Chapter 1 — Introduction (plain-language version)
The puzzle
Here is a moment that captures the problem this thesis is about.
A junior consultant on a railway project in Portugal produces the risk section of a client presentation. The writing is polished. The structure is clear. The regulatory categories are named correctly. But the senior partner reads it and stops.
The section treats the main risk as a routine contract-renewal question — textbook, defensible, wrong. The senior partner had spent three meetings reading the signals from the Portuguese rail regulator: the real concern was about timing, specifically how a contract renewal would interact with a separate parliamentary inquiry the regulator couldn’t name openly but kept circling back to with its questions. The junior consultant hadn’t been in those meetings. He had used an AI tool to draft the section from the brief.
The output was good enough to survive a first read. It was not good enough to survive a read by someone who had actually been in the room.
The partner corrects the section, says nothing, and then spends two weeks thinking about what this means. The junior couldn’t have known what he didn’t know. The AI tool couldn’t have known it either. And the quality-control process had nearly missed it.
That moment is small. What it points to is not.
This thesis is about a gap between two ways of talking about what AI tools are doing to professional consulting work.
In the first version — the one you hear at conferences, in marketing materials, from management consultants — AI tools are mainly a productivity story. Drafting goes faster. Research gets done in minutes instead of hours. Slides get polished more efficiently. The tools are useful and manageable, and the main question is how to deploy them sensibly.
In the second version — the one practitioners talk about in less guarded conversations — something more unsettling is happening. The junior consultants coming up now produce better-looking work than their predecessors did at the same stage, but their grasp of the work is sometimes thinner than the polish suggests. Senior consultants worry about how the next generation is being formed. Clients are changing how they buy consulting services. And underneath all of it, something in the texture of the profession feels like it is shifting in ways the productivity story doesn’t capture.
This thesis takes the second version seriously. It argues that what practitioners are sensing is real, that it matters, and that understanding it requires a different kind of framework than the existing literature provides.
The framework at the core of this thesis is philosophical. Its two main pillars are Bernard Stiegler, a French philosopher who argued that expertise gets progressively absorbed into technical systems under specific economic conditions, and Hans Jonas, a philosopher who argued that the ability to act at technological scale creates obligations to future generations. Supporting these two are Alasdair MacIntyre, who wrote about the values that are internal to a professional practice and that can be corrupted when a practice is degraded; Michael Polanyi and Harry Collins, who wrote about tacit knowledge — the expertise that practitioners cannot fully articulate but that distinguishes a senior from a novice; and John Dewey, an American philosopher who argued that the people affected by consequential decisions form a public with a legitimate stake in those decisions, even when they aren’t party to them. Each of these thinkers does specific work at specific points in the argument.
What the thesis is called, and why each word matters
The full title is: The Proletarianisation of Advisory: Judgment, Sovereignty, and Responsibility in European Infrastructure Consulting under AI Mediation.
Proletarianisation is a word from the philosopher Bernard Stiegler, who reworked it from its origins in Marx. It doesn’t mean job losses here. It means something more specific: the gradual transfer of practical know-how — and, beneath that, the know-how-to-live and ultimately the know-how-to-think — out of the people doing the work and into technical systems they don’t own or control. A factory worker was “proletarianised” when the craft knowledge that had lived in their hands was transferred into machines they operated but couldn’t shape. The thesis argues that something analogous is happening to consulting — not by replacing consultants wholesale, but by absorbing into AI tools the specific activities through which consultants have historically developed their judgment. The process is selective, not total.
Judgment, Sovereignty, and Responsibility name what is at stake. Judgment is the kind of expertise a senior consultant has that an AI tool doesn’t — not just knowing the frameworks, but knowing when the frameworks are wrong, reading what a regulator is really worried about, sensing when a client is saying one thing and meaning another. Harry Collins, a sociologist who studies expertise, calls this “contributory expertise”: knowledge formed through sustained engagement with real consequences, not just pattern-recognition from a large training set. Sovereignty operates at two levels: whether consulting firms retain their own distinctive way of seeing the world, and whether the European professional tradition maintains its own interpretive framework rather than gradually defaulting to the analytical assumptions embedded in AI tools built elsewhere. Responsibility is about who the work ultimately answers to — including the people who are never in the room but whose lives are shaped by infrastructure decisions, and whose conditions of life are being set now for decades into the future.
European Infrastructure Consulting is the specific context the thesis uses to develop and test its argument. Infrastructure — railways, energy, water, urban development — is chosen for particular reasons: it involves long time horizons, public money, regulatory oversight, and consequences for people who never hired a consultant. The stakes of getting it wrong are high and long-lasting. If the argument holds here, where the case for AI assistance should be hardest to make, it is more convincing as a general argument about consulting work broadly.
Under AI Mediation refers to a class of tools — large language models and the systems built on them — rather than any specific product. The thesis deliberately avoids naming particular AI tools because those tools are changing faster than a thesis can be written and examined. The argument is about what this class of technology does to professional formation, not about any particular version of it.
The central claim
The thesis makes a claim at two levels.
At the general level: AI tools, under the conditions in which they currently exist, are hollowing out the kind of situated, answerable judgment that makes professional advisory work legitimate.
In European infrastructure specifically: AI tools, under current conditions, are doing this to infrastructure advisory in ways that carry particular consequences — because the judgment at stake here answers not only to clients but to the public, to regulators, and to the people who will live with the consequences of decisions made today.
Each part of that specific-register claim is doing work. “Generic AI mediation” names the mechanism without locking it to any particular product. “Under current conditions” signals that the argument is not technologically determinist — it depends on the political and economic conditions under which AI models are produced and owned, and later chapters show that different conditions would produce different trajectories. “Proletarianising” means selective displacement of expertise, not wholesale replacement of practitioners. “Situated, answerable judgment” names the contributory expertise that develops through sustained engagement with real consequences. “Legitimacy” means that what is at stake is not efficiency — it is the basis on which advisory recommendations are received as genuinely warranted rather than merely produced.
The “hardest case” logic matters here. If this is happening in the context where AI substitution should be least attractive — where the stakes are highest, the time horizons longest, the public accountability most visible — then the general argument is stronger, not weaker, for being tested there.
The thesis is careful about what it is and isn’t claiming. It is not saying consulting was better before AI tools existed. It is not saying AI tools should be refused. It is saying that the conditions under which consultants are formed into good consultants, and under which firms maintain their distinctive expertise, are being structurally altered — and that the current conversation about AI in consulting doesn’t have the right framework to see this clearly.
Why this matters
Three reasons.
First, the existing research on AI in consulting looks at the wrong level. It asks: which tasks are well-suited to AI tools, and how do you govern their use? This is a useful question, but it misses what happens to the people who are supposed to be learning by doing those tasks. The existing literature can see efficiency gains and skill-retention concerns, but it cannot see what happens to professional formation, to a firm’s distinctive way of seeing, or to the answerability of advisory work to the people who bear its consequences. When an AI tool produces the financial model, the slides get done faster — but the junior consultant who would have spent two weeks wrestling with the model hasn’t learned what they would have learned. This thesis works at the level of formation, not just output.
Second, infrastructure decisions bind the future. A railway built today, a water tariff reform designed today, a grid interconnection commissioned today — these shape how people live for decades, sometimes generations. The advisory profession that supports these decisions has historically been embedded in institutional frameworks that take the public-interest character of the work seriously. If that profession loses the kind of judgment that earns trust — not just competent output but genuine, answerable expertise — the institutions it serves become less answerable too. This is not an abstract concern: it is the practical question the thesis puts on the table.
Third, the argument leads somewhere. The thesis develops practical recommendations for four groups of people: individual consultants, consulting firms, infrastructure clients, and European policymakers. The analysis generates specific things each group can do, not just a warning that things are going wrong.
Six things this thesis contributes
The thesis makes six specific contributions to the analytical conversation.
First, it applies a philosophical framework about deskilling — Stiegler’s proletarianisation thesis — to management consulting for the first time. This framework has been applied to factory workers, platform workers, and cultural producers, but never to the advisory professions. Applying it here requires developing an account of how consulting expertise is structured at three levels simultaneously: the individual practitioner, the firm, and the professional culture. The argument is that all three levels are connected, and that what is happening to formation at the individual level is also happening to identity at the firm and cultural levels through the same mechanism.
Second, it applies Hans Jonas’s account of long-term responsibility to AI-mediated advisory in infrastructure contexts for the first time. Jonas argued that the ability to act at technological scale creates obligations to the future — to people who don’t yet have a voice but whose conditions of life are being set now. His framework has been deployed in environmental ethics and bioethics, but not in infrastructure advisory, where it fits very naturally: infrastructure decisions are precisely the kind of long-horizon, consequence-laden decisions Jonas was writing about.
Third, it maps the professional qualification framework for management consultants against what AI tools can and cannot do, for the first time. The ChMC framework — Chartered Management Consultant, the qualification that structures professional development in the field — describes three stages: Foundation (early career, learning the basics), Applied (mid-career, managing delivery), and Chartered (senior, authoritative judgment). The thesis goes through each stage and identifies specifically what AI mediation does there. Analogous mappings have been done in law and medicine; this is the first for consulting.
Fourth, it identifies and theorises pipeline rupture as a mechanism operating across professions. Software development and law have had AI tools at scale for longer than consulting. Looking at what happened there — not at whether headcount fell, but at what happened to the formation of junior practitioners — reveals a consistent pattern: routine tasks got absorbed into AI tools, senior judgment became more concentrated, and the pipeline from junior to senior got disrupted independently of whether total employment changed. The thesis calls this pipeline rupture: when the formative work through which juniors become seniors is absorbed by AI tools, the disruption propagates forward, shaping the senior practitioners of future decades. This mechanism is the thesis’s most analytically robust finding, because it holds even when the labour-market data is ambiguous.
Fifth, it develops an account of what sovereignty means for advisory work at two levels at once. Existing discussions of AI sovereignty focus mainly on data infrastructure and regulation. The thesis argues that the deeper sovereignty question is about identity and interpretive independence: whether European advisory practice retains its own analytical framework, or whether it gradually defaults to the cultural and intellectual assumptions embedded in AI tools built elsewhere. This happens through the same mechanism at both the national-cultural level and the individual firm level — a process the thesis calls dis-individuation, the gradual erosion of a distinctive way of seeing under the pressure of generic mediation.
Sixth, it constructs four future scenarios for European infrastructure advisory as philosophical thought experiments, not predictions. The scenarios describe four coherent possible futures and then identify what the analysis has to say about all four — the structural trends that hold regardless of which trajectory unfolds. This produces recommendations that are robust to uncertainty, because they don’t depend on correctly predicting which future arrives.
How this research was done
The thesis is reflexive practitioner research. The author is a practising consultant in the European infrastructure space. The puzzle that drives the research is one the author has lived, not observed from outside.
This methodological position draws on the tradition established by Donald Schön, an American theorist of professional practice who argued that competent practitioners don’t merely apply theory to practice but reflect in action, generating and revising knowledge through their engagement with the situations they actually confront. The practitioner has epistemic access that the external observer cannot have: the texture of bids, the political weight of client relationships, the quiet ways in which judgment is formed or foreclosed in daily practice.
That access comes with obvious risks. It is easier to mistake your own situation for the situation of the field, and vested interests are hard to see clearly. The thesis tries to discipline this by being explicit about where it draws on lived experience and where it draws on published evidence, by engaging seriously with research that complicates the argument, and by being honest about the limits of what reflexive practitioner research can claim.
One methodological choice needs explaining: the thesis is deliberately not about specific AI tools. It doesn’t name particular products or models, and it doesn’t try to catalogue what different AI systems can currently do. The reason is simple: those tools are changing faster than a thesis can be written. The argument is about a class of technology and what it does to professional formation — and that argument needs to operate at a level of abstraction that survives the specific tools changing. Where particular products are mentioned, they appear as examples of practice, not as the object of analysis.
The thesis is conceptual and normative, not a systematic empirical study. It doesn’t include interviews, surveys, or longitudinal data. Its claims are about what is structurally at stake, not about measuring exactly how much of it has already happened. That measurement is identified as important further work.
How the thesis is structured
Nine chapters, each building on the previous ones.
Chapter 2 sets the scene: what consulting is, what European infrastructure advisory specifically looks like, and who else is affected by infrastructure decisions besides the client and the consultant. It introduces the “third box” — the publics who bear the consequences of infrastructure decisions without ever being party to the contract — as a normative constituency the thesis keeps returning to.
Chapter 3 lays out the philosophical framework: the ideas from Stiegler, Jonas, and the supporting thinkers that give the thesis its analytical vocabulary, including a careful distinction between two ways of thinking about tacit knowledge — the dominant view in management research, which treats it mainly as knowledge that is hard to transfer, and the thesis’s own view, which treats it as something constitutive of professional becoming, such that displacing the activities through which it develops is a more serious matter than simply automating a task.
Chapter 4 describes how expertise is actually formed in European infrastructure advisory — at the level of individual practitioners, at the level of firms, and at the level of the wider professional domain. It argues that the three levels are mutually constitutive: the individual is formed by the firm, the firm by the domain, the domain by the firms that work within it.
Chapter 5 develops the political-economic argument: who owns AI tools, what cultural and interpretive assumptions they carry, and what happens to the distinctiveness of European advisory practice as they become embedded in daily work. It also examines the structural risk that knowledge produced by consulting firms flows outward into the training data of the AI tools those same firms are using — a kind of extraction that the thesis treats as an extension of the proletarianisation mechanism.
Chapter 6 looks at what has already happened in software development and law — the professions where AI tools have been operating at scale longest — and identifies the consistent patterns: selective displacement, pipeline rupture regardless of headcount changes, and the gradual concentration of value in senior judgment that the tools cannot replicate.
Chapter 7 is the analytical core. It goes through each of the three stages of the professional formation pathway — Foundation, Applied, Chartered — and traces what AI mediation does at each stage and why it matters.
Chapter 8 constructs four scenarios for the future and identifies what holds across all of them: four structural trends that operate regardless of which trajectory unfolds, and that provide the ground for the recommendations.
Chapter 9 draws the conclusions and makes specific recommendations to the four groups whose decisions will shape what happens: individual consultants, consulting firms, infrastructure clients, and European policymakers. It also acknowledges what the thesis has not done and identifies the further work it opens.
The puzzle is one the author has lived. The framework is offered to make visible what the productivity story misses. The chapters that follow develop it.