadsheet instead of the ERP. The mechanism is a *technical workaround*. The employee wants to do the work that has been asked, and picks the fastest, easiest, most accessible tool, even when it isn't approved. The motivation is ergonomic. The work being asked stays the same. **BYOAI** is of a different nature. The employee is not bringing a device, the employee is bringing a productive capability. They are not picking a tool to do faster what they have been asked to do, they are importing a resource that changes what can be asked of them. The break is not material, not ergonomic, it is *cognitive*. The LLM they plug into their inbox writes notes, summarizes files, produces analyses, translates, codes, proposes, rewrites. It modifies the perimeter of what the employee can produce on their own. And it modifies, by ricochet, the very definition of the role. ### BYOD brought a device. BYOAI brings a power. This distinction is not academic. It has immediate consequences. An organization that treats BYOAI as a fancy BYOD will try to regulate it through usage charters, security policies, tool audits. These devices miss the target, because they treat hardware and ergonomics when the subject is cognitive power. An organization that treats BYOAI as Shadow IT will try to bring it back into the system, validate the right tools, ban the others. This effort is useful but insufficient, because it misses what the employee *does* with the tool once it is plugged in. The real question is not *which tool is my employee using*. It is *what work are they now producing, and to whom does that work belong*. Our friend at the media group answers that question every day, without ever putting it in words. When he develops the tool the dedicated service would take more than six months to deliver, he is not bypassing a Shadow IT process, he is not doing BYOD with a bit of AI on the side: he is repatriating cognitive power whose formal ownership is still unclear, and he is using it to redraw his own role, in silence. ## Reclaiming work time

We sold our time for two hundred years. The contract no longer holds.

The story of our friend at the media group, opened earlier, is not an anomaly of profile. It is the living illustration of four relationships that flip simultaneously when BYOAI takes hold. Four lines of tension that hadn't moved since the invention of modern wage labor, and that generative AI, almost silently, is recomposing. Each of them needs to be named in turn, because their articulation is where the strictly technical diagnosis fails to look. ### Time

You sold your time. You're discovering it has started to be worth more.

The scarcity of human time is the implicit foundation of the work contract for two centuries. The employee sells their time, the organization buys it at the agreed cadence. When AI does in ten minutes what used to take two hours, the surplus of one hour fifty has no designated owner in the contract, because the contract never anticipated that an employee's productivity could vary by an order of magnitude through their own initiative. BYOAI, in its simplest form, is the tacit choice of the employee not to return that surplus. They keep some of the gained time for themselves: to breathe, to learn, to plan, or simply to set the rhythm that suits them. The line between sold time and personal time is being redrawn without an addendum. Our friend at the media group has reclaimed his work time in entire blocks: three thousand six hundred articles produced in a single night, while he slept, is the equivalent of months of human writing compressed into eight hours of sleep. His leadership doesn't return that surplus to him in the form of free time, because his leadership doesn't see it pass. The value he creates while he sleeps will never come back to him, and he experiences it, rightly so, as an injustice. ### Skill

You don't sell what you can do anymore. You sell what you can have done.

Value slips from *knowing how to do* to *knowing how to have it done*. Orchestration takes the place of execution. The employee no longer sells only the skill they spent years acquiring, they sell the capacity to summon, on demand, skills they have never internalized and that they call up through the machine. This shift changes what we mean by the value of a role, because it changes what we expect from whoever holds it. Our friend is not a developer by training. Yet he codes, because he knows how to ask the machine to code for him, and how to recognize when the output is right or wrong. He hasn't learned to design slide decks, he has learned to formulate his presentation needs. He doesn't possess each of the skills his deliveries mobilize. He possesses the meta-skill that calls them all up. The breach is open, and [The five levels of AI adoption](/en/pensees/five-levels-ai-adoption) describes its stages in detail. ### Role

You don't redraw your role mentally anymore. You re-tool it in practice.

Amy Wrzesniewski and Jane Dutton, in 2001, named a long-overlooked professional behavior: *job crafting*^[12]. Their definition is precise, and it must be honored without distortion. Job crafting refers to the physical and cognitive changes individuals make in the task and relational boundaries of their work, modifying in the process the meaning and identity of their job. Three dimensions are at play: *task* (the activities), *relational* (the relationships), and *cognitive* (the perception). In its original formulation, job crafting is a bottom-up reorganization that does not change the technical nature of the work, it reorganizes its frame. BYOAI extends this mechanism by giving it a lever it never had before. The employee no longer just redraws the boundaries of their role mentally, they *remake them in practice*, because they now hold a productive capability that transforms the *task* dimension at depth. They don't reformulate their role, they re-tool it. Our friend at the media group did not say to himself, "I'm going to redefine my role to include software development." He just developed the tool his colleague had come to ask for. The role redrew itself through the gesture, not through the HR sheet. This articulation, *technologically armed job crafting*, is the joint that this essay sets between a classic literature of work and the generational phenomenon of AI. ### Organization

The company has lost its monopoly on automation. It doesn't know it yet.

This is the most destabilizing of the four relationships, because it touches the structure of technological power. For two centuries, it was the company that automated. Today, it is the individual who automates. And the company is discovering, with delay, that it has lost its monopoly on the call. The most visible form of the imbalance is the one that [The revolution that never happened](/en/pensees/the-revolution-that-never-happened) documents in depth: the asymmetry of power between equipped employees and a leadership still talking strategy. > BYOAI is not the migration of a tool, it is the reclaiming of work time the wage contract assumed it had locked. What these four relationships compose, taken together, is the *reclaiming of work time by the worker, through their own means of production*. The phrase is worth weighing, because it is not innocent. Through most of industrial history, the means of production belonged to the organization, and the worker sold their time to whoever held them. BYOAI scrambles that equation. The observer Matt Warren has coined a phrase for the phenomenon that now circulates in the debate: *capital formation at the edge of the company*^[13]. The worker accumulates, at the periphery of the company that employs them, a personal productive capital, a stack of augmented skills they rent in part to their employer, and keep in reserve for the rest. The systemic break of that migration is outside the scope of this essay. [Will capitalism survive AI?](/en/pensees/will-capitalism-survive-ai) documents the crisis of the three structural pillars (property, wage labor, value capture), and what comes out of it is not redeveloped here. But the joint must be named: what plays out in BYOAI, at the scale of an individual, is the micro-gesture that, multiplied by tens of millions, triggers the macro-fracture. ## What the organization sees, and what it does not

The organization watches productivity. It misses ownership.

What the organization sees is the tip of the iceberg. A diffuse productivity gain, hard to quantify but constantly noticed. A few subscriptions paid by employees on personal cards. A vague sense that things move a little faster, without quite knowing where. AI charters that struggle to keep up, because they describe a usage that no longer existed by the time they were signed. A strategic roadmap that talks about partners, platforms, enterprise licenses, and that lands on a terrain where the employees have already made their calls. What the organization doesn't see is the submerged base. A balance of power that has flipped below the waterline. A cognitive capital accumulating at the periphery, off the books, off the contract, off any pilot dashboard. A silent redistribution of technological power between equipped employees, those who learn fast, those who learn later, and those who don't equip themselves at all. A new asymmetry in the real hierarchy that the official org chart does not reflect and that management does not yet have the tools to measure. The friend in the media group has already done the math. He knows that his own augmented productivity, through the tools he has brought, exceeds by several orders of magnitude the value reflected on his payslip. It is not a claim, it is a measurement. And as long as his leadership cannot perform that measurement, the asymmetry plays in his favor alone. He keeps the call. The day he decides that he prefers to rent his capital elsewhere, or keep it for his own projects, his organization will discover the submerged base by the only means left to it: his departure.

Individual the employee who brings their AI doesn't just bypass a procedure, they accumulate a personal productive capital they can rent in part or withdraw The wage contract no longer describes the entirety of the relationship, because part of the produced value runs through a tool the company has neither bought nor configured.

System organizations that don't make a call on their relationship to BYOAI are not absorbing a passing disorder, they are silently ceding the monopoly on automation they had held for two centuries The structure of economic power is being redrawn, profession by profession, at variable speeds, but in a single downward direction.

Across three years of close observation of the turning point, through more than two hundred and fifty days of training and over twelve hundred professionals accompanied, the same pattern repeats in every organization visited. Leadership talks strategy while the teams have already made their calls. The terrain runs ahead of the plan, almost always. And the terrain will not roll back to wait for the strategy to catch up.

If the worker, for the first time in two hundred years, no longer needs the organization to get equipped, who still owns what they produce ?

Jean-Jérôme DANTON

#### Sources 1. Georges Friedmann, *[Le Travail en miettes](https://www.gallimard.fr/Catalogue/GALLIMARD/Idees/Le-travail-en-miettes)*, Gallimard, 1956. 2. On Tesla automation: « [Elon Musk Says Tesla's Production Lines Already Over 75 Percent Automated](https://www.torquenews.com/11826/elon-musk-says-tesla-s-production-lines-already-over-75-percent-automated) », *Torque News*, on the Fremont site, and « [Tesla highlights Giga Shanghai automation and production cycle time](https://www.teslarati.com/tesla-giga-shanghai-automation-production-cycle/) », *Teslarati*, reporting close to 95% automation at the Shanghai site. 3. Harry Braverman, *[Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century](https://monthlyreview.org/product/labor_and_monopoly_capital/)*, Monthly Review Press, 1974. 4. Marie-Anne Dujarier, *[Le Management désincarné. Enquête sur les nouveaux cadres du travail](https://www.editionsladecouverte.fr/le_management_desincarne-9782707178442)*, La Découverte, 2015. 5. Microsoft & LinkedIn, « [2024 Work Trend Index Annual Report: AI at Work Is Here. Now Comes the Hard Part](https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part) », May 8, 2024, on 31,000 respondents in 31 countries. 6. Aaron Chatterji, Tom Cunningham, David J. Deming et al., « [How People Use ChatGPT](https://www.nber.org/papers/w34255) », *NBER Working Paper* No. 34255, September 2025, on 1.1 million messages. 7. Aurélie Leclercq-Vandelannoitte, « [Managing BYOD: How do organizations incorporate user-driven IT innovations?](https://www.emerald.com/insight/content/doi/10.1108/ITP-11-2012-0129/full/html) », *Information Technology & People*, vol. 28, no. 1, 2015. 8. Software AG, « [Half of all employees are Shadow AI users, new study finds](https://newscenter.softwareag.com/en/news-stories/press-releases/2024/1022-half-of-all-employees-use-shadow-ai.html) », survey on 6,000 knowledge workers in the US, UK and Germany, September 2024. 9. Stack Overflow, « [2025 Developer Survey: AI section](https://survey.stackoverflow.co/2025/ai) », December 2025. 10. Zheyuan Kevin Cui, Mert Demirer, Sonia Jaffe et al., « [The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4945566) », *Management Science* (in press), 2025. 11. Mario Silic & Andrea Back, « [Shadow IT, A View from Behind the Curtain](https://doi.org/10.1016/j.cose.2014.06.007) », *Computers & Security*, vol. 45, 2014. Steffi Haag & Andreas Eckhardt, « [Shadow IT](https://doi.org/10.1007/s12599-017-0497-x) », *Business & Information Systems Engineering*, vol. 59, 2017. 12. Amy Wrzesniewski & Jane E. Dutton, « [Crafting a job: Revisioning employees as active crafters of their work](https://journals.aom.org/doi/abs/10.5465/amr.2001.4378011) », *Academy of Management Review*, vol. 26, no. 2, 2001. 13. Matt Warren, « [Bring Your Own Agent: The New Workplace Shift Nobody's Naming Yet](https://www.mattwarren.co/2026/04/bring-your-own-agent/) », April 2026.

The Audaces