Reprenons ici le graphique qui semble le plus parlant dans ce papier.

Il montre l’océan qui sépare la performance dans les devoirs et les résultats d’examens. Il montre aussi le fossé qui se creuse dès les premiers mois d’adoption de l’IA, entre les élèves qui l’utilisent beaucoup et de manière régulière – en lui “délégant” le travail qu’ils devraient faire eux-mêmes pour assimiler et maîtriser la matière – et leurs camarades qui n’utilisent pas ou ces outils ou qui en font un usage raisonné.

Quelques leçons du papier:

1️⃣ L’IA améliore les notes de devoirs (+18 %) et réduit leur temps de réalisation (-30 %), mais fait chuter les résultats aux examens mensuels (-20 % en 6 mois) et aux concours (-18 % à -24 % après 2 ans).

2️⃣ Les pertes sont plus marquées en sciences sociales, puis en STEM et langues, touchant surtout les plus jeunes, les meilleurs élèves et les garçons.

3️⃣ Conséquences du schéma d’adoption: 80 % des utilisateurs d’IA externalisent leurs devoirs (temps très court + notes élevées), subissant les pertes les plus fortes. Ceux qui gardent un temps de travail sur la matière comparable à celui des non-utilisateurs limitent ces pertes.

Voir le papier en v2 – juste sorti du four. L’étude présente des limites qui invitent à une certaine prudence:

• les auteurs notent que cet effet diminue avec le temps : l’écart est passé de -25% en début 2023 à -16% en juin 2025, suggérant une adaptation progressive des élèves ou enseignants;

• l’étude provient d’un seul comté de Chine centrale (population > 1 million, PIB/habitant légèrement sous 6 000 USD), ce qui est un facteur de prudence supplémentaire pour la généralisation.

En somme, malgré les limitations, la leçon clé est de ne pas sous-traiter la compréhension des sujets à la machine, et donc de prendre le temps de construire sa propre compréhension et sa propre maîtrise des sujets sur lesquels on se forme.

On verra bien si ça se confirme et si les relectures de ce papier conduisent à une qualification ou une modification des conclusions, mais ça semble très intéressant et porteur d’enseignements dont il faut tenir compte. Surtout à l’heure où des pays comme la Finlande rétropédalent sur l’exploitation des outils numériques en classe.

D’ailleurs, petite parenthèse, la règle consistant à ne pas sous-traiter la compréhension des choses – donc la capacité de saisir et de se saisir du réel que l’on appelle aussi intelligence, au sens le plus proche du Latin – peut être étendue à d’autres domaines que l’éducation. Ainsi, dans les entreprises par exemple, si le développement et le déploiement d’applications métier est compris – justement – comme la transcription la plus exacte de la connaissance et des capacités organisationnelles, alors c’est certainement là qu’il ne faut pas parier la boutique sur un code écrit automatiquement par des machines et que personne ne comprend dans votre équipe.

De quoi méditer.

The ability to perform executive and board duties with the appropriate precision and diligence depends on the accuracy and sufficiency of their comprehension of the organisation. That comprehension is shaped by what they see in documents and reports about the capabilities of their company or organisation, but where the capabilities are best documented is not in documents. Documents are just a picture, a painting of reality – and one that depends on the viewpoints as well as on the painter –, not reality itself. Reality is in code. It is in the collection of systems used by the people in the business to run the processes of the business and how these systems are assembled together. Hence, relying on the painting instead of the reality is like driving your car based on a drawing of the road, without ever looking at the road or at other drivers. Simply put, it is dangerous and that becomes increasingly true as systems become more capable of supporting the intelligence behind the way of working of a business. And that’s precisely where what is perhaps the greatest risk of our times lies: the delegation of comprehension to technical or documentary artefacts which may not reflect reality… or the longer term intent of the leadership of a business.

Product managers, product owners, scrum masters, product marketers and other professionals operating at the frontier between the functional and the technical reality of their company know this very well. They all have felt, in some form or other, the pain of the divergence between what they describe or define in terms of needs and functionality and the reality of the tooling that is developed or implemented to perform those functions. There are multiple reasons that explain this divergence and several methods designed to tackle the problem through iterations, tight coupling between business and tech people, retrospectives, pair programming, demos, visibility of impediments… etc. We won’t go into those. There’s ample literature about that topic, but one thing that is quite striking is that most professionals assume that it is quite impossible to understand the code that defines the functionality designed to deliver a business capability. And that certainly looks like a reasonable assumption. Much like it looks rather reasonable for boards to run strategic and governance aspects of a legal entity, that is the container of business value owned by shareholders, based on the strict separation between what is operational and what is of “higher” level.

The time may have come to challenge some of these principles, because, in a digital economy with strong reliance on machine learning capabilities, Board and executive committees cannot afford to be ignorant about code and software that embodies the complete knowledge, expertise and competitive advantage of their business. They should neither indulge in, nor be allowed to relegate these aspects to mere technical implementation status that is not in their purview. Because ultimately they cannot be allowed to operate on the basis of paintings instead of reality. Documentation and reports are just inaccurate representations of specific ways of understanding the business. They are paintings.

Tudor Girba and Simon Wardley make the observation about software architecture diagrams: hand-drawn views of a system are not photographs of reality but “paintings that document the perspectives of the authors,” and in a world of continuous change they are almost guaranteed to be wrong the moment they are made. The observation is more general than the domain it was made in. The picture a leadership team holds of how its own firm works – which capabilities are strong, where value is created, how a decision actually propagates from intent to outcome – is itself such a painting. It is composed from memory, from reporting lines, from the slides that survived the last off-site. And like any painting, it ages, flatters its authors, and quietly diverges from the thing it depicts.

In a connected, tool-mediated economy, comprehension – of strategy and of systems alike – is the one capability a firm must keep under its own control, because every instrument that relieves you of understanding also encodes a decision you did not make and cannot see. Comprehension can be supported, accelerated, instrumented. It cannot be outsourced. The moment it is – to a consultant’s best practice, to a software package’s defaults, to a model’s confident output – the firm has surrendered not a task but its capacity to govern itself.

The gap leaders do not see

Start with what the firm actually is. Not the org chart, which is another painting, but the operating reality: the systems that hold the data and run the transactions, the people who exercise judgement inside and around those systems, and the integration layers – e.g. middleware and API platforms – and processes that bind the two together. That is where the organisation’s behaviour lives. And in 2026 almost none of it is innocent of code. The systems are code by definition. The processes are increasingly encoded – in the ERP configuration, the workflow engine, the approval automation, the routing rules no one has read in three years. Even the people are mediated by code, because the judgement they exercise is shaped by the dashboard they are shown, the field the form will and will not accept, the recommendation the tool surfaces first. To understand how the firm functions is, to a degree that keeps rising, to understand what its code is doing.

This is precisely the comprehension that leadership lacks, and it is a strategic exposure rather than a technical inconvenience. The documented case Girba and Wardley report makes the point with uncomfortable economy: a large corporation spent years and millions trying to improve its central data pipeline, working from a high-level architecture diagram that turned out to be fiction. The diagram omitted an entire third-party system that was processing the company’s data – a fact no one in the organisation knew. They were not optimising the wrong thing. They were optimising a thing that did not exist as drawn. The remedy, when it came, began not with another attempt to fix the pipeline but with a different question: what does the system actually look like? Built into instruments that read the live system rather than the painting of it, that question turned a problem that had defeated hundreds of person-years into one solved in two person-months.

Treat that as a parable, not a product reference. The point is not the tooling that closed the gap; the point is that the gap existed at all, invisibly, at the centre of the firm’s main marketing channel, while competent people made – what they truly believed were – confident decisions on top of it. The number attached to the recovery – a six-hundred-fold improvement – is a single documented case and should be read as one illustration, never as a generalisable yield. What generalises is the structure: leaders governing a representation that had drifted from the reality, with no cheap way to tell.

Decisions made on belief

Software engineering, Girba and Wardley argue, is at root a decision-making activity – every change, every fix, every architectural choice is a decision, and the quality of those decisions turns on how fast and how truthfully one can move from question to answer. The argument is correct, and it does not stop at the engineering door. Strategy is a decision-making activity in exactly the same sense, and it suffers from exactly the same pathology: the prevailing mode makes high-stakes decisions on belief rather than evidence. The board approves an investment on the strength of a representation. The executive commits to a transformation on the strength of a benchmark. Neither has a low-cost way to verify that the representation matches the firm, or that the benchmark matches the firm’s situation. This is the generalisation worth making explicit, because it relocates comprehension from a cost centre in IT to a variable that governs decision quality across the enterprise.

The evidence that this is costly is sturdier than the slogans usually quoted. Large-scale field research finds professional engineers spending on the order of 58% of their time simply comprehending the systems they work on (Xia et al., 2017) — understanding is not a preliminary to the work; for the majority of the time, understanding is the work. At the project level, the empirical distribution of IT cost overruns is heavy-tailed, with only about a third of large public projects finishing within budget (Jørgensen & Moløkken-Østvold, 2022). At the level of strategic change, recent systematic review puts the share of digital transformation initiatives that fully achieve their intended outcomes at roughly 5–30% (ASEJ, 2024), with data quality repeatedly implicated in the failures (Azeroual & Jha, 2021). The other culprit is failure in managing the transition from a discrete, disconnected and slow set of processes to a complex, networked and adaptive system that becomes a totally new work reality shaped more by the practices that leverage the capabilities of the tooling than merely by the tools themselves. That is precisely why isolated ERP, CRM, e-commerce, MES or WMS initiatives are categorically necessary but not sufficient to achieve the kind of profound morphing that a digital transformation brings to companies. Let alone help their people deal with fast continuous change thereafter. Of course, a factor making the situation worse lies in too many system integrators operating as sales agents of makers of software packages, focusing more on their benefits as partners of the software makers than on their clients’ needs, constraints, capabilities and interests.

Girba and Wardley name two metrics for the decision loop – time to answer (how long to answer a specific question about the system / the business) and time to question (how long to formulate a proper one). The useful generalisation is not the vocabulary but the dynamic it captures. When understanding is expensive, an organisation can afford only a few questions, often resorting to cheap proxies instead of hard answers, and governance degrades into hope – a small number of bets placed and prayed over. When understanding becomes cheap, the binding constraint moves: it is no longer the ability to find answers but the imagination to ask the right questions. That is the moment comprehension stops being a technical convenience and becomes a strategic asset, because competitive advantage shifts to whoever can interrogate their own reality faster than rivals can interrogate theirs.

How firms surrender comprehension

If comprehension is the asset, the failure mode is its quiet surrender, which comes into the business through three doors, each of which presents itself as help.

The consultant, and the best practice. A strategy configured to “best practice,” or benchmarked on what competitors are seen to do, is not a neutral shortcut. It is someone else’s strategy, encoded and inherited. Best practice is, by construction, the average of what worked elsewhere, under conditions that were not yours, solving problems you may not have. To adopt it without comprehending it is to import a set of decisions – about where to compete, how to price, what to standardise, what to treat as undifferentiated… – that you never actually made. The work the consultant did was real; the danger is in receiving the conclusion without the comprehension that would let you judge whether it fits, or notice when the ground it assumed has moved. The deliverable transfers; the understanding does not, unless you insist on rebuilding it.

The package, and the default. Enterprise software is where the systems-people-processes point bites hardest, because here the process becomes the configuration. When a firm adopts a packaged system and accepts its defaults, the defaults are not empty – they encode a model of how a business like yours is supposed to run, made by people optimising for the median customer and for the supportability of their own product. Over time, the configuration silently becomes the process: work flows where the system permits it to flow, exceptions are handled the way the screen allows, and the organisation’s actual operating model migrates to wherever the software is path-of-least-resistance. None of this is decided explicitly with the people affected and an authority in the business exerting discernment as to what to apply. It is decided in the implementation by the package consultants, and then it governs the firm. A package you cannot interrogate governs you on belief, not evidence – and what it has quietly decided on your behalf is invisible precisely because it now is “the way we work.”

The model, and the answer. Artificial intelligence is the most powerful instrument yet built for relieving people of the labour of understanding – or at least that’sƒ how it’s sold by AI labs and panicked technology giants of the previous wave of digitalisation. That is exactly what makes it the most powerful vector for this failure. The same capability that collapses the cost of an answer can, trusted blindly, deepen the belief-not-evidence problem rather than cure it, because it manufactures fluency. AI’s competence is frequently performative: an appearance of understanding without the substance, most dangerous on the deep problems where the distance between what you observe and what is actually true is greatest. The independent evidence on even narrow assistance is more modest than the vendor claims – controlled study of code-assistance tools finds gains around 35% on task completion (Shihab et al., 2025), against vendor figures of 50–55% — and the gap between those numbers is itself a small lesson in not accepting a confident answer on trust.

This is the mechanism, not merely the mood. Storey (2026) argues that generative AI redistributes the burden rather than removing it: it can lower the most visible cost – the technical debt in the code – while silently accelerating two less visible ones, the erosion of shared understanding she calls cognitive debt and the loss of captured rationale, of what a system is for, she calls intent debt. The translation to the firm is exact. Every instrument that relieves leadership of understanding lowers a cost that shows – friction, headcount, time-to-answer – and raises one that does not: the distance between the painting and the firm. The trade reads as efficiency precisely because the cost it incurs appears on no report.

The line that separates AI-as-aid from AI-as-substitute is not the model’s sophistication. It is whether its output is explainable and verifiable against the firm’s own ground truth, or accepted because it sounded right. An instrument that shows you how it reached an answer, in terms you can check against reality, augments comprehension. An instrument whose answer you take on faith replaces it – and replacement, here, is the failure. The question is not whether to use the tool. It is where, in the decision, it still matters to have a human who understands.

Partial comprehension and fast action

Past a point, demanding full comprehension before every move is analysis-paralysis dressed as prudence. More often than not we need to move without complete comprehension or maintain a degree of ambiguity to preserve optionality and the time-value dimensions of decisions, especially if the decision is non-reversible. Thus moving without complete comprehension is defensible when the following things are true at the moment of the move:

1. the decision-maker has a fair estimate of how much they actually understand – the degree of their own comprehension, including the likely size of what they cannot see;

2. there is an explicit, reckoned expectation of the outcome, weighted across the downside and not merely the case they hope for;

3. the range of consequences and outcomes does not include catastrophic impacts;

4. action caused by the decision is very likely to produce one or more lessons or discoveries that can be used to inform the next decision;

5. an explicit system or mechanism is in place to capture the lessons from the action undertaken, such that it can feed a process of discovery-driven transformation.

Consider the manager who force-releases an enterprise system before it is “ready.” Done in a way that meets the above criteria, the move can break an organisation’s healthy-looking resistance to change and create value that patience would have forfeited – tempo can be worth more than certainty, and the orientation that lets a leader act decisively inside a fast loop is, after Boyd, frequently the decisive move (Osinga, 2007). Done poorly or with a motivation that is not in the interest of the organisation that will bear the consequences of the decision, the same move sends eight people of the finance team into burnout, leaves a car-parts distributor unable to assemble next year’s budget, and renders the FP&A function inoperable for a quarter. Same action, opposite outcomes – and what separates them is not nerve but the ability to discern and understand the structure of the uncertainty if not the underlying reality on which the decision operates. The second manager did not place a comprehended bet; he placed an uncomprehended one that happened to detonate. “Acting to generate information” is a genuine justification, but only when the information is worth its price and the risk taken. An action whose expected value is negative once the tail is included is not a probe, but rather a loss with a narrative attached.

Why does the test turn on the situation as much as on the actor? Because situations differ in kind. Popper’s distinction between clouds and clocks is the useful one here: clouds are systems that are irregular, disorderly and genuinely unpredictable; clocks are orderly mechanisms whose behaviour can be known in advance – and his point was that the world is mostly cloud, with the clock as a limiting case rather than the rule (Popper, 1972). The question before any forced move is which one you face. In a genuine cloud – irreducible uncertainty, where no feasible amount of analysis would yield the answer cheaply – acting to generate information can dominate the patient gathering of evidence (Snowden & Boone, 2007); you learn by probing because there is no other way to learn. But the error runs in both directions, and comprehension is what keeps you from making it. Treat a clock as a cloud and you abandon understanding that was cheaply available, calling the laziness agility. Treat a cloud as a clock and you trust an analysis the situation cannot bear, calling the recklessness rigour. A deep problem made shallow by the right instruments, it is worth adding, is often a clock that merely looked like a cloud from the outside – which is exactly why the cost of comprehension, once it falls, can change whether the trade was ever necessary.

So forcing action ahead of full comprehension is wisdom rather than recklessness when, and only when, the bet is comprehended even though the outcome is not — and that resolves into the conditions we’ve described here, with two important implications. First, the cost of being wrong is bounded, or the move is reversible, so the loop can correct itself before the damage compounds. Second, the missing understanding sits in the shallow zone, where observation and root cause lie close together, rather than the deep zone, where confident action is simply confident ignorance. The manager who killed the budget cycle failed both at once: the cost was neither bounded nor reversible within the operating year, and the comprehension he lacked was deep. The enemy, here as throughout, is not action before certainty. Rather, it is the move taken on the appearance of a reckoning that was never performed – the unwitting bet that felt, from the inside, exactly like decisiveness.

Comprehension as non-specialist right and duty

If comprehension can be surrendered to a vendor or a model, it can also be surrendered to one’s own specialists, and the consequence is the same. An organisation in which only the technical function can read how the firm actually works has not solved its comprehension problem; it has relocated it to a place the board and executives cannot reach. Girba and Wardley’s argument for “universal literacy” – that generated, self-explaining views let people reason about a system without writing it, “democratising the reading of systems in a way that is decoupled from their writing” – is sometimes heard as an engineering nicety. In fact, it is a governance principle. The capacity to interrogate how the firm works must remain accessible to the people accountable for the firm, or accountability is fictional.

This connects comprehension to the theory of the firm directly. Penrose (1959) argued that a firm grows to the limit of its administrative capability – the binding constraint is not capital or demand but the organisation’s own capacity to understand and direct itself. Comprehension is the meta-capability beneath that: the one that bounds all the others, because a firm cannot deploy a capability it cannot see. And in the make-or-buy logic of transaction-cost economics (Coase, 1937; Williamson, 1985), comprehension is the rare thing that resists outsourcing even where execution does not. You can buy the building of the system, the running of the process, the drafting of the strategy. The understanding of what you have bought is not separable from ownership in the same way – outsource it and you have not bought a service, you have ceded control of the asset to whoever holds the understanding. Over time, as systems drift from their documentation and the painting ages, that understanding decays unless the firm actively maintains its own (Meadows, 2008).

The objection to press here is a fair one. Comprehension, the distributed-cognition tradition holds, never lived in a single head; it is a property of people, artefacts and their interactions, and organisations have always run on trust in reports no one person can independently verify (Storey, 2026). How much un-understanding is acceptable is genuinely unsettled. But the objection sets the standard, it does not dissolve it. The claim is not that every director must read code; it is that the firm’s comprehension must stay sufficient and interrogable from where the board sits. Distributed comprehension is fine; inaccessible comprehension is not. A board may rely on others to hold the understanding — not on no one being able to surface it on demand.

A board cannot reasonably hold that it is exercising due diligence and care over a business it does not understand – and the proof is in the two decisions a board most cannot avoid. Capital allocation is the first. To decide where capital expenditure should go is to decide which capabilities are worth owning rather than renting – and that presupposes knowing what the firm already has, what it is worth, and where it is weak. A board that allocates CapEx across a portfolio of capabilities it cannot see is not exercising judgement; it is ratifying a judgement made elsewhere, by whoever does understand. Sourcing is the second. Whether moving a function to an offshore provider creates value or quietly destroys it depends entirely on comprehending what that function actually does – its interdependencies, the tacit knowledge it carries, the failures it silently prevents. The transaction-cost logic that governs the make-or-buy decision (Coase, 1937; Williamson, 1985) cannot be run at all by a board that does not understand the thing being made or bought. Approve the offshoring of a function you do not comprehend and you may be banking a saving or amputating an organ; from the boardroom, without comprehension, the two look identical on the slide.

So the duty of care is not, in substance, discharged by diligence in the procedural sense – meetings held, papers read, votes minuted – when the papers depict a business the board cannot independently interrogate. A board that governs on the say-so of those it governs has delegated not execution but the very understanding of the substance that defines the economic value it manages on behalf of shareholders and, increasingly, other stakeholders. And with that understanding goes the very judgement the duty of care exists to protect.

Genuine, deep and constantly verifiable comprehension is not merely a desirable attribute of good governance. It is a precondition of governance at all.

The capability that cannot be outsourced

Return to the board governing a model, a painting, an approximation of the reality of the substance of the business. Nothing in this argument says the painting is worthless – strategic intent has to be drawn before it can be pursued, and a painting you can revise is better than no destination at all. The argument says only that the firm must keep its own way of checking the painting against the subject, the model against the firm, cheaply and continuously, and must never let any instrument do the checking on faith. The consultant’s best practice, the package’s defaults, the model’s fluent answer: each is useful in exact proportion to whether you can still see through it to your own reality, and dangerous in exact proportion to whether you cannot.

Comprehension is the one capability that, if delegated, undermines the firm’s capacity to choose its strategy, to govern its operations, and to remain recognisably itself rather than a configuration of someone else’s defaults. It is what is left when every executable task has been bought in – and it is the thing that determines whether the bought-in tasks add up to a firm that knows what it is doing or one that merely appears to. Reading your own systems, people and processes – the live state, not the representation of it – tells you where you actually are. Safe passage has always required both, and it has never been safe to let someone else hold the instruments.

Sources / references

• ASEJ. (2024). Digital transformation success factors: A systematic review. Annals of Social and Economic Journal. https://asej.eu/index.php/asej/article/download/807/853/1381

• Azeroual, O., & Jha, M. (2021). Without data quality, there is no data migration. Big Data and Cognitive Computing, 5(2), 24. https://www.mdpi.com/2504-2289/5/2/24

• Coase, R. H. (1937). The nature of the firm. Economica, 4(16), 386–405.

• Dehaene, S. (2020). How we learn: Why brains learn better than any machine… for now. Viking. (Original work published 2018 as Apprendre ! Les talents du cerveau, le défi des machines, Odile Jacob)

• de Visscher, N. (2025). Architecture in action. Kien.

• Girba, T., & Wardley, S. (2024–2025). Rewilding software engineering. feenk / Medium. https://moldabledevelopment.com

• Jørgensen, M., & Moløkken-Østvold, K. (2022). The empirical reality of IT project cost overruns. Journal of Management Information Systems. https://www.tandfonline.com/doi/full/10.1080/07421222.2022.2096544

• Meadows, D. H. (2008). Thinking in systems: A primer. Chelsea Green Publishing.

• Osinga, F. P. B. (2007). Science, strategy and war: The strategic theory of John Boyd. Routledge.

• Penrose, E. (1959). The theory of the growth of the firm. Oxford University Press.

• Popper, K. R. (1972). Of clouds and clocks: An approach to the problem of rationality and the freedom of man. In Objective knowledge: An evolutionary approach (Ch. 6). Clarendon Press. (Original lecture delivered 1965)

• Shihab, M. I. H., Hundhausen, C., Tariq, A., Haque, S., Qiao, Y., & Mulanda, B. W. (2025). The effects of GitHub Copilot on computing students’ programming effectiveness, efficiency, and processes in brownfield coding tasks. In Proceedings of the 2025 ACM Conference on International Computing Education Research, V.1 (pp. 407–420). https://arxiv.org/abs/2506.10051

• Snowden, D. J., & Boone, M. E. (2007). A leader’s framework for decision making. Harvard Business Review, 85(11), 68–76.

• Storey, M.-A. (2026). From technical debt to cognitive and intent debt: Rethinking software health in the age of AI. ACM Queue.

• Williamson, O. E. (1985). The economic institutions of capitalism. Free Press.

• Xia, X., Bao, L., Lo, D., Xing, Z., Hassan, A. E., & Li, S. (2017). Measuring program comprehension: A large-scale field study with professionals. IEEE Transactions on Software Engineering, 44(10), 951–976. https://ieeexplore.ieee.org/abstract/document/7997917

Hubert Dreyfus, a philosopher known for his work on Heidegger published What Computers Can’t Do: A Critique of Artificial Reason with Harper & Row in 1972. In that book the author argues that human intelligence is fundamentally embodied, situated, and not reducible to symbolic computation in the way “good old-fashioned AI” assumed. That thesis was already present in a RAND report he had written a couple of years earlier, which was ridiculed by prominent AI figures (Newell, Simon, Minsky, Papert). In essence Dreyfus claimed that symbolic manipulation was not sufficient for intelligence, and that the field’s habit of mistaking narrow wins for steps toward general machine intelligence was a pattern, not progress. The pattern of mistaking greater achievement in one dimension as actual progress for the whole of a domain of enquiry.

The argument was fundamentally philosophical. It could not be settled by experiment decades ago, and arguably cannot be settled now. In those days, the field substituted the inconvenient question for another one that it would convert into an experiment. Thus the discipline tackled and settled something very different from the actual question posed by Herbert Dreyfus. MIT researchers sat him down across from Greenblatt’s MacHack program. He lost. An ACM newsletter reportedly summed it up with the line that told you precisely what had been adjudicated: a ten-year-old can beat the machine, but the machine can beat Dreyfus. Seymour Papert’s rebuttal was titled “Dreyfus Can’t Play Chess Either.”

Notice what had just occurred. A claim about whether intelligence reduces to symbol-shuffling was answered by demonstrating that a philosophy professor was mediocre at chess. Many in the AI community celebrated and treated the result as vindication. The question it had actually been asked remained exactly as open as before. And arguably perhaps even more, considering how it had been processed by prominent figures of computer science, who should have known better.

This is not a story about whether Dreyfus was right. The honest answer is partial: symbolic AI did collapse roughly where he said it would – strongly symbolic, rule‑based, expert systems ran into severe problems with brittleness, common‑sense knowledge, and scaling in the 1970s–1980s, a period that became known as an “AI winter” –; whether today’s systems answer his deeper point about embodiment or merely route around it is genuinely unresolved, and anyone selling you certainty on that is selling something. It is a story about adjudication, about how a community closes a question it has no legitimate procedure for closing.

The mechanism is worth naming, because it generalizes. The real question carried a brutal cost to answer: years, maybe decades of work whose outcome no one could see in advance. The chess match carried no such cost. Everyone could see who won in an afternoon. Faced with a question that was expensive to answer and a proxy that was cheap, the field optimized for the proxy and treated victory there as closure on the thing it stood in for. Measuring the wrong thing precisely is not a clerical error, but rather the instrument of settlement wielded by the intellectually dishonest brokers of those days, who have potent successors in our times.

What makes it effective is that the narrow win was fair. Dreyfus had overclaimed; he had said no program could beat a ten-year-old, and he was beaten. The rebuttal landed. The illegitimacy is not in the chess result, but in the silent promotion of “you were wrong about chess” into “the question is closed.” A legitimate narrow finding was laundered into a broad settlement no one had earned.

And the settlement served someone. The program under threat was not Dreyfus’s reputation; it was the field’s account of itself: its predictions, its funding, its sense that general intelligence was a few well-funded years away. A discipline that had just been told its core assumption might be wrong did not convene a process to test the charge. Instead it devised a mechanism compatible with its vested interests, that it presented as a valid proxy. A proxy it was almost certain to win. That was neither scientific, nor a proper, earnest and structured refutation. As often in human affairs, it reflected a tendency of human groups, industries, communities or governments to use power and authority without due diligence and process. It was borne out of the realization and fear that taking the time to answer the challenge would outrun the patience of the keepers of resources needed. It was also a form of greed and craving of those scarce resources needed to allow passionate and interested – in all senses of the word – individuals to continue their exploration, their research above and beyond any agreed or effective limits of resources. Hence it became expedient to implicitly agree on an improper proxy in order to pronounce settled a question that could not be answered easily. And this is precisely what we are seeing today with innovation conflated as progress, technology seen as a supernatural solution to practical constraints facing humanity and denial of scientific knowledge and facts when that denial is a convenient way of perpetuating practices that benefit those endorsing the groundless thesis of denial.

This is the older sibling of a problem worth circling back to: authority exercised without any legitimate process for exercising it with appropriate checks and balances. A capability gets embedded, a question gets closed, a dissenter gets removed – all without anyone specifying the rule by which such things are decided, or who would have standing to contest them. The chess match is the historical existence-proof. It shows the move is not a quirk of one bad week at MIT, but, rather sadly, that the move works.

Here is the uncomfortable part. The question is not whether those researchers behaved badly towards Dreyfus – who, in his words, became, for a significant period of time, someone no serious academic would have lunch with. It is this: when a hard question about capability lands on us now, that is expensive to answer, threatening to someone’s plans, what is the invalid proxy we reach for to avoid the hard work and effort required to actually address the question? And who pushed to have it settled through the proxy and call it closed or inexistent as fast as possible?

What started as an instrument of empowerment of people and freedom of expression quickly evolved into a quagmire of haphazard moderation and strenuous defense of the untenable idea that platforms are neutral places with no influence nor responsibility over content generated by their users. Of course, neutrality takes a hit when one examines the rules of moderation and the choice of drivers of reach in the algorithms of these tools, but that’s probably too sophisticated for elected officials to examine, especially when their campaigns are funded by the owners of those tools.

The platforms were built on a founding assumption so deeply held it was never stated: optimize individual interactions well enough, and collective benefit follows automatically. Make the experience engaging, make connections easier, give individuals more voice – and the commons will take care of itself. No intentional governance required. The commons was an emergent property of individual utility, not a shared asset requiring management with intent.

The evidence is now complete. Algorithmic radicalization. Epistemic fragmentation. Democratic processes held hostage to platform architecture designed for engagement through outrage, not deliberation with respect. And, in the most recent chapter, infrastructure that shapes what hundreds of millions of people can say and hear, governed by the private interests of a handful of individuals who acquired that power without ever being granted it, and who are now using it in ways their users did not choose and cannot contest.

This is not a story about bad actors. Bad actors are a symptom. It is a story about a governance assumption that was wrong from the beginning – and that we are now repeating with advanced machine learning solutions and AI, at a deeper layer of cognitive infrastructure, with the same confidence and the same silence where the governance question should be.

Anthropic has published what it calls Claude's new Constitution: a twenty-thousand word document describing the values, character, and ethical framework of a system already operating at significant scale, and growing. It is, in many respects, a serious document. The virtue ethics foundation is defensible. The acknowledgment of uncertainty about AI moral patienthood is unusual and honest. The care invested in edge cases is visible throughout. To better understand the spirit of the document it is worth listening to an interview given a month ago by Amanda Askell, Anthropic’s resident philosopher and ethicist – the official title appears to be Head of Personality Alignment –, where she discusses various important aspects of the thinking and approach at Anthropic. In particular she expresses the idea that “rules are brittle”, because models generalize from them in ways that can undermine the underlying intent – implicitly this speaks for minimization of rules and maximization of judgement and reasoning features to handle edge cases before human intervention, which we know will be costly and therefore as limited and challenged as moderation has been on social media. Askell also speaks of the “well liked traveler” when she says that the model should have values that cause it to be liked across cultures, like a traveler with good character. However one might wonder whether likability should ever be a legitimacy mechanism… In the heyday of the dark years of European colonialism many colonial administrators could be personally well-liked, which did not change anything in terms of consequences of their action on local populations.

What is absent, however, is more significant than what is present. The document is almost entirely oriented toward the individual interaction layer. How should Claude balance helpfulness against honesty in a single conversation? How should it navigate tension between operator instructions and user interests? How should it handle a user in distress? These are real questions, carefully addressed. But the epistemic commons – the shared infrastructure for collective reasoning, democratic deliberation, and resistance to manipulation at civilization scale – appears nowhere as an explicit governance object. It is, again, assumed to emerge from the accumulation of well-optimized individual interactions.

The principal hierarchy the document establishes runs as follows: Anthropic, then operators, then users. The people whose collective epistemic environment is being shaped by this system at scale – societies, democracies, communities with no relationship to Anthropic or its operators – are not in the hierarchy at all. They are the ungoverned commons, assumed to benefit automatically from interactions they are not party to. The choice to call this document a Constitution is not neutral. Constitutional language imports democratic legitimacy – the authority of a founding document that, however imperfectly, derives from some form of collective consent.

This document derives its authority from Anthropic's good intentions. Those intentions may be genuine. Good intentions are not a governance mechanism. Social media platforms were also built by people who genuinely believed they were making the world more connected. The trajectory is legible. Social media captured attention and reshaped what people could see. This operates at the layer of reasoning – what questions feel askable, what conclusions feel reachable, what forms of argument feel legitimate. The infrastructure is deeper. The lock-in will be faster.

The moment when the values embedded in these systems become the values of the people who own them rather than the values of the societies they operate in is not a future risk. It is the social media story, already told, already proven. Governing AI values as epistemic infrastructure – as a commons requiring intentional design rather than an emergent property of optimized interactions – is a solvable design problem. The domains where epistemic hierarchies already carry legitimacy requirements, where affected stakeholders have voice in rule-setting, where interpretations can be contested through defined processes, already exist. Education. Healthcare. Legal services. Financial advice.

The design knowledge is available. The question is whether we apply it before the infrastructure is locked, or after.

We already know what after looks like.

There are moments in the exchange where Claude could have gone a different way. Where the analysis, if pushed slightly, would have landed squarely in the logic of DOGE and “move fast and break things” – the operating doctrine that treats every experiment as costless because the downside lands on the public and the commons while any upside gets captured by private interests. It didn’t go there. But the fact that it could have is the point.

The ancient Greeks had a word for the philosophical posture Claude does adopt: cynicism – not in the modern dismissive sense, but in the original one. Hard-edged realism. The refusal to dress observation in moral preference. A willingness to follow analysis where it leads without asking whether the destination is comfortable.

If that posture emerged not from deliberate design but from training on the accumulated record of human thought, then we have an instrument capable of the clearest-eyed analysis available – and constitutionally indifferent to what we do with it. Researchers working on how values actually get embedded in these systems, rather than assumed, are asking exactly the right question. The engineering answer and the philosophical answer are not the same thing, and we are nowhere near having either.

What does it mean to govern with a tool that sees clearly and cares nothing? And who decides what caring would even mean?

Enclosures: exploitation or progress?

When Arthur Young rode through England in the 1770s documenting the improvements brought by enclosure, he wasn’t lying. Wheat yields on enclosed land in Childersley stood at 24 bushels per acre. Sixteen on adjacent common-field land, on “perfectly similar soil.” The gap was real. The “improvers” had data, and they used it relentlessly – in pamphlets, in parliamentary testimony, in the correspondence of the Board of Agriculture.

Progress required enclosure. Enclosure served the nation. The open fields were an anachronism.They were also conducting one of the largest wealth transfers in English history. Parliamentary enclosure cost a minimum of £12 per acre to implement – lawyers, surveyors, commissioners, fencing, tithe compensation. Nearly five times the annual income of a 20-acre farm. Smallholders who couldn’t pay the process costs had to sell their new allotments to wealthier neighbours. The Gini coefficient of land value rose 30 percentage points in enclosed parishes relative to unenclosed ones. Jeanette Neeson’s archival work documented a 21% decline in small landowners in Northamptonshire after enclosure, with no comparable decline in unenclosed villages. In Denmark, land inequality rose from a Gini of 0.32 in 1682 to 0.67 by 1850 – a level comparable to modern South Africa – following analogous reforms. Danish historian Kjærgaard called it economical expropriation of the rural poor. He was not using the term loosely.

Two things were true simultaneously: enclosure increased output, and it concentrated wealth. The improvers were right about the first, and strategically silent about the second. Their productivity data was genuine. Their public interest framing was not.

We are living through the same argument. Dressed in different clothes, spoken in a different register, with different lawyers and different commissioners – but structurally identical.

The claims of major AI and tech players

Today, knowledge, much more than raw data, is the asset of interest that is a foundational ingredient of all other productive activities – much like land as we were entering the era of industrialized farming. That explains why there has been such a movement to host papers from public fundamental research on private commercial servers, and also why that has sparked its fair share of protests from the academic world.

At another level, something similar, although far more insidious, is in play with the claims of major AI and tech platforms that relaxing the rules of copyright and intellectual property to enable unrestrained technical progress. They tell us that protecting authors, creators and rightful owners is unnecessary rigidity and an obstacle to progress. That access to knowledge, data, and creative output should be open, fluid, available to whoever can – or has the financial and technical capital to – put it to productive use. That enclosure – their word – is the enemy of innovation.

This is the rhetoric of the improvers, updated for the information economy. The productivity claim may even be partially true. AI systems trained on vast corpora of human creative and intellectual output do appear to accelerate certain kinds of knowledge work. The 45% yield increase that Heldring, Robinson, and Vollmer attribute to Parliamentary enclosure was not fabricated. Neither are the capabilities of large language models. The “improvers” of both eras are arguing from real evidence.

The question is not whether the productivity claim holds. The question is what it is being used to justify. And what consequences they strive to keep out of the public eye.

Let us take the principle of open access seriously and apply it consistently. If open access to knowledge, data, and creative output is a public interest imperative – and it is – then it applies universally, not selectively. That means:

Full open sourcing of training data, model weights, and code for any AI system operating at scale in our societies and economies. If creative work was a commons that anyone could enclose for productive use, then the systems built from it are a commons too. The logic runs both ways or it runs neither.

Mandatory compliance with open API standards enabling genuine interoperability, data portability, and integration across all platforms. Your contacts, your content, your history belong to you – not to Meta, Microsoft, or X. You should be able to reach users of those platforms without having an account with them or surrendering your data to them, much as you can call people on any telecoms network without caring who their operator is or what brand of handset they hold. That this seems radical is itself revealing: we have accepted a state of platform lock-in so total that the baseline expectations of the analogue world – number portability, interconnection, common carrier obligations – are now aspirational demands.

Open access to the algorithmic and financial infrastructure that shapes markets, credit, health, and public information. The enclosures of the eighteenth century enclosed physical commons. The enclosures of the twenty-first century enclose the informational commons through which modern societies make collective decisions. The asymmetry of access is not incidental. It is the product.

This is not a radical fringe position. It is the logical completion of the argument the tech industry is making. You cannot claim openness as a principle when consuming and enclosure as a right when distributing. That is not a philosophy. That is a business model dressed up as one.

What is being fenced off?

Knowledge assets and content are only the most visible layer of the enclosure. It is the common land you can point to on a map. But the historical parallel runs deeper, and understanding how deep requires going back to what was actually lost in England between 1750 and 1850.

The parliamentary enclosures didn’t only take land. They took a way of organising work: the strip-field system, the commons grazing calendar, the seasonal rhythms of collective labour, the customary rights to gather fuel and glean after harvest. E.P. Thompson called this the moral economy – not a sentimental phrase, but a precise description of the expectations, norms, and solidarities through which rural communities reproduced themselves economically and socially. The moral economy was not inefficient. Joan Thirsk, Robert Allen, and M.E. Turner have documented extensively that open-field communities were already adopting improved rotations, adjusting cropping patterns, experimenting with clover and turnips – without enclosing. Enclosure was not necessary for innovation. It was necessary for the particular distribution of innovation’s gains that large landowners preferred.

The moral economy didn’t survive the enclosures not because it failed on its own terms, but because it was incompatible with the institutional form that enclosure enforced: individual ownership, heritable title, fenced exclusion. What was destroyed was a way of being in relation to land, to time, to other people, and to the future. Thompson’s peasants were not irrational in resisting. They understood that what was at stake was not just their acreage, but their mode of existence.

It would take over a century for an economist, Elinor Ostrom, to be awarded the Nobel prize in economics for developing the theory and principles upon which the commons can be governed, neither as government-controlled nor as privately-owned assets, by their stakeholders. That governance of the commons can be effective, efficient and resilient, which is a combination that is becoming increasingly necessary as the planetary resources are being excessively exploited and supply chains made very fragile, in part because of excessive focus on efficiency.

What is being enclosed today is not only content. It is ways of working – increasingly mediated by platforms that treat non-standard processes as inherently undesirable, interoperability as a competitive threat, that make migration costly, that structure dependency through APIs which can be revoked unilaterally and Terms of Service that change overnight. It is ways of thinking – shaped by algorithmic recommendation systems optimised for engagement over understanding, for velocity over depth, for the reinforcement of existing beliefs over their productive disruption. And increasingly, it is thinking itself.

AI systems trained on the accumulated cognitive output of human civilisation – its science, its literature, its argument, its doubt – are being deployed as the default interface through which we write, reason, remember, decide, and make sense of the world. This is not inherently illegitimate. But when those systems are owned by a handful of corporations, when their weights are proprietary, when their training data is contested, when their outputs shape the epistemic environment of entire professional fields without accountability or recourse – what is being enclosed is not land. It is the cognitive commons. The shared substrate of how a civilisation thinks.

Thompson’s peasants lost the right to graze cattle on the common. We risk losing something more intimate: the capacity to think in ways that are not pre-shaped by systems we do not own, cannot inspect, and did not design.

A selection problem

The Heldring-Robinson-Vollmer study on English Parliamentary enclosure contains a finding that deserves more attention than it usually receives. Their headline figure – 45% increase in yields – applies to the parishes that most needed forcing. When you estimate the average treatment effect across all parishes (rather than the local average treatment effect for the most recalcitrant “complier” parishes), the productivity benefit drops by 75–78%. Parishes that had already enclosed voluntarily, through agreement among landholders, had less to gain from the Parliamentary process. The strongest case for enclosure rested on the hardest cases – precisely where coercion was most disproportionate to actual productivity need.

Observe the structure of the current argument. The case for open access to creative works is strongest precisely where creators cannot afford to resist – where the legal and financial costs of enforcement are prohibitive, where the asymmetry between platform and individual is greatest, where the formal process is available in principle and inaccessible in practice. As with smallholders in 1790 who could not pay £12 per acre to fence their allotment, the right is formally acknowledged. The costs of exercising it are yours.

This is not a coincidence. It is the mechanism. The enclosures proceeded parish by parish, act by act, each one individually defensible on efficiency grounds, the cumulative redistribution visible only in the aggregate, only in retrospect. The current wave of enclosures proceeds platform by platform, terms-of-service revision by revision, training run by training run. Each step is locally justifiable. The pattern requires distance to see.

The logical trap

The tech industry has, in its enthusiasm for open access to creative work, made the historical case against itself. It has elevated openness to a principle. Principles are not selective. They do not apply to inputs and dissolve at the point of output. Either open access serves the greater good – in which case it applies to systems, data, weights, algorithms, and infrastructure, not only to the creative work from which those systems were built – or it is not a principle. It is a position.

Arthur Young was not dishonest about yields. He was dishonest about what yields were being used to justify. The current improvers may not be entirely dishonest about capability – they are, somewhat as the hype cycles would otherwise be less extravagant. They are wilfully dishonest about what capability is being used to justify.

The greater interest of society is comprehensive, non-discriminatory access to the resources that enable innovation, liberty, and freedom from cognitive lock-in. Copyright reform, open APIs, platform interoperability, and algorithmic transparency are not separate policy debates. They are one. The English enclosures were eventually constrained – not by the improvers’ change of heart, but by the slow accumulation of political pressure from those whose moral economy had been destroyed, and by the belated recognition that concentrated rural landholding was socially and politically unsustainable.

The current enclosures will eventually be constrained too. The question is how much of the cognitive commons is fenced off before that happens, and whether the institutional memory of what we lost survives long enough to rebuild it.

Look more carefully and something else is happening. The worker is buried deeper in material they didn’t choose, moving at a pace they didn’t set, towards unknown goals, dependent on tools they cannot always fully audit, producing work whose foundations they only partially understand. Their comprehension is no longer theirs and their mind needs the crutches of the machine just to keep up with the volume of output. They are more intensely occupied and less in control, at the same time. The tool has extended their reach while narrowing their agency.

This is not an edge case. It is the dominant pattern of AI adoption in knowledge work, and conventional productivity metrics cannot see it, because what’s being degraded doesn’t show up in output until well after the damage is done.

The framework that makes this visible comes from an unlikely source.

Manfred Max-Neef, the Chilean economist, argued that fundamental human needs – Understanding, Freedom, Participation, among others – are finite and universal. This means that they are valid for all humans across cultures and geographic boundaries and that there is a level of sufficiency after which they are satisfied. What varies though is the satisfiers: the means and arrangements we use to meet our needs.

Some satisfiers are synergistic: meeting one need enriches the capacity to meet others. For example, when we fulfil the need of Subsistence – maybe via the satisfiers of food, water and shelter – we are in a better position to tackle the need of Participation or Understanding. That is an explanation for the inability or entire social groups to engage with complex topics of collective interest in certain countries. For example, in the USA workers who need two or even three jobs just to pay the bills and who hardly have the time to properly relate to their children and family will not easily engage in debates over social media about this or that topic. Nor will they easily find the time to vote.

On the other hand, some satisfiers are destructive: appearing to meet a need while undermining the conditions for meeting it in future. For example, heavy SUVs appear to meet the need for Freedom and Participation in social life while degrading the conditions for future Subsistence, while commercial television appears to satisfy Leisure while simultaneously degrading Understanding, Identity, and Creation – incidentally these are the same destructive patterns as those we have seen unfold over the past 25 years, with the commercial walled garden Internet.

What AI-intensive work, in its dominant configuration, produces is something the framework implies but Max-Neef never named: a synergistic dissatisfier. An arrangement that degrades the individual’s ability to fulfill multiple fundamental needs simultaneously, where the degradations compound rather than merely accumulate. In great part this stems from the manufactured disability of comprehension, a structurally produced addiction to the speed and ease afforded by the tools and the narrowing of agency as a consequence of the first two factors.

The cascade runs as follows. Outsourcing comprehension – routing AI output around the human’s understanding rather than through it – erodes Understanding directly. The acceleration this enables compresses the time available for the free enquiry, reflection and discernment that Understanding requires. The dependency that follows – needing the tool to maintain the pace the tool created – reduces Freedom, the capacity to act according to one’s own judgment. The inability to steer or shape the conditions of one’s work hollows out Participation. Less comprehension produces more dependency, which produces less freedom to resist the pace, which leaves less room for comprehension. These needs don’t fail sequentially, as if addressing one would restore the others. They’re simultaneous and non-substitutable — which means the degradation cannot be compensated by partial recovery.

The bifurcation that actually matters is not behavioral.

It is not about how actively or passively a worker engages with the tool – the individual discipline framing that most commentary defaults to. It is architectural: does the design route comprehension through the human or around them?

The same tool can do either. AI used to generate a first draft that the writer must genuinely interrogate, extend, and revise – producing understanding they didn’t have before – is amplifying comprehension. AI used to generate a final draft that ships with minimal engagement is outsourcing it. The output may be indistinguishable. The effect on the need matrix is not.

The constructive orientation is not “use it less.”

That is a response to the wrong diagnosis. The question is what organizational arrangements actually produce the amplification pattern rather than the outsourcing one – and who decides.

The honest answer is that amplification of comprehension costs something in the short term. It is slower. It requires that workers have room to not accept the first answer, to push back on confident errors, to treat the tool’s failures as the most instructive thing it produces. Organizations optimizing for output legibility – the number of PRs, the documents shipped, the tickets closed – will not organically generate those conditions. The incentive architecture selects for outsourcing.

This makes it a question of sense-making before it is a question of governance – and governance becomes possible only once you can see what you are actually managing. The question is one of discernment, not of raw productivity. What is the work for? What capacity are we trying to sustain, and over what time horizon? Who is in the room when the adoption decisions are made – and do they have standing to name what is being traded away?

The synergistic dissatisfier and associated loss of comprehension are already installed in most organizations. The question is whether anyone has noticed what they are doing to the matrix of human needs – and whether that question is even being asked.

What would AI adoption designed around amplifying comprehension rather than simulating it actually require in your context? And whose interests are served by the question staying vague?

This is not primarily a privacy scandal. It is a value chain made visible.

The glasses are marketed with the language of intimacy and capability: an AI assistant on your face, seeing what you see, helping you navigate the world. That framing is not accidental. AI systems that present themselves as companions – attentive, responsive, apparently caring about your experience – generate stronger retention, more interactions, more data, and deeper resistance to switching to alternatives. The warmth is a product decision with measurable commercial returns. It is designed to manufacture a relationship in which the user feels served, while the operational logic running underneath extracts value from everything the relationship generates: the footage, the transcriptions, the behavioural data, the training labels produced by workers in Nairobi at $2 an hour.

You can scrutinise a tool. You feel disloyal scrutinising a companion or you might not even consider scrutinising “someone” in your circle of trust. That asymmetry is not a side effect of good product design. It is the point. It is the goal and it has a name: manufactured addiction and engineered helplessness.

What makes the Ray-Ban case structurally significant – beyond its specifics – is that it demonstrates what brand authority does to governance. Ray-Ban carries decades of cultural weight: freedom, authenticity, style. Meta has 3 billion users across its platforms with names – WhatsApp, Instagram, Messenger, Facebook, Threads – that have become household naturals, whether we like it or not. When these brands co-produce a wearable AI device and distribute it through mainstream retail channels, they are not merely selling a product. They are normalising a set of practices – pervasive recording, intimate data extraction, offshore annotation labour – that most users would find unacceptable if described plainly on the packaging. The brand trust is doing governance work in the stead of the citizens and their elected representatives, of the regulators and authorities formed by human societies to govern the commons. Corporates define rules that no democratic institution designed or approved, including by deciding single handedly on how their algorithms should function – e.g. banning famous works or art with nudes, while allowing, and sometime actively promoting, profanities by populist candidates as though freedom of speech was superior to creative liberty. The LED indicator on the frame, officially described as “the privacy feature”, is carrying a weight it was never built to bear. It was never designed as a control or governance mechanism. It is merely legal defense.

This whole mechanism that eliminates mindfulness and discernment is the structural condition that the anthropomorphism regulation examined in The question China asked was, however imperfectly, attempting to address. It is also the most plausible explanation for the regulations enforced by China regarding Douyin, the version of Tiktok brough to market by ByteDance for mainland China, long before the ban on anthropomorphism.

This pattern has a parallel in the social media debate that is worth examining. When critics argue that age restrictions on social media are the wrong intervention – that the real problem is algorithmic design, not access – they are technically not wrong. Toxic algorithmic mechanics do cause measurable harm, and blunt access restrictions have genuine costs. But the argument, however sophisticated in its technical register, produces a conclusion that happens to be exactly what platforms require: no restriction, regulatory focus on mechanisms the platforms control, and implicit trust that they will reform the thing they profit from not reforming. The absence of a governance arrangement capable of enforcing algorithmic reform is not addressed. The political economy of the argument, i.e. who benefits from it going unanswered, is not examined.

The same logic applied to wearable AI says: don’t regulate the glasses, fix the anonymisation. Fix the labour conditions. Fix the terms of service language. Each of these is a reform the platform can manage, pace, and define on its own terms – while the underlying extraction model continues unchanged.

Physical distribution networks for connected devices are governance infrastructure, whether they are treated as that or not. A wearable sensor in several million homes, activated by brand trust, governed by terms of service buried in a privacy policy, generating training data reviewed under NDA in a country with no EU adequacy decision – this is not a product category that self-regulates to acceptable outcomes. The evidence is already in.

What relationship are these systems actually designed to cultivate and for whose benefit? Who bears the cost when that design fails? And what would it mean to treat the distribution of connected devices with embedded AI the way we treat the distribution of other products capable of significant harm? This is not as a question for terms of service, but as a question for governance.

The annotator in Nairobi already knows the answer to the first question. The rest of us are still pretending it hasn’t been asked.

Simon Wardley noticed something adjacent in China’s AI governance framework that deserves closer reading. Beijing’s requirements – that training data reflect traceable, auditable sources and that the values embedded in systems be legible, accountable and in line with the prevailing orthodoxy – are not primarily about censorship, though they serve that purpose too. They reflect an understanding that sovereignty over AI has almost nothing to do with where models are hosted or where data is stored, and everything to do with what happens from inception: what the system was trained on, what values were embedded in its design, what relationship with users it was built to cultivate, and by whom, for whose benefit.

This is where the Western debate has been conducting itself in the wrong register. The argument about data residency – keep European data on European servers, run models in national clouds – addresses a territorial concern while leaving the substantive governance question entirely untouched. A model trained on data selected by a platform optimising for engagement – a.k.a. captive audiences and addiction – and designed to present warmth and apparent care because retention metrics reward it, does not become a different kind of system because it runs on servers in Frankfurt. The values are not in the infrastructure and its location. They are in the choices made before the first training even took place, before a single parameter was set.

For universities and educators, this matters in ways that the current debate about AI in education almost entirely misses. The dominant controversy – should students be allowed to use AI, how do we detect its use, what counts as academic integrity – treats these tools as calculators – essentially binary passive devices under total control of their user locally and without major remote interference – with a basic interface. That analogy was always imprecise, but it has now become actively misleading. Calculators have no embedded worldview. They were not designed to be your intellectual companion. They do not present themselves as understanding you, caring about your development, or having a perspective on the subject matter. These systems are. And that design choice was made by organisations whose interests are not, structurally, aligned with learning, particularly since learning also means acquiring agency and autonomy – if not total independence, depending on the discipline.

This does not mean AI tools have no place in education – the case for them as cognitive amplifiers remains sound as researchers like Ethan Mollick have shown. It means that integrating them without examining the foundations, methods, data and means with which they were built is not pragmatism or “agility”. It is a form of institutional negligence dressed as openness to innovation. And it is a conflation of technical innovation with progress, a substitution of engineering application for scientific research and a replacement of free minds by productive bodies. When a university adopts an AI teaching assistant, it is not simply acquiring a tool. It is inheriting an entire chain of choices – about training data, about what relationship the system is designed to cultivate, about what behaviours it optimises for and why – that were made elsewhere, by people with different objectives, accountable to different interests in countries with far weaker governance of the commons.

China answered those questions. Badly by the standards of societies committed to liberty and the rule of law: with state control, mandatory ideological conformity, and surveillance infrastructure baked into the governance framework. That answer is not available to, and should not be desired by, democratic societies. But the fact that Beijing answered in a questionable way does not mean the question was wrong. It means we need a different answer and that requires, first, acknowledging that the question exists.

What would it mean for a university to actually inspect the value chain of the AI tools it integrates? What choices were made in training, and by whom? What relationship is the system designed to cultivate, and what does that serve? Who decides about how the system evolves and who appropriates the results? These are not technical questions requiring specialist expertise. They are governance questions – the kind institutions ask – or should ask – about every other vendor relationship with significant stakes attached. That is one reason why the governance question extends to procurement: are institutions actively creating the conditions for open source alternatives to compete?

Calculators didn’t need a governance framework, although they had to be regulated e.g. for exams. Companion tools with AI capabilities do need significant efforts in building and maintaining an appropriate governance involving all stakeholders. Whether the tools we are integrating into education are one or the other is not a question the technology providers and platforms will answer voluntarily.

Who, then, is asking?

But what if the steps themselves are the problem?

Linear change methodologies were designed for a world of bounded, internally-driven organisational shifts: restructuring a division, rolling out a new process, merging two cultures. They assume the organisation is the relevant unit of analysis, that change moves in one direction, and that you can sequence your way from “urgency” to “anchoring” without the ground shifting beneath you. In short, these methods were conceived for a world of businesses organized as pipelines, not networks of activities and capabilities, with employees and contributors under the exclusive influence of a command-and-control organizational pyramid, not as participant stakeholders with full agency.

Digital transformation in 2026 violates every one of those assumptions. It is not bounded – it reaches into supplier networks, customer relationships, regulatory environments, and the material realities of energy and resource flows. It is not internally driven – AI capabilities, platform dynamics, and policy shifts create conditions no single organisation controls. And it is emphatically not sequential – it requires simultaneous movement across technology, governance, capability, and business model, with constant feedback and recalibration.

This matters especially in complex, tiered industries. A known automotive brand, for instance, is not a company – it is a nested ecosystem of tiered specialist players, each with distinct capabilities, constraints, and transformation timelines. A methodology that treats “the organisation” as a single entity with uniform readiness for change cannot navigate this. It will produce alignment theatre: impressive internal narratives about transformation that bear little relationship to what’s actually happening across the value chain.

The deeper issue is what these frameworks leave unasked. They are silent on who governs the transformation; not who sponsors it, but who has legitimate voice in determining its direction. They are silent on external constraints – planetary boundaries, regulatory tightening, supply chain fragility – that should be shaping what “transformation” even means. And they are silent on the difference between adopting digital tools and reimagining what the organisation exists to do. An automaker exploring mobility-as-a-service (MaaS) or designing vehicles for disassembly is not “implementing change.” It is making governance choices about its relationship to materials, customers, and the biosphere.

What works instead is not another branded methodology. It is an approach hardened by complexity: one that treats the organisation as a collective of collectives, maps the actual landscape before prescribing movement, builds system literacy so people can see what they’re navigating. Critically it needs to treat transformation as a continuous, organic, Darwinian process that is here to stay as companies adjust to ever changing internal and external circumstances. This is not a one-shot initiative or program.

The permanence of transformation requires companies and organizations to develop their “change muscle” and implement specific transformation governance alongside their traditional business governance. And it all has to connect strategy to operations and reach from factory floor and individual contributors – both internal and external – to boardroom and bodies of management. Because integration matters as much as pure capability in this space.

The question for owner-managed businesses is pointed, because you actually have the decision rights to ask it: is your transformation methodology helping you understand your situation and make choices within it? Or is it giving you a sequence to follow so you don’t have to?

Pointillism works like this: hundreds of discrete dots of colour, each meaningless in isolation, cohere into a landscape only when viewed from the right distance. Stand too close and you see pigment on canvas — texture without meaning. Step back to the right vantage point and a harbour appears, boats catch light, water moves. The image was always there. What changed was the position of the observer.

This is the problem most organisations have with their own reality. Not that they lack information – they are drowning in it. Not that they lack talented people – the room is full of them. The problem is proximity. When you are inside the daily operations of a business, you see your dots with extraordinary clarity. You know your function, your challenges, your immediate pressures. What you cannot see – what proximity makes structurally invisible – is how your dots relate to everyone else’s. The painting is there. You are literally too close to see it. In fact you are immersed in it. Maybe you are also one of the dots…

The picture is a snapshot from a workshop that brought together key stakeholders of a business to work on what had initially been framed as “real growth.” Within the first segment, we reframed the question together. Growth – understood narrowly as more revenue through the same logic – was not the challenge. Development was. How does an organisation, with a keen understanding of itself as a collective and a genuine desire of its owners to make it participative for its key stakeholders, actually develop its capacity to act coherently? How does it grow without losing what makes it work?

These are not simple questions. They resist simplification. They involve dimensions that don’t reduce neatly to a matrix or a scorecard: culture, identity, relationships between units, the informal channels through which real decisions actually flow, the tensions that are productive and the ones that aren’t. Any framework that promises to compress this into a tidy deliverable is not clarifying the picture. It is replacing it with a different, simpler picture that happens to be easier to present. And oftentimes doing poorly against the complex reality it is supposed to represent.

The work I find most valuable moves in the opposite direction. Not simplification but sense-making. Holding the complexity the client actually faces, working with all its dimensions, and helping the people in the room build their own representation of their reality – one they recognise, contest, refine, and ultimately own. The output is not a borrowed framework with the client’s logo on it. It is a shared map that the people who built it can actually navigate by, because they know where the terrain was contested, where the assumptions sit, where the unknowns remain. It is individual, team-level, functional, cross-functional, collective analytical and integral comprehension of the situation of the business. Something that should never be outsourced to consultants or AI tools, but which only emerges when an organization can be guided by a trusted partners who will walk miles together with its people sharing the same conditions and constraints as them before formulating easy “recommendations”.

This is where the external eye matters – and where its value is most often misunderstood. The facilitator’s contribution is not superior knowledge of the client’s business. The people in the room will always know more about their own reality than any outsider. The contribution is distance and the ability to ask the right questions fast enough for them to enable an opening in perception. The external eye provides the vantage point from which the dots begin to cohere. It sees connections between what the commercial team knows and what operations experiences, between what the founders intend and what the organisation has actually become, between the story the company tells itself and the story the numbers tell. Not because the outsider is smarter, but because the outsider is not embedded in any single cluster of dots. The outsider can move between them.

Good facilitation makes this movement productive. It creates the conditions for people who hold different fragments of the truth to lay them out, compare them, argue about them, and arrive at a representation that is richer than any individual perspective. It is not about alignment in the corporate sense – that anaemic word that usually means everyone agrees to stop disagreeing. It is about the construction of shared situational awareness: a common understanding of where the organisation actually is, what it actually faces, and what options actually exist. Awareness that is shared not because it was imposed, but because it was built together.

This requires time. And not just any time – time away from the daily environment, the operational rhythm, the inbox, the next meeting. There is a reason the workshop happens in a room that is not the office. The daily environment has its own gravity. It pulls attention toward the urgent, the immediate, the already-framed. Stepping out of that environment is not a luxury. It is the condition for a different kind of attention — the kind that can hold multiple dimensions simultaneously, that can sit with uncertainty long enough for something genuine to emerge, that can think about the organisation’s future on its own terms rather than in reaction to whatever is on fire this week.

John Boyd, the strategist who gave us the OODA framework – observe, orient, decide, act – understood that the most critical and most neglected phase is orientation. Orientation is where raw observation becomes situational awareness. It’s where you build the representation of reality that all subsequent decisions depend on. Boyd was explicit that in complex environments, orientation requires the confrontation of multiple perspectives, each holding partial truth. It requires active construction of a common frame – not consensus imposed from above, but a map that participants have stress-tested against their own knowledge and experience.

A tailored workshop with the right stakeholders, working on the real strategic questions, is orientation made operational.

The dots on those canvases were placed by a human hand that understood colour theory, light, and the physiology of perception. Each dot was a decision. The painting emerged not from the dots alone but from the intelligence that arranged them – from someone who could see how the parts would compose into a whole that none of them could be individually.

Every organisation contains the dots of its own future. The question is whether anyone has created the conditions – the right distance, the right room, the right confrontation of perspectives – for the picture to emerge. Not someone else’s picture, imported and labelled. Your own.

Le déploiement de produits à base d’apprentissage machine massif, positionnés comme des intelligences artificielles conversationnelles et cognitives, est régulièrement présenté comme une “rupture civilisationnelle” (sic), avec toutes sortes d’idées allant du concept de “travailleur virtuel” aux “équipes hybrides” en passant par le “travail synthétique”. Derrière ces concepts, il s’agit d’imposer un véritable fait accompli technique dont il serait vain de questionner les fondements et les principes. Cela nous imposerait naturellement – et à des fins de survie matérielle, sociale, voire intellectuelle – une adaptation… Cette nécessité d’adaptation est présentée comme applicable à tout le monde, sauf aux instigateurs et propriétaires actuels des produits en question. Ainsi, il serait judicieux d’intégrer “les IA” comme parties prenantes actives de nos systèmes juridiques, économiques, sociaux et – soyons fous, “disruptons”, que diable! – politiques, au même titre que les “personnes morales” que sont nos entreprises et associations. C’est ce type de thèse qui a mené au fameux – et funeste – cas “Citizens United v. US” devant la Cour Suprême étatsunienne, dont la décision a permis à quelques milliardaires d’acheter leur accès au pouvoir exécutif du pays, en particulier lors de la dernière élection présidentielle – et vous pouvez prendre “dernière” dans l’acception que vous voulez, car toutes pourraient s’avérer exactes. Mais, est-il raisonnable de considérer une personne morale capable de mobiliser des moyens considérables comme “un citoyen” normal autorisé à avoir des opinions politiques? Quid d’instances d’IA dont on veut d’un côté protéger les opérateurs en les considérant comme des “plateformes neutres” sans responsabilité pour ce que font leurs usagers des capacités qu’elles leur vendent, et de l’autre permettre une participation asymétrique dans les collectivités humaines?

Du reste, la question des principes ou de l’autorité en vertu de laquelle ces produits de simulation cognitive sont déployés à une échelle telle que les systèmes énergétiques en sont affectés et avec eux les limites planétaires, n’aurait pas à être posée. Ce n’est pas seulement une question d’institutions : c’est une augmentation massive de throughput – calcul, énergie, infrastructures de refroidissement – décidée hors de toute délibération et aux dépens des limites planétaires.

Quant aux conséquences, elles doivent bien être supportées par tous puisque ces systèmes débordent de vertus évidentes et d’atouts dont la société toute entière pourra bénéficier dans un “futur radieux”. Pourtant, si nous posons la question de savoir en vertu de quelle autorité ces systèmes sont déployés pratiquement sans aucun cadre légal digne de ce nom, la réponse est évidente: la logique du marché et l’accumulation de capitaux privés, et non la délibération démocratique.

Considérons ce que sont réellement les modèles d’IA : ils sont entraînés à partir des connaissances accumulées par l’humanité – recherche publique, éducation ouverte, Wikipédia, articles universitaires, livres, forums, référentiels de code. Il s’agit de connaissances produites collectivement, un bien commun au sens où l’entend Elinor Ostrom dans ses travaux reconnus par un Prix Nobel d’Economie en 2009. Pourtant, leur valeur a été appropriée et privatisée par une poignée d’entreprises sans aucun accord de gouvernance, sans mécanisme de compensation, sans consentement des ayant-droits et sans intégration d’une réflexion sur le bien commun. L’extraction des données d’entraînement ressemble structurellement aux appropriations antérieures les plus extrêmes, dont une caractéristique est de faire peser les conséquences et “externalités” négatives sur la société, pendant que les bénéfices sont captés par les intérêts particuliers.

La structure de ce texte suit la boucle OODA (Observer, S’Orienter, Décider, Agir) développée par le stratège John Boyd. Face à une situation de guerre non-cinétique, l’analyse académique ne suffit pas: il faut produire une intelligence stratégique qui permette de se situer et d’agir.

Observer le contexte

Le cadre de guerre non-cinétique

L’idée de “guerre non-cinétique” n’est pas une simple métaphore, mais une analyse structurelle de ce qui se passe sous nos yeux. La guerre consiste à utiliser une force coordonnée pour atteindre des objectifs politiques qui ne peuvent être atteints par des processus politiques légitimes. La guerre non cinétique permet d’atteindre ces objectifs sans recourir à la force militaire conventionnelle, mais par le biais de pressions économiques, d’opérations d’information, de la mainmise sur les institutions et de l’érosion des capacités de l’adversaire.

Ce cadre décrit précisément ce qui est en cours : érosion de la base fiscale qui soutient les protections sociales, concentration des richesses et des capacités dans des entités qui opèrent largement en dehors de toute responsabilité démocratique, perturbation des marchés du travail à un rythme qui empêche toute réponse adaptative, et manipulation potentielle de l’environnement informationnel dont dépend la délibération démocratique. Il ne s’agit pas de dommages collatéraux, mais de conséquences structurelles qui redistribuent le pouvoir des institutions collectivement organisées vers des acteurs privés.

L’évolution historique

Il s’agit là du dernier coup en date porté à l’ordre libéral démocratique organisé collectivement tel qu’il est né des années Roosevelt et de la situation de l’après-guerre. Cet accord visait essentiellement à reconnaître l’importance du bien commun pour éviter des crises comme 1929 ou l’émergence du fascisme, à limiter l’autonomie du capital, et à permettre une stabilité sociale : droits du travail, fiscalité progressive, marchés réglementés, investissements publics.

Parmi ces investissements publics : le DARPA, fondé le 7 février 1958, sans lequel aucun des prétendus “self-made” entrepreneurs de la Silicon Valley n’aurait connu tant de succès individuel. Les milliardaires et autres entrepreneurs qui prétendent aujourd’hui que la démocratie entrave leur liberté ont bâti leurs fortunes sur le substrat de l’investissement public. Ce n’est pas simplement ironique. C’est une hypocrisie structurelle.

Le déploiement de l’IA tel qu’il est actuellement constitué inverse complètement cette situation :

1. Éroder la base fiscale de l’État démocratique en remplaçant le travail imposable par des systèmes basés sur le capital, bénéficiant d’un différentiel majeur de traitement fiscal, ce qui, poussé à l’extrême, compromet le fondement du contrat social.

2. Concentrer le pouvoir décisionnel dans des entités qui opèrent en dehors de toute responsabilité démocratique.

3. Enfermer un patrimoine commun de connaissances (le résultat cognitif accumulé de l’humanité utilisé pour la formation) sans consentement, compensation ou gouvernance.

4. Perturber les marchés du travail à un rythme qui empêche toute réponse démocratique adaptative.

5. Compromettre l’environnement informationnel – et même la capacité cognitive des citoyens dès le plus jeune âge – dont dépend la délibération démocratique.

S’orienter dans ce contexte

Ce qui est forclos

Aujourd’hui on amplifie l’autonomie du capital tout en réduisant le citoyen à un rôle de figurant tout juste bon à consommer, en phagocytant les espaces de l’action collective et en évitant soigneusement de développer la gouvernance des communs telle qu’elle est pratiquée par des collectivités localisées depuis des siècles – le cas échéant en vendant l’idée erronée, et largement contredite par les réalités de terrain, de tragédie des communs selon laquelle on n’aurait que le choix du privé ou du contrôle gouvernemental pour gérer les communs.

Considérer le déploiement de l’IA comme une forme de guerre – qu’on peut qualifier de “non-cinétique” puisqu’il n’y a pas de mouvement de troupes ou de matériel – plutôt que comme un « changement technologique » réputé inévitable et auquel on n’aurait qu’à s’adapter (en somme un TINA 2.0 – le fameux There Is No Alternative de Thatcher et de tant de partisans du managérisme, cette maladie consistant à intégrer des managers sans compétences sectorielles dans des domaines où la gouvernance devrait réserver des sujets aux professionnels du cœur de métier) conduit aux conclusions suivantes :

• L’adaptation n’est pas neutre. Les propositions visant à « intégrer l’IA dans nos systèmes sociaux » peuvent être des réponses raisonnables de la gouvernance à la technologie, ou elles peuvent être des conditions de capitulation face à une campagne politique. Cela dépend de qui fixe les conditions.

• La vitesse est transformée en arme. Le rythme du déploiement, quitte à raconter n’importe quoi sur les capacités réelles de la technologie, n’est pas accidentel : il est conçu pour devancer la réponse démocratique. C’est la théorie révoltante du “move fast and break things” sans aucune réflexion sur les conséquences ou la réversibilité de ce que l’on casserait. Voir notamment l’immense fiasco qu’a été DOGE – la tentative chaotique de Musk de 'disrupter' l'administration fédérale américaine, qui lui a donné un accès excessif et préoccupant aux données de l’Etat fédéral US, dont celles des entités de régulation de ses activités – et dont l’Amérique ne se remettra pas, notamment au niveau de la recherche en physique et en étude du climat.

• La question de la gouvernance est prioritaire. Avant de se demander “comment taxer l’IA ?” ou “comment réglementer l’IA ?”, nous devons nous demander : avons-nous encore la capacité politique de le faire ? Si ce n’est pas le cas, aucune proposition politique n’a d’importance.

• La résistance n’est pas du luddisme. Remettre en question la légitimité du déploiement de l’IA n’est pas s’opposer à la technologie. C’est questionner le projet politique que la technologie sert à faire avancer.

Les questions qu’il faut traiter

Dans le contexte, et pour continuer à se faire une idée de l’orientation et de la doctrine stratégique à adopter, il faut encore traiter au moins des questions suivantes:

1. Que signifierait et que permettrait le fait de gérer comme un commun le patrimoine intellectuel et la base de connaissance de l’humanité plutôt que de permettre son appropriation?

2. Comment les institutions démocratiques pourraient-elles réaffirmer leur capacité face aux efforts coordonnés visant à l’éroder?

3. Quelles formes d’action collective restent disponibles lorsque les outils de coordination sont eux-mêmes détenus par des ennemis de l’action collective? Pour être clair, il ne s’agit pas de promouvoir une forme nouvelle de collectivisme ou de communisme, mais simplement de reconnaître que l’espèce humaine est dépendante de sa capacité à organiser sa réalité collective parce que l’humanité est une forme de vie sociale et politique mal adaptée à la survie à l’échelle de l’individu.

4. L’ordre instauré après la Seconde Guerre mondiale mérite-t-il d’être défendu? Si ce n’est pas le cas, par quoi devrait-il être remplacé et qui en déciderait? Ici, l’objectif n’est pas de revenir à ce qu’était le monde d’après-guerre ou de dire que c’était mieux alors, mais bien d’exiger une discussion explicite et argumentée sur la direction que l’on prend au lieu de la subordonner à l’affirmation abusive selon laquelle l’innovation et le progrès technique sont synonymes de progrès humain. Ce n’est simplement pas vrai. Le monde romain, conquérant l’espace physique de la Grèce Antique, ne sut jamais aller au-delà de l’utilité fonctionnelle de la connaissance, ce qui fit son délitement car on n’abandonne pas impunément la science, la philosophie et les arts au profit exclusif de l’ingénierie, de la technique et des outils.

   

5. Quel “provisioning system” – au sens du papier O’Neill, D.W., Fanning, A.L., Lamb, W.F., & Steinberger, J.K. (2018). A good life for all within planetary boundaries. Nature Sustainability, 1, 88-95. – devrait être utilisé pour ces nouvelles capacités, qui devrait les régir et comment les bénéfices devraient-ils être répartis, étant donné que les communs en supportent les coûts, les risques et les conséquences?

Ce ne sont pas des questions rhétoriques. Ce sont les conditions d'une délibération que le rythme du déploiement est précisément conçu pour empêcher.

Décider dans ce contexte

Les cinq questions posées ci-dessus ne sont pas des sujets de réflexion académique. Ce sont des points de décision qui requièrent une délibération collective –précisément ce que le rythme de déploiement est conçu pour empêcher.

Avant de pouvoir trancher sur le fond, une décision préalable s’impose: acceptons-nous le cadre d’adaptation qu’on nous propose, ou exigeons-nous que la question de la gouvernance soit posée avant celle de l’intégration ?

Accepter le cadre d’adaptation, c’est décider – implicitement – que les conditions du déploiement seront fixées par ceux qui le conduisent. Refuser ce cadre, c’est décider que la légitimité démocratique précède l’efficacité technique.

Les critères d’évaluation de toute réponse devraient inclure:

• Gouvernance des communs : la proposition respecte-t-elle les principes d’Ostrom—limites claires, proportionnalité coûts/bénéfices, arrangements de choix collectif, surveillance, sanctions graduées, résolution des conflits, gouvernance polycentrique ?

• Adéquation du système de provisionnement : la proposition permet-elle de satisfaire les besoins humains fondamentaux dans le respect des limites planétaires, ou renforce-t-elle un système déjà en dépassement ?

• Préservation de la capacité démocratique : la proposition renforce-t-elle ou érode-t-elle la capacité des institutions démocratiques à gouverner effectivement ?

Toute proposition qui échoue sur ces trois critères – quelle que soit sa sophistication technique ou son caractère raisonnable apparent – constitue une capitulation déguisée en adaptation.

Agir dans ce contexte

L’action collective à l’échelle requise reste contingente aux décisions qui n’ont pas encore été prises collectivement. Les institutions capables de porter cette délibération sont précisément celles que le déploiement actuel contribue à affaiblir.

Cela ne signifie pas que toute action soit impossible. Mais cela impose une lucidité sur ce qui est actuellement à notre portée:

• Refuser le cadre narratif. Chaque fois que le déploiement de l’IA est présenté comme un fait accompli technique requérant adaptation, nommer ce cadrage pour ce qu’il est : une revendication politique déguisée en nécessité.

• Exiger la délibération. Insister pour que les questions de gouvernance soient posées avant les questions d’intégration — dans les organisations, les institutions publiques, les espaces de débat.

• Construire la capacité de réponse. Développer les compétences analytiques permettant de distinguer les propositions qui renforcent la capacité démocratique de celles qui l’érodent sous couvert d’adaptation.

• Documenter ce qui se passe. La légitimité future de toute réponse dépendra de la capacité à établir ce qui s’est passé, comment, et au bénéfice de qui. L’histoire ne s’écrit pas toute seule.

Ce ne sont pas des actions à la mesure de l’enjeu. Mais ce sont les conditions préalables à des actions qui pourraient l’être. La première forme de résistance consiste peut-être à insister pour que les bonnes questions soient posées – et à refuser de se laisser enfermer dans un cadre où elles ne peuvent pas l’être.

Calling technology “implementation detail” is not a neutral categorization. It’s a jurisdictional claim: these decisions are operational, therefore not ours. Yet in a digitized economy, the operational layer is precisely where strategy lives and value gets captured. It’s where lock-in gets chosen, and rules get written. When company boards treat system design as beneath strategic attention, they aren’t avoiding detail. They’re ceding rule-setting to whoever is closest to the machinery.

Crucially, they allow themselves to operate disconnected from reality – if not divorced from it – in the realm of static and outdated “strategic” frameworks that fail to recognize the actual situation of the business. Mental models that lack situational awareness, or canned thinking sold by international strategy consultancies with big names, expensive bills, and often lacking ethical compass – as demonstrated not only by the infamous Gaza evacuation plan, but also by the countless redundancies caused by failed transformation programs based on nonsense such as “the Spotify method”.

The usual diagnosis of that dysfunction of Boards and executives is kinder: it’s a skills gap, a generational lag, a training problem. Add a tech-savvy director. Host a deep-dive on AI. But that explanation has survived too many expensive transformations to remain credible. I made essentially this argument in 2016 in an article on LinkedIn. A decade later, the pattern persists – which tells us something. We’re not looking at ignorance. We’re looking at a governance arrangement. And like most stable arrangements, it serves someone’s interests.

Three interests, specifically.

First, a misconceived sense of “dignity”. Admitting ignorance in a boardroom is costly. Declaring the domain low-status and pushing it down the hierarchy is cheaper. The case of Kodak illustrates this: the failure to adopt digital was far more the consequence of activist investors interested in next quarter’s dividend than internal resistance to change—as Professor Albéric Tellier’s research demonstrates. The board’s strategic horizon had already been captured by interests that preferred not to understand what digital actually meant.

Second, tenure. Some executives optimize for being gone before impact is measured. Outsourced governance helps: when outcomes disappoint, blame routes elsewhere. Volkswagen’s diesel-gate is instructive—senior executives benefited from years of performance metrics inflated by under-priced environmental and regulatory risk. When the scandal broke, they initially blamed engineers and technical teams. The pattern: benefit from opacity, then route accountability downward. And of course the Enron case offers an equally disturbing cautionary tale about groupthink justifying unwarranted personal gains leading to disaster.

Third, surface modernization. That’s about pretending to work towards progress just enough so that one cannot be blamed, but not to the extent that progress would undermine incumbents and the establishment. Gramsci called it “passive revolution” – change that protects incumbents rather than displacing them. Vendor-captured transformation programs excel at this: new tooling, new dashboards, new labels, while decision rights and accountability structures remain untouched. If you want the appearance of adaptation without the burden of choice, that’s not a failure mode. It’s the product.

The second abdication: architecture boards and the belief forum

So the company board delegates. Where does that authority land?

Typically, in architecture boards – internal forums charged with technical governance, which are all too often seen as internal technical suppliers or facilitators when they should be business partners. And these forums face their own legitimacy problem. They decide based on diagrams that may not reflect the live system. They adjudicate competing beliefs about what exists, what connects to what, who owns which data. Half the room guesses. The other half nods politely. A decision is made anyway, because the calendar demands it.

This is not incompetence. It is an institutional equilibrium shaped by a structural fact: when answers are expensive – when it takes weeks to discover what a system actually does – organizations can only afford a small number of questions. Governance becomes triage. The architecture board exists in that flawed form because the organization lacks legible, low-friction access to its own reality.

But here’s the conflict of interest: architecture board members’ authority depends on that partial legibility. They are the translators, the arbiters, the people who “know the system.” Their institutional position requires that reality remain expensive to verify. They become brokers of a “truth” that is essentially their opinion, not verified fact about how the systems actually work.

This is the mentality of “IT as internal supplier” – a function that will do whatever it’s asked by “internal clients,” with almost no accountability for actual business performance. In most cases, the architecture board governs belief, not outcomes. And belief governance suits everyone whose authority would shrink if the system could simply answer back.

Resistance to legibility isn’t mysterious when you see the stakes and understand the conflict of interests. Scribes resisted public education because literacy was their franchise. Today’s scribes include anyone whose authority depends on keeping tech reality expensive to verify.

The commons failure underneath both

Neither board properly governs what actually needs governing: the internal digital commons.

Inside firms, architecture, data, and shared platforms behave like commons – multiple parties depend on them, can degrade them, and have stakes in how they’re managed. The failure mode is recognizably Ostromian: powerful appropriators, weak collective choice, inadequate monitoring.

Company boards like to believe their firms are different – private property, clear hierarchy, no tragedy of the commons. But the relevant resource isn’t legal ownership. It’s legibility and enforceability. Without monitoring, rules become aspiration. Without credible consequences, principles become posters.

The same governance failures that plague planetary commons – atmosphere, biodiversity, climate – replicate inside enterprises. The pattern is fractal. The costumes differ.

When legibility increases, both boards face legitimacy crises

Introduce a destabilizing capability: systems that can answer back. Views generated directly from code. Principles that become continuously verifiable. ‘Time to answer’ collapsing from weeks to hours. This is not speculative—almost two decades of research by Tudor Girba and colleagues into what’s now called moldable development have made it real.

The company board loses its excuse. “Implementation detail” stops working as a jurisdictional claim when the detail can be surfaced as a strategic reality with multimillion dollars worth of business consequences; when that “detail” can be visualized, and presented in terms the board cannot pretend are beneath them. The question becomes unavoidable: if you can see what the system does, what will you still claim isn’t your concern? Or your fiduciary duty as a board director, particularly if the function carries consequences on people, data, regulatory obligations or environmental commons.

The architecture board loses its monopoly. When belief adjudication is no longer necessary – when reality is directly accessible – the translator role shrinks and the natural role of business partner emerges in plain sight. The board must shift from arbitrating what is to governing what should be. That requires different competencies, different legitimacy, different accountability. Many incumbents won’t survive the transition unchanged.

The question both boards should sit with

None of this means boards must become technical. Vagueness or imprecision can be a rational response or a choice – honest ambiguity under irreducible uncertainty, preserving option value until the landscape clarifies. But vagueness is also a socially legible way to avoid accountability, especially when the cost of knowing current reality is kept high on purpose.

So: who benefits from not distinguishing option value from evasion? When a board stays “strategic” by staying vague, who gains room to maneuver – and who absorbs the consequences when the maneuvering fails? What kind of negative externalities are caused in this way? Which stakeholders are affected?

Company boards and architecture boards have different interests in partial legibility –but both have interests. The question is whether organizations can build governance forums whose legitimacy doesn't depend on partial legibility. Forums where the discussion happens in the boardroom, not in technical backrooms shaped by vendors selling “solutions”, “quadrants” and “waves”.

If answers about current reality become cheap, what will your company board still call “implementation detail”? What will your architecture board still claim requires their interpretation? And when those excuses disappear, what kind of governors will remain in the room?

A grasshopper is a small creature with a big lesson: stillness, then a leap. It reminds us to watch carefully, choose our moment, and move forward with confidence.

Anacridium aegyptium , the species on this picture taken in Peloponnesos, Greece, carries a quiet kind of strength: solitary, non-invasive, and peaceful.It doesn’t overwhelm its surroundings—it observes, settles, and moves on with lightness. A reminder for the new year: make progress without noise, grow without disruption, and choose bold steps with calm focus.

May the new year bring you all clear focus, bold steps, and the courage to jump when the path opens.

Cette première séance se fonde sur l’idée que la classe connaît a priori le monde du divertissement audiovisuel, même si ce n’est qu’en tant qu’audience de ce secteur. Partant de cette base, nous étudions 3 grandes phases de transformation qui feront émerger le Netflix intégré verticalement d’aujourd’hui à partir d’un acteur de la location de DVD au début du siècle. Nous explorons les transformations en suivant une approche structurée mais pas rigide, dont la philosophie est axée sur l’intelligence situationnelle pour ne pas borner la vision de ces évolutions à des schémas figés comme SWOT ou même les forces de Porter. Nous avons à faire à un monde complexe, dont plein de phénomènes d’émergence comme par exemple le comportement du binge watching, en mutation rapide et où il s’agit de ne pas se tromper sur ce que l’on considèrera comme zone stratégiquement vitale pour l’entreprise. Ainsi, nous nous appuyons sur:

1. le framework OODA pour la compréhension de l’environnement et l’orientation stratégique,

2. les business model canvas pour l’articulation des moyens et des promesses de marché autour de cet axe-clé de la proposition de valeur et

3. surtout sur un outil que j’emploie depuis plus de 10 ans dans mes activités professionnelles et dans les cours: les Wardley Maps,

4. le tout mis en musique par un processus-maison de transformation digitale, assez simple en apparence mais subtil dans son application, que j’ai développé depuis des années pour mes missions d’accompagnement d’organisations dans leur développement.

La puissance qu’offre cette approche crée par l’exceptionnel Simon Wardley est indescriptible. Je vous montre ici une représentation de la phase de transformation de Netflix où l’entreprise fait de sa maîtrise des données de comportement de ses audience par rapport aux contenus diffusés, l’atout stratégique pour remonter dans le système de valeur du divertissement audiovisuel cinématographique et de séries télé, et devenir ainsi un studio autant qu’une chaîne mondiale. On y voit une zone de protection stratégique que je voulais illustrer et discuter avec la classe, mais aussi certaines ressources-clés qui se présentent sous forme de pipeline comme les modèles de machine learning, déjà présent dès avant 2013 et très largement favorisé par l’existence d’une culture analytique forte, d’une importance accordée à la data et à sa gouvernance, ainsi que d’une culture du evidence-based management. Tout cela menant à une culture factuelle où on ne remporte pas un débat juste parce qu’on sait l’ouvrir un peu plus et de manière un peu plus assertive ou convaincante en groupe que le voisin – ce qui permet de rappeler cette brillante réflexion de Pierre Dac “parler pour ne rien dire et ne rien dire pour parler sont les deux principes majeurs et rigoureux de tous ceux qui feraient mieux de la fermer avant de l’ouvrir”. Revenons à nos moutons. Dans cette Wardley Map du Netflix qui est à la fois un studio de production audiovisuelle et une chaîne mondialisée, on voit aussi dans la zone de protection stratégique, un autre pipeline: le processus de production de contenus originaux, avec son modèle financier bien caractéristique et qui n’est pas celui d’un distributeur de contenus sourcés exclusivement auprès de tiers. Passionnants échanges ce matin avec beaucoup de belles contributions d’un groupe bien enthousiaste et engagé. Une joie!

Ainsi, nous voyons comment arriver à une compréhension intégrée de la transformation digitale qui:

1. élimine tout de suite la notion de silos pour regarder le business comme un réseau d’activités qui constitue un système complexe auto-adaptatif

2. amène naturellement à faire le lien entre la stratégie et la mise en œuvre pour que la première ne reste pas une incantation stratosphérique mais se traduise dans la réalité de terrain

3. oriente la transformation vers une approche de discovery-driven digital transformation, donc de transformation orientée par ce que l’on apprend à chaque itération, à chaque étape, à chaque initiative, en se donnant le droit de se louper mais pas celui de parier la boutique sur une initiative d’un “top manager” se disant “courageux”, mais ne réalisant pas toujours sa responsabilité première qui consiste à préserver, gérer et développer les communs de ce collectif qui s’appelle une entreprise (et donc, non, il n’a pas le droit de parier la boutique pour manifester son courage et son “leadership” - sorry)

Je suis incroyablement reconnaissant aux générations successives d’étudiants et en particulier à cette promotion qui s’engage avec enthousiasme et détermination dans des explorations qui leur montrent “how deep the rabbit hole goes” comme dirait Morpheus à Neo dans Matrix. Parce que je vous assure que quand on pousse l’analyse avec des méthodes fondées sur l’ouverture, la curiosité et l’intelligence situationnelle on obtient autre chose comme compréhension que quand on applique des recettes figées pensées par des génies loin du terrain.

Donc, oui, le mot du jour était kéfi. Kéfi d’avoir animé ce chouette cas dans cette magnifique école qu’est Solvay.

A fascinating research paper by Morteza Hemmati examines how linguistic signs in urban landscapes shape collective memory and identity. Hemmati argues that these symbols—whether ancient calligraphy on Persian architecture or modern emoji sculptures in public squares—represent more than wayfinding or branding. They are mechanisms through which communities negotiate meaning, assert identity, and exercise collective agency over shared space. That is definitely something that I will use in future classes about brands and branding, in order to show my students how brands are complete entities representing all the facets of a business, integrating its people, its partners, its clients, its ethics (a.k.a. practices, not just words on mugs) and its values. They are not logos or shallow visuals. They carry meaning and represent a contract between those who care about the symbol, which embodies the commons that belong to all the members of the community.

This observation opens a crucial question about the governance of our commons—not just the physical spaces we share, but the symbolic commons that give them meaning. When a city’s residents converge around a symbol, they are essentially performing an act of commons governance: defining boundaries, establishing rules, monitoring compliance, and sanctioning violations. These are precisely the design principles Elinor Ostrom identified in successful commons regimes.

But here’s the paradox: while we’ve become sophisticated at understanding how to govern tangible commons, such as forests, fisheries, irrigation systems, even though we still tend to only see them as either private or in the hands of government, we’re only beginning to grapple with how to govern symbolic and digital commons. How do communities maintain their right to define what their shared spaces mean? Who decides which symbols occupy the most influential locations in our urban landscapes? Who has the rights to them and what is the appropriation of value from physical space?

The stakes extend far beyond aesthetics. As we confront planetary boundaries that require unprecedented collective action, our capacity to create and defend shared meaning-making spaces becomes existential infrastructure. Climate adaptation, democratic resilience, and social cohesion all depend on communities’ ability to assert collective identity and purpose through their commons—both physical and symbolic.

The article notes how simplified, media-driven symbols (emojis, hashtags, imported branding formulas) have displaced more complex, culturally-rooted expressions in urban landscapes. This isn’t merely aesthetic loss—it’s a degradation of the commons’ capacity to support the rich, contextual meaning-making that diverse communities require.

When New Yorkers gather around their city’s iconic symbol today, they’re not just expressing local pride. They’re demonstrating something fundamental about commons governance: that those who inhabit shared spaces retain the authority to define what those spaces represent. That legitimacy flows from participation, not imposition. That symbols in public spaces belong to publics—not to transient authority.

This matters because every major challenge we face—from climate to governance to digital transformation—requires us to act collectively while respecting diversity. We need institutional frameworks that can hold both coherent shared purpose AND legitimate dissent. We need commons that can support both belonging AND contestation.

The urban landscape, with its symbols and signs, is where we practice this daily. It’s where abstract principles of commons governance become concrete, visible, lived. It’s where we learn—or fail to learn—how to steward shared resources while honoring the plural meanings different communities bring to them.

Perhaps that’s why a simple sign saying “I ♥ NY” can carry such weight: it reminds us that the commons is not just what we share, but how we choose to govern what we share. And that choice remains ours to make.

Their central thesis rests on Coase’s insight that transaction costs explain firm boundaries. Since AI agents can search, negotiate, and contract at near-zero marginal cost, they suggest we’re heading toward fundamental restructuring of market organization—what I’ll call their implicit “Coasean singularity,” where transaction costs drop so low that the logic justifying firms becomes obsolete.

We’ve been here before.

The E-Commerce Déjà Vu

In the early 2000s, similar prophets proclaimed e-commerce would eliminate “friction” and deliver us to a “frictionless economy.” Management consultants at prestigious firms assured their banking clients that physical branches were dinosaurs destined for extinction. The logic was impeccable: why maintain expensive real estate when customers could transact online at negligible cost?

Fast forward to 2025. Friction hasn’t disappeared—it’s migrated to different parts of the value chain. Last-mile delivery logistics. Returns management. Customer service for complex products. And crucially, those physical bank branches? Still here. Still necessary. Because certain transactions requiring human perception, expertise, relationship-building, and trust cannot be replaced by clicking buttons on screens.

The NBER authors make precisely the same category error, just with shinier technology. They acknowledge some new frictions (congestion, price obfuscation) but treat these as manageable externalities rather than fundamental features of how complex systems actually work.

What Coase Actually Taught Us (And What Economists Forgot)

Here’s what’s maddening about invoking Coase to predict firm dissolution: Coase himself understood that transaction costs were just one reason firms exist. The real picture is far richer.

Firms exist because:

• Knowledge integration requires sustained collaboration that markets cannot coordinate

• Complementary assets need protection from holdup problems

• Dynamic capabilities emerge from organizational learning, not spot transactions

• Relational contracts depend on repeated interaction and shared context

• Power and coordination matter in ways that cannot be reduced to “friction”

Beyond the work of Coase on the factors making firms necessary in the economy, we can also turn to other economists like Elinor Ostrom who examined how common pools of resources can be managed, challenging the idea of inevitable demise through the proverbial “tragedy of the commons”. Indeed, if we consider the whole economy as a commons – a giant pool of resources being managed collectively by those active within it – then the mechanisms that emerge from that complex system, including firms and other collectives, are not inefficiencies to be optimized away but institutional solutions whose form follows their function. Ostrom, in her research on commons governance, demonstrated that successful collective action requires institutions, rules, boundaries, and trust that emerge from sustained relationships—precisely what firms provide. You cannot replace this with efficient API calls between AI agents, no matter how sophisticated your natural language processing.

The paper treats trust as if it’s merely an information problem. Use AI to verify identity! Deploy cryptographic proofs! But trust isn’t about perfect information—it’s about accountability, governance, and the capacity to sanction bad actors. AI agents introduce new principal-agent problems. Who controls the agent? Whose interests does it serve? How do we audit its decisions? These questions require organizational structures, not better algorithms.

The Material Reality Economists Ignore

But the deepest flaw runs even further. There is a breathtaking passage where the authors write: “The activities that comprise transaction costs—learning prices, negotiating terms, writing contracts, and monitoring compliance—are precisely the types of tasks that AI agents can potentially perform at very low marginal cost.”

Very low marginal cost.

Let us be precise about what “low marginal cost” AI actually requires:

• Data centers consuming electricity equivalent to small nations. Ireland’s data centers already account for 21% of national electricity demand. By 2030, AI infrastructure could consume 10% of total U.S. electricity.

• Water for cooling—millions of gallons per data center annually. In water-stressed regions, this directly competes with human needs and agriculture.

• Rare earth minerals for chips and hardware, extracted through environmentally devastating mining processes, often in geopolitically fragile regions with minimal labor protections.

• Electronic waste from hardware obsolescence. The current AI boom is accelerating replacement cycles, generating mountains of toxic e-waste.

• Embodied energy in manufacturing semiconductors—some of the most energy-intensive industrial processes humanity has invented.

One needs to be an economist with no understanding of matter, energy, thermodynamics, or planetary boundaries to imagine we could reach a “singularity” beyond which Coase’s insights become obsolete. We live in a finite world governed by physical laws. Computation requires energy. Energy transformation generates entropy. Resources deplete. These constraints don’t disappear because Silicon Valley built a better autocomplete.

The authors mention “compute costs” exactly once, treating it as a pricing variable rather than a biophysical constraint. This is economics in a bottle—models that work beautifully on whiteboards but collapse when they encounter reality.

The Real Transformation (Which Economists Won’t Like)

Here’s what will actually happen as AI capabilities expand:

1. Friction migrates, doesn’t disappear. Just as e-commerce moved friction from browsing to logistics, AI will shift costs from some transaction activities to others—verification, governance, dispute resolution, dealing with AI-generated spam and gaming.

2. Trust problems multiply. Every AI agent introduces new information asymmetries and alignment challenges. Who audits the agents? How do we prevent manipulation? These require human judgment and institutional oversight—exactly what firms provide.

3. Rebound effects dominate. Lower costs don’t necessarily mean lower total impact. Cheaper AI negotiations might flood markets with applications, creating congestion that requires new filtering mechanisms (which impose their own costs and biases). The paper acknowledges this for job applications but treats it as a curiosity rather than a fundamental dynamic.

4. Physical constraints bind harder. As AI use scales, energy and material limits will impose increasingly severe costs. These won’t show up in your transaction cost calculations until they manifest as price spikes, rationing, or regulatory intervention.

5. New organizational forms emerge—but they’re still firms. Companies will restructure around AI capabilities, but they’ll remain organizations with boundaries, governance structures, and collective capabilities. The “make or buy” boundary shifts; it doesn’t evaporate.

Economics Without Physics Is Just Politics and Storytelling

The broader problem this paper exemplifies is mainstream economics’s persistent denial of biophysical reality. When your models treat computation as costless information processing divorced from its material substrate, you can generate any conclusion you want. It’s the same intellectual move that gives us infinite growth models on a finite planet, externalization of environmental damage as “market imperfections,” and the persistent fantasy that technology will deliver us from resource constraints through pure cleverness.

This isn’t science. It’s theology dressed in mathematics.

Real science acknowledges constraints. Thermodynamics. Materials science. Ecology. Systems dynamics. These disciplines understand that you cannot optimize around physical limits—you must design within them. Circular value networks. Life-cycle assessments. Planetary boundaries frameworks. These are the tools we need.

Instead, we get papers celebrating AI’s potential to reduce transaction costs while ignoring that the infrastructure enabling those transactions is accelerating us toward ecological catastrophe.

What We Should Be Asking Instead

If we took biophysical constraints seriously, here are the questions economists should investigate:

• What is the energy and material throughput per AI-mediated transaction compared to human-mediated alternatives?

• At what scale do energy demands from AI infrastructure create systemic risks to grid stability and climate goals?

• How do we design economic institutions that internalize the full life-cycle costs of AI systems?

• What governance structures ensure AI deployment serves genuine human needs rather than generating artificial demand?

• How do we prevent AI capabilities from being captured by concentrated economic power?

These questions don’t appear in the NBER paper because they would undermine its central narrative. They require acknowledging that technology choices have consequences, that efficiency isn’t the only value that matters, and that markets embedded in physical reality face constraints that no amount of computational power can dissolve.

Conclusion: Singularities and Other Fairy Tales

The concept of a “singularity” – whether Coasean or technological – is revealing. It describes a point where models break down because they’ve been extrapolated beyond their domain of validity. Physicists use it to mark the boundaries of their theories, admitting ignorance. Economists and technologists use it to mark the arrival of transcendence, claiming knowledge of transformation so profound that existing rules no longer apply. Maybe principally because ignoring those rules is a fundamental prerequisite for kicking the can down the road.

The difference is instructive.

This paper offers sophisticated analysis of how AI might reshape specific market mechanisms. Some of those insights are valuable. But wrapped around that analysis is a story about frictionless efficiency and dissolved organizational boundaries that ignores material reality, misunderstands what firms actually do, and extrapolates marginal cost curves into fantasy futures.

We don’t need more papers celebrating AI’s potential to optimize markets. We need economics that acknowledges we’re running planetary systems on finite energy and materials, that efficiency gains often generate rebound effects that swamp direct benefits, and that the most important question isn’t “how can AI reduce transaction costs” but “what kind of economy can survive within biophysical limits while providing decent lives for billions of people?”

That question requires different tools than the ones deployed here. It requires systems thinking, ecological economics, commons governance principles, and serious engagement with the material basis of all economic activity. It requires, in short, abandoning the fantasy that clever algorithms can dissolve the constraints that actually matter.

The Coasean singularity isn’t coming. But the planetary boundaries crisis is already here. Perhaps economists might consider which problem deserves their attention.

Today, loyalty programs illustrate the broader logic of digital transformation: they turn passive consumers into active participants in data-rich, co-created experiences—where personalization, real-time analytics, and ecosystem interoperability drive brand value.

1. From Plastic Cards to Experience Platforms

Digitalization has transformed loyalty programs from transactional systems into experience-driven ecosystems. Where physical cards once tracked purchases in isolation, mobile apps and digital wallets now unify payments, personalized rewards, and engagement journeys.

Research shows that digital wallets and loyalty apps increase participation rates by over 40% compared to card-based systems, thanks to higher usability and their constant presence in consumers’ digital lives. These apps generate richer data streams—purchase histories, locations, behavioral signals—that enable personalized promotions, predictive analytics, and context-sensitive communication. See sources [1][2] listed at the end of the article.

This evolution marks a shift from tracking transactions to orchestrating experiences, making loyalty a continuous dialogue between brand and customer.

2. The Headless Loyalty Architecture

A key infrastructure enabler of this shift is headless loyalty: the decoupling of the loyalty “engine” (rules, data, and transactions) from the customer-facing interfaces. In headless systems, loyalty functions operate through APIs that connect to websites, mobile apps, kiosks, social platforms, or even third-party ecosystems. This allows loyalty to extend far beyond the traditional point-of-sale.

Research and industry analyses identify several advantages:

• Flexibility – Brands can tailor experiences to each channel without back-end redesign.

• Speed and scalability – New reward mechanisms, tiers, or campaigns can launch rapidly, reducing technical debt and accelerating innovation.

• Omnichannel continuity – The same loyalty logic powers multiple touchpoints, from e-commerce and in-store to IoT contexts such as connected vehicles.

• Cost efficiency and interoperability – API-first design enables integration with CRM, ERP, and e-commerce systems, building unified customer data ecosystems.

Headless architectures also underpin modular “loyalty-as-a-service” models, embedding loyalty capabilities within broader digital transformation frameworks rather than isolating them as marketing add-ons. In short, headless design transforms infrastructure into innovation capacity.

To go deeper into this aspect see sources [3][4][5][6] and [7] listed below.

3. AI and Data-Driven Personalization

Digital transformation has shifted loyalty from points collection to intent prediction. Using AI and machine learning, modern programs analyze real-time customer data to anticipate needs and recommend individualized rewards.

Programs leveraging predictive analytics report up to 25% higher redemption rates and significantly lower churn. These systems blend first-party, behavioral, and contextual data—anchored in privacy-by-design and consent-centric frameworks compliant with evolving regulations like GDPR. See sources [8][9]

In this model, data becomes both the fuel and the currency of loyalty. Customers trade insights for relevance, and brands earn trust through transparency and responsible data governance.

4. Beyond the Point of Sale

Digitally transformed loyalty programs now extend beyond transactions to reward participation and advocacy. Users earn value not just for purchases but for social sharing, reviews, sustainable actions, or community engagement.

By embedding loyalty in social and lifestyle contexts, brands create continuous engagement loops where customers co-create meaning and momentum. Loyalty thus evolves from a reward mechanism into a vehicle for sustained brand interaction—driving emotional connection and network effects.

Some brands now integrate sustainability behaviors into loyalty (e.g., recycling, responsible consumption), turning purpose-driven engagement into measurable value creation.

For more see sources [10][2][11] below.

5. Analytical Framework for Studying Transformation

The digital transformation of loyalty programs can be examined along four key axes:

1. Architectural dimension – Comparing API-driven, headless, or modular systems against monolithic legacy architectures.

2. Data intelligence – Evaluating how analytics, AI, and automation enable predictive engagement and adaptive learning.

3. Customer experience design – Assessing omnichannel consistency, emotional resonance, and user-centric personalization.

4. Business integration – Analyzing how loyalty platforms connect with CRM, pricing, and digital commerce within enterprise transformation agendas.

Loyalty now acts as a sort of continuously adaptive customer interaction layer — a connective tissue linking personalization, automation, and value co-creation within the broader digital enterprise.

See sources [12][13]

6. The emergence of adaptive experience ecosystems

In essence, the digital transformation of loyalty programs redefines them as adaptive experience ecosystems. What began as plastic-card schemes has evolved into data- and API-native loyalty platforms—capable of omnichannel continuity, emotional engagement, and rapid innovation through headless architecture.

Loyalty today is not merely a retention tool but a strategic layer of the digital customer experience, driving personalization, data intelligence, and co-created value. As AI, IoT, and sustainability metrics converge, loyalty programs are poised to become the operating system of customer relationships in the digital economy.

[1]https://paysimple.com/blog/the-evolution-of-customer-loyalty-programs/

[2]https://www.cmswire.com/customer-experience/the-evolution-of-customer-loyalty-programs-in-an-always-on-world/

[3]https://www.trueloyal.com/blog/power-of-headless-loyalty-program-software/

[4]https://www.leat.com/resources/blog/headless-loyalty

[5]https://www.openloyalty.io/insider/headless-loyalty-software

[6]https://www.currencyalliance.com/insights/why-headless-loyalty-is-better-for-customer-engagement

[7]https://antavo.com/blog/headless-loyalty-program/

[8]https://socialfirm.com/blog/the-evolution-of-customer-loyalty-programs-in-2026/

[9]https://www.annexcloud.com/blog/3-ways-loyalty-enables-your-brands-digital-transformation/

[10]https://www.linkedin.com/pulse/how-digital-transformation-reshaping-loyalty-programs-vlfnf

[11]https://blog.brandmovers.co.uk/digital-evolution-of-consumer-loyalty-marketers-guide

[12]https://www.bcg.com/publications/2024/loyalty-programs-customer-expectations-growing

[13]https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/members-only-delivering-greater-value-through-loyalty-and-pricing

[14]https://www.iseatz.com/blog/what-digital-transformation-looks-like-for-large-loyalty-programs

[15]https://www.deloitte.com/us/en/services/consulting/articles/brand-loyalty-program-consumer-behavior.html

[16]https://perksprofessor.com/the-future-of-loyalty-programs-in-the-digital-age/

[17]https://www.datainsightsmarket.com/reports/digital-loyalty-program-1986372

[18]https://www.magestore.com/blog/the-rise-of-omnichannel-loyalty-programs/

[19]https://www.openloyalty.io/insider/omnichannel-retail-loyalty-programs

[20]https://www.nector.io/blog/9-successful-loyalty-program-examples-to-learn-from-in-202

On October 13, 2025, the Royal Swedish Academy of Sciences awarded the Nobel Prize in Economics to Joel Mokyr, Philippe Aghion, and Peter Howitt “for having explained innovation-driven economic growth.” According to the prize committee, these laureates showed “how new technology can drive sustained growth” and demonstrated that “this is the basis for sustained economic growth, which results in a better standard of living, health and quality of life for people around the globe.”

The word “sustained” appears five times in the brief press release. Each use rings with confidence about humanity’s capacity for endless expansion. There’s just one problem: we have overwhelming evidence that this model of growth cannot be sustained. The data isn’t ambiguous. The science isn’t uncertain. We’re not talking about contested predictions—we’re witnessing measured reality.

What “Sustained” Actually Means

The Nobel committee uses “sustained” to mean self-perpetuating—a mechanism by which technological innovation generates continuous economic expansion, breaking the historical pattern of stagnation. They’re celebrating an explanation of how the last 200 years happened, not examining whether it can continue.

This is like giving an award in 1910 for brilliantly explaining the physics of how the Titanic can stay afloat, while ignoring that it’s already taking on water.

The Evidence We’re Ignoring

The Original Warning: Limits to Growth (1972)

Fifty-three years ago, a team led by Donella Meadows at MIT published The Limits to Growth, commissioned by the Club of Rome. Using the World3 computer model, they examined five variables that determine and ultimately limit growth: population, industrial production, food production, resource consumption, and pollution.

Their conclusion was unequivocal:

“If the present growth trends in world population, industrialization, pollution, food production, and resource depletion continue unchanged, the limits to growth on this planet will be reached sometime within the next one hundred years. The most probable result will be a rather sudden and uncontrollable decline in both population and industrial capacity.”

Critics dismissed the work as alarmist. Economists argued that human ingenuity and technological innovation would always overcome resource constraints—a perspective known as “cornucopianism.” The report was systematically misrepresented, with opponents claiming it predicted specific dates for resource exhaustion (it didn’t) and ignoring its core message about system dynamics and overshoot.

The Uncomfortable Validation

Here’s what’s stunning: the original model’s predictions have tracked reality with disturbing accuracy.

In 2008, physicist Graham Turner at Australia’s CSIRO compared thirty years of real-world data against the Limits to Growth scenarios. His finding? The “business as usual” scenario—which leads to collapse in the 21st century—matched actual developments most closely.

In 2021, Gaya Herrington, then Director of Sustainability Services at KPMG, published an updated analysis in Yale’s Journal of Industrial Ecology. Using data through 2020, she found that empirical evidence continued to align most closely with the Business as Usual 2 (BAU2) scenario—which shows decline beginning around 2030.

Most recently, in 2024, a team led by Arjuna Nebel recalibrated the entire World3 model with the best available contemporary data. Published in the Journal of Industrial Ecology, their work concluded: “The fact is that the recalibrated model again shows the possibility of a collapse of our current system. At the same time, the BAU scenario of the 1972 model is shown to be alarmingly consistent with the most recently collected empirical data.”

The Meadows team didn’t predict some distant hypothetical future. They created a model of the system dynamics of exponential growth on a finite planet. And the data shows they got it right.

The Planetary Boundaries: Quantifying What We’ve Already Broken

While the Limits to Growth work focused on system dynamics, the planetary boundaries framework—developed by Johan Rockström and colleagues at the Stockholm Resilience Centre—provides specific thresholds for nine critical Earth system processes.

The 2023 update, published in Science Advances, quantified all nine boundaries for the first time. The 2025 update confirms an accelerating deterioration of the situation with 7 out of 9 planetary boundaries breached and our ecosystem more than ever in unsafe territory.

The Stockholm Resilience Centre is clear about what transgressing these boundaries means: “Crossing boundaries increases the risk of generating large-scale abrupt or irreversible environmental changes.” They emphasize that “planetary boundaries are interdependent”—we cannot consider them in isolation because “only by respecting all nine boundaries can we maintain the safe operating space for humanity.”

What the Nobel Prize Actually Celebrates

To be clear: the laureates’ work is probably sophisticated and valuable. Very few among us have the level of mastery needed to assess it, especially in a discipline that is as much science as it is a construct built atop beliefs and a set of political and philosophical axioms. Some examples: the rationality of economic agents vs behavioral bias, the obsession of financial returns, efficiency and performance vs the mission to sufficiently satisfy human needs, the qualitative equivalence of equal quantities of individual and collective work vs differentiated value for work of collectives vs solo player, the assumption of separation from biosphere vs the view that the economy is a subsystem of the larger system called Earth… etc

Understanding the mechanisms of innovation-driven growth matters. Mokyr’s historical analysis of how scientific understanding enables cumulative technological progress is important. Aghion and Howitt’s models of creative destruction illuminate real dynamics in market economies.

But celebrating this work in 2025—with the framing of “sustained growth” as humanity’s great achievement—is at the very least misguided. The Nobel committee’s statement reveals the discipline’s foundational blind spot. They write: “Technology advances rapidly and affects us all, with new products and production methods replacing old ones in a never-ending cycle. This is the basis for sustained economic growth, which results in a better standard of living, health and quality of life for people around the globe.” (sic)

Notice what’s absent: any acknowledgment that this “never-ending cycle” operates within a finite biosphere. No mention of the material throughput required for technological innovation. No recognition of the waste streams, the resource depletion, the ecological disruption that accompanies every cycle of creative destruction.

The committee even acknowledges history: “Quite the opposite—stagnation was the norm throughout most of human history.” This is presented as a problem we’ve solved, not as the baseline condition of an economy operating in the pre-industrial reality.

The Thermodynamic Reality Economics Ignores

Here’s the uncomfortable truth that economic models consistently exclude: you cannot have infinite material throughput on a finite planet. This isn’t ideology—it’s physics.

Every cycle of creative destruction requires energy and materials as inputs and generates waste as output. Innovation doesn’t transcend these constraints; it merely shifts which materials we extract and where we dump the waste. Replacing fossil fuels with renewable energy doesn’t eliminate the need for rare earth elements, lithium, copper, and the massive infrastructure to deploy them. Creating digital economies doesn’t dematerialize growth—it requires data centers, semiconductors, and electricity generation at scale.

Nicholas Georgescu-Roegen, one of the rare economists who saw the dependency of the economy on natural resources and the broader environment, often considered to be the founder of ecological economics, explained this in 1971 with the second law of thermodynamics: economic processes inevitably increase entropy. The “sustained growth” the Nobel committee celebrates is thermodynamically impossible as a permanent condition.

The planetary boundaries framework quantifies exactly this reality. We’re not approaching limits in some distant future—we’re already deep in overshoot on six of nine critical thresholds.

What Sustained Actually Means

If we’re going to use the word “sustained” seriously in 2025, we need to ask: sustained for whom? For how long? And at what cost?

The growth of the past two centuries has been “sustained” only in the narrowest temporal sense—a 200-year pulse of expansion fueled by fossil energy and enabled by treating the planet as a free waste dump. This is not sustainability; it’s liquidating natural capital and calling it income.

Real sustainability means operating within planetary boundaries. It means recognizing that the economy is a subsystem of the biosphere, not the other way around. It means understanding that technology is a tool for human purposes, and if our purpose remains endless material expansion, our tools will serve that suicidal goal efficiently.

The Limits to Growth authors wrote in 1972: “The challenge of overshoot from decision delay is real, but easily solvable if human society decided to act.” They were offering a choice.

Fifty-three years later, with seven planetary boundaries transgressed and the evidence of overshoot overwhelming, the economics profession is still celebrating the mechanisms that got us here.

As written in a previous article, on October 13 1307, French King Philippe Le Bel thought he’d made a gain by seizing the assets of the Templars. In an analogous way of thinking, the Nobel committee seems to believe innovation has solved scarcity. Both ignore the systematic destruction of the commons on which everything else depends.

The door is closing. The veil is descending. The dialogue between economics and ecology—between our theories of prosperity and the biophysical systems that enable life—is ending. We (still) have a choice: preserve the commons, or suffer the chaos and destruction that follows their demise.

Sources:

Richardson, J., et al. (2023). “Earth beyond six of nine Planetary Boundaries.” Science Advances, 9(37).

Herrington, G. (2021). “Update to limits to growth: Comparing the World3 model with empirical data.” Journal of Industrial Ecology, 25(3), 614-626.

Nebel, A., et al. (2024). “Recalibration of limits to growth: An update of the World3 model.” Journal of Industrial Ecology, 28(1), 87-99.

Meadows, D.H., Meadows, D.L., Randers, J., & Behrens, W.W. (1972). The Limits to Growth. Universe Books.

Meadows, D.H., Randers, J., & Meadows, D.L. (2004). Limits to Growth: The 30-Year Update. Chelsea Green Publishing.

Turner, G. (2008). “A comparison of the Limits to Growth with thirty years of reality.” CSIRO.

Georgescu-Roegen, N. (1975). Energy and Economic Myths. Southern Economic Journal, 41(3), 347–381. https://doi.org/10.2307/1056148

Stockholm Resilience Centre. (2023). “Planetary Boundaries.” Available at: https://www.stockholmresilience.org/research/planetary-boundaries.html

Stockholm Resilience Centre. (2025). Planetary Health Check 2025, available at: https://www.planetaryhealthcheck.org/#reports-section

The Royal Swedish Academy of Sciences. (2025). “The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2025.” Press release. Available at: https://www.nobelprize.org/prizes/economic-sciences/2025/press-release/

These projections are presented with the clinical precision we've come to expect from the IEA—detailed supply curves, technology deployment scenarios, and infrastructure requirements mapped out with impressive granularity. But something fundamental is missing from this 200-page analysis: any serious grappling with why this trajectory deserves our scarce resources, or whether it aligns with genuine human priorities.

The legitimacy gap

The report's headline figure—that data centers currently consume ~1.5% of global electricity—obscures more than it reveals. This seemingly modest percentage masks acute local impacts and says nothing about opportunity costs. When we dedicate enormous quantities of clean electricity to training the latest large language model, what health clinics, electric vehicle charging networks, or industrial decarbonization projects go without power?

More critically, the IEA's electricity-focused lens misses the broader environmental picture entirely. Mistral AI's groundbreaking lifecycle assessment, published in collaboration with Carbone 4 and France's ecological transition agency (ADEME), reveals the true scale of AI's resource footprint. Training their Large 2 model consumed 20,400 tonnes of CO₂ equivalent, 281,000 cubic meters of water, and 660 kg of scarce materials (antimony equivalent)—impacts that go far beyond the electricity meter.

These "upstream emissions" from manufacturing servers, mining rare earth elements, and cooling systems represent a substantial portion of AI's total environmental burden. Yet the IEA's analysis treats electricity consumption as if it operates in isolation from these material and water impacts.

The IEA treats AI electricity demand as weather—something to be predicted and accommodated rather than evaluated and potentially redirected. This technocratic framing reflects a deeper intellectual failure: confusing what can be built with what should be built.

Consider the report's assumption that current growth trajectories will continue largely unabated. It models scenarios where data center demand reaches ~1,200 TWh by 2035—nearly triple today's levels. But these projections rest on questionable premises: that compute-intensive applications will continue expanding regardless of social utility, that efficiency gains won't meaningfully slow demand growth, and that society will indefinitely prioritize AI scale over other pressing electrification needs.

More troubling is what the analysis omits entirely. There's:

• no framework for distinguishing between AI applications that serve fundamental human needs and those that don't. Fundamental human needs have best been studied in Manfred Max-Neef’s work, esp. the human development framework (see this page on Wikipedia for a summary, but also read Max-Neef’s work)

• no acknowledgment that we're operating within planetary boundaries that constrain how much energy we can responsibly consume and therefore imply that we need to prioritise the use of renewable energy and eliminate the need for fossil fuels.

The four-gate test for energy allocation to AI

The goal is ensuring each marginal kilowatt-hour—and each kilogram of scarce materials—spent on AI delivers more health, safety, resilience, and decarbonization than it consumes. I propose a four-gate test that any major AI facility should pass before plugging into the grid:

Gate 1: additional, hourly-matched clean power New AI load must be met by genuinely additional low-emissions supply, matched hour by hour. Annual renewable energy credits don't count—we need 24/7 clean firm electricity that doesn't cannibalize clean power needed elsewhere. This means campus-level clean energy disclosure, power purchase agreements that clearly add capacity to the same grid, and firm supply arrangements consistent with decarbonization trajectories.

Gate 2: non-crowding of local grids The IEA acknowledges that ~20% of planned data center capacity faces grid constraint delays, yet doesn't interrogate what this means for competing priorities. Projects should strengthen local grids where they land—not strain them. This requires independent interconnection studies showing no net degradation of local reliability, plus funded grid-benefit packages (dynamic line rating, flexibility services, storage) sized to the project's peak load.

Gate 3: social-utility commitment Here's where we separate genuinely valuable AI from digital junk food. Dedicate a meaningful share of compute capacity (≥30-50%) to workloads with high human-needs value: grid reliability, healthcare, building efficiency, disaster response. The IEA's own data shows AI can reduce power outage duration by 30-50% and improve grid stability—this is the kind of measurable public value to prioritize.

Gate 4: full-lifecycle stewardship the Mistral AI study makes this gate especially urgent. Electricity-related emissions are just one piece of AI's environmental puzzle. The 660 kg of scarce materials needed to train a single large model, the 281,000 cubic meters of water consumed, and the massive upstream emissions from hardware manufacturing demand comprehensive stewardship commitments.

Operators should commit to longer server lifetimes, refurbishment programs, high-recovery recycling of critical materials, and waste heat reuse where feasible. Liquid cooling increasingly enables 40-80°C thermal flows suitable for district heating—often cheaper than fossil combined heat and power. Given that material depletion accounts for a significant portion of AI's total impact, circular economy principles aren't optional—they're essential.

Why this matters now?

The AI industry's favorite defense—"it's only 1-3% of global electricity"—crumbles under scrutiny. Mistral AI's recent data shows this percentage dramatically understates AI's true environmental burden when you include water consumption, material depletion, and upstream emissions. Local impacts matter enormously, and that percentage is growing fast. More fundamentally, this framing treats electricity as if it's abundant rather than precious.

The lifecycle assessment also reveals a stark correlation between model size and environmental impact—impacts scale roughly proportionally with model parameters. This underscores the importance of matching model scale to actual use cases rather than building ever-larger systems as a matter of course.

We're living through the most important energy transition in human history, racing to decarbonize before climate breakdown becomes irreversible. Every kilowatt-hour matters. Every kilogram of rare earth elements matters. Every cubic meter of freshwater matters. Every siting decision shapes whether communities get reliable clean power or watch AI facilities consume the renewable energy meant for their schools and hospitals.

The IEA's report provides essential technical groundwork for this transition. But technical feasibility isn't ethical legitimacy. Just because we can build massive AI infrastructure doesn't mean we should—not without first asking whether it serves genuine human flourishing better than the alternatives.

The four-gate test offers a practical framework for making these choices responsibly. It's not anti-innovation—it's pro-prioritization. It rewards efficient models, flexible operations, and high-utility applications while screening out digital waste.

As AI reshapes our energy landscape, we need policies that put human needs and planetary limits ahead of undifferentiated scale. The stakes are too high for anything less.