Chapter 3 — The Analytical Framework (plain-language version)

Why a framework is needed

The existing research on AI in consulting asks a reasonable question: which tasks are well-suited to AI tools? The research then maps tasks against the characteristics that AI handles well (structured, verifiable, high-volume) and those it handles less well (context-dependent, relational, novel). This produces useful guidance for firms deploying AI tools sensibly.

But it can’t see something important. It can see which tasks are being absorbed by AI mediation. It can’t see what happens to the people who would have done those tasks — specifically, whether they’re developing the kind of professional judgment that senior advisory work requires. The question it can’t ask is the question this thesis is most concerned with.

To ask that question properly requires different conceptual tools. This chapter assembles them.

The core idea: proletarianisation

The central concept comes from the French philosopher Bernard Stiegler, who spent decades thinking about what technology does to human expertise and identity.

Stiegler took a concept from Marx — proletarianisation — and extended it beyond the factory. In the nineteenth century, factory workers were “proletarianised” when the craft knowledge in their hands (how to work timber, how to read a loom) was encoded into machines. The machine then did what the craftsperson used to do. The craftsperson remained employed but was doing something different — operating a machine rather than practising a craft. The knowledge that had defined them was now living in the machine, not in them.

Stiegler’s argument is that the same mechanism operates in contemporary capitalism across a much wider range of work. It’s not just about physical craft. When practical know-how, professional judgment, and the capacity for independent thought get absorbed into technical systems that someone else owns, that’s proletarianisation in his sense.

Crucially: proletarianisation doesn’t abolish the worker. It displaces specific components of their work — the components that used to constitute their expertise and professional identity — while leaving other components apparently intact.

This is the framework’s central claim applied to consulting: AI tools are not replacing consultants wholesale. They are absorbing specific components of the work — specifically, the components that used to form junior practitioners into experienced ones.

The three types of know-how at stake, in Stiegler’s vocabulary:

• Savoir-faire: practical skill — how to do the work
• Savoir-vivre: how to conduct oneself professionally — the ethical and cultural dimension
• Savoir-penser: how to think independently — not just operating a system but reasoning from first principles

The thesis argues all three are affected.

What gets proletarianised: not the content, but the time

There’s an obvious problem with applying Stiegler’s framework to consulting. In the nineteenth century, the craftsperson’s knowledge was in their body — you couldn’t write down how a master cabinetmaker reads grain, or how a weaver knows when the tension is right. That embodied knowledge was what the machine captured.

But consulting knowledge has never been like that. Consulting firms have spent decades deliberately writing their knowledge down: frameworks, playbooks, methodology documents, training curricula, case databases. McKinsey’s 7S framework, the BCG growth-share matrix, every firm’s proprietary analytical approach — these are prior instances of consulting know-how being externalised into documents. The externalisation was the product the firm was selling.

So if proletarianisation requires capturing embodied, tacit knowledge, and consulting knowledge was already substantially written down before AI arrived — what exactly are AI tools capturing?

The answer is: not the content, but the time.

Imagine the difference between reaching the sixth floor by stairs and by lift. The lift delivers you to the destination. It does not build the leg muscles the stairs would have built.

When a junior consultant spent three weeks wrestling with a tariff model from scratch — making mistakes, asking questions, getting corrections, trying again — the output was the deliverable. But the formation happened in the process. The formative content was not the model itself; it was the duration of struggle, the corrections, the witnessed senior judgment, the gradual development of an instinct for how regulatory problems are structured. None of that is in the spreadsheet. None of it gets transmitted through the spreadsheet.

AI mediation collapses the time. The model is produced in hours rather than weeks. The output is adequate. The formation doesn’t happen.

This is what the thesis means by temporal structure collapse: AI mediation eliminates the temporal structure of formative engagement, not the content of what the engagement produces.

The philosopher Walter Benjamin captures this from a different direction. He distinguished between counsel (practical wisdom that comes from sustained experience and that carries the mark of the practitioner’s formation within a specific tradition) and information (which arrives complete and self-explanatory, with no requirement for experience on either side). What AI tools produce is information. What a formed senior adviser gives is counsel. Sophisticated clients can often sense the difference even when they can’t articulate it.

The individuation argument: becoming a particular kind of practitioner

The other philosophical anchor is Gilbert Simondon, a French thinker whose ideas Stiegler developed extensively.

Simondon argued that becoming a person — or a professional — is not a matter of arriving pre-formed. It’s an ongoing process of development through engagement with the world and with other people. A junior consultant becomes a senior consultant through years of consequential work, challenge, mentorship, and error. The identity is not static; it’s continuously being formed.

Stiegler called the disruption of this process dis-individuation: when technical systems reduce a person’s role to operating a tool rather than genuinely engaging with a problem, the development stalls. The practitioner doesn’t become the practitioner they would have become.

The thesis uses this idea at three levels:

• Individual — the consultant whose professional identity is at stake
• Firm — the consulting firm, which also has an identity (its distinctive way of seeing problems) that can be diluted
• European professional tradition — the broader advisory culture, with its specific public-interest values, that can be gradually displaced by other cultural patterns

All three are at risk under generic AI mediation, through the same underlying mechanism.

The ethics anchor: Jonas and responsibility to the future

Hans Jonas, whose argument about the third box was introduced in Chapter 2, contributes the normative dimension of the framework.

Jonas argued that modern technology has created a genuinely new ethical situation. Our actions can now affect people far in the future — people who don’t yet exist, who have no voice in current decisions but who will live with the consequences. A fifty-year infrastructure asset being designed today will shape how people live in 2070. The advisory work that shapes it carries obligations to those future people, not just to the client who commissioned it.

This is not just the standard “consider future consequences” advice. Jonas’s claim is more radical: when present decisions alter the conditions under which future people will develop and exercise their own capabilities, the ethical obligation is not merely to maximise their welfare — it is to preserve the conditions that make them capable agents at all. For advisory work, this means preserving the formation pathways through which future practitioners will develop the judgment on which the work’s legitimacy depends.

Supplementary ideas that do specific work

Three other thinkers contribute to specific parts of the argument.

Alasdair MacIntyre (philosopher) distinguished between what he called internal goods and external goods of any practice. External goods — money, status, reputation — can be obtained by any means and don’t depend on doing the work well. Internal goods are different: they’re the forms of excellence achievable only through genuine engagement with the practice on its own terms. A consultant who produces a polished deck using AI shortcuts might get the external goods (a satisfied client, a fee). But they’ve foregone the internal goods — the genuine understanding of the problem, the judgment that only comes from wrestling with it. The thesis argues AI mediation most directly threatens the internal goods of consulting practice.

Michael Polanyi (scientist and philosopher) developed the concept of tacit knowledge: the things we know but can’t fully put into words. “We know more than we can tell.” Senior consultant judgment — when to hold a position, when to retreat, what a regulator’s hesitation signals — is largely tacit in this sense.

Harry Collins (sociologist of science) made a critical distinction between two kinds of expertise:

• Interactional expertise: knowing enough to talk competently with experts in a field — you speak the language, follow the argument, produce outputs that read as competent
• Contributory expertise: knowing enough to actually contribute something to the field — to add genuine judgment, to make decisions that wouldn’t have been obvious to anyone else

AI tools are very good at interactional expertise. What they cannot produce is contributory expertise — and contributory expertise is what senior advisers actually deploy, and what junior advisers need to develop.

The most important distinction: two ways of thinking about tacit knowledge

The framework’s most direct contribution to the existing research on AI in consulting is a distinction between two ways of understanding tacit knowledge.

The constraint view: tacit knowledge is information that’s currently hard to transfer, but technology keeps getting better at capturing it. AI tools that make tacit knowledge more explicit are, under this view, straightforwardly useful.

The formation view: tacit knowledge isn’t information at all — it’s the result of a process. The knowledge exists as a capacity that develops through sustained engagement with difficult work. Bypassing the process doesn’t just skip a step — it eliminates what made the step valuable.

The existing research on AI in consulting operates with the constraint view. The thesis argues the formation view is correct for consulting, and that this is why the existing research misses the most important consequence of AI mediation.

The distinction matters practically. Under the constraint view, the question is: which tasks can AI do? Under the formation view, the question is: what does AI mediation do to the formation of the practitioners who would have done those tasks?

The thesis is asking the second question.

A working model of what AI tools actually are

Rather than arguing about specific tools or making predictions that become outdated quickly, the thesis uses a working model of what large language models are:

A large language model is a collective body of human discourse — billions of documents, conversations, and texts — inscribed into a technical system and recirculated under specific ownership conditions.

Three things about this matter:

1. The outputs are recirculations of collective human work, not independent reasoning. When an AI tool produces a regulatory analysis, it is statistically reconstructing patterns from all the regulatory analyses it was trained on. It is not reasoning.

2. The training data carries cultural patterns. The dominant patterns in most large AI models are Anglo-American — because that’s what the training data is. When European advisers use these tools, they’re working with systems whose defaults are shaped by a different professional and cultural tradition.

3. Ownership matters. The conditions under which these systems are built and deployed — who owns them, under what terms, who controls what they’re trained on — shape what happens to the knowledge that flows through them.

This model underpins the sovereignty argument in Chapter 5 and the extraction argument throughout the thesis.

The framework assembled

The pieces fit together like this:

• AI mediation is proletarianising consulting by absorbing the formative components of advisory work — not the content of what experienced practitioners know, but the temporal process through which practitioners develop into experienced ones.

• This operates through temporal structure collapse: the output appears without the formation that producing it would have created.

• The mechanism operates at three scales simultaneously: individual practitioners, consulting firms, and the wider professional tradition.

• The framework for evaluating this is tacit knowledge as formation — the question is not what AI can do, but what AI mediation does to the formation pathway.

• The normative ground is Jonas’s responsibility to futures: preserving the conditions under which future practitioners can develop the judgment on which the work’s legitimacy depends.

Chapter 4 applies this framework to describe how expertise is actually formed in European infrastructure advisory — and sets up the analysis of what AI mediation is changing.