Annex A - Proletarianisation

A.1 Why this annex exists

Proletarianisation is the concept this work leans on hardest. Chapter 3 builds the framework around it, and it is the lens through which what is happening to advisory becomes visible at all. The term is Marxist in origin, but I use it in Bernard Stiegler’s reworked sense 1 , which takes it well beyond the industrial economy Marx was writing about. This annex sets out what I mean by it, what it is not, and where its claims stop. 💡

One qualification carries most of the weight: selective displacement. As I use the term, proletarianisation is the displacement of savoir-faire into systems the practitioner does not own, rather than the wholesale replacement of practitioners by machines. That is why it produces none of the dramatic effects a headcount analysis would look for, and why it can be at work on professional formation whatever happens to employment.

A.2 The Marxist origin and its limits for this work

For Marx, proletarianisation was the loss of control over the means of production. The industrial worker’s capacity to produce value was separated from ownership of the tools through which value was produced. The separation changed the work itself: the artisan whose skill lived in the tools and materials of the craft became the factory hand whose capacity was subordinate to the factory.

I draw on that account, but it was built for industrial conditions that do not map onto consulting. A consultant is not separated from the means of production in Marx’s sense; the capacity at stake is intellectual and relational, and a partnership is not a factory. I needed a version of the concept that works for intellectual and professional labour without losing what Marx saw.

There is a Marxian route to that terrain. Marx’s later notion of the general intellect, the knowledge a society accumulates and the processes by which it is developed, supports an account of “general intellect subsumption”, the capture and control of societal knowledge itself; Ngwenyama and colleagues (Chapter 5) apply it to generative AI and expert labour 2 . That route reaches intellectual work. But it asks a different question from the one this work needs. The Marxian account asks who owns and appropriates the knowledge; Stiegler’s asks what happens to the knower when the knowledge passes into systems they do not own, and above all to the formation through which a practitioner becomes one. The first is a question of appropriation, and a change of ownership would answer it. The second is a question of individuation, and a change of ownership would leave it open. The formation argument depends on the second, so the version of the concept this work needs comes from Stiegler, with the Marxian reading kept in view. 💡

A.3 Stiegler’s reworking

Stiegler supplies that version. For him, proletarianisation happens wherever practical know-how (savoir-faire) is captured into technical systems owned elsewhere, and wherever professional or cultural identity is at stake in that capture. The mechanism is no longer specifically industrial; it runs across contemporary capitalism in ways Marx did not anticipate. The loss is of the savoir-faire itself, the practical, ethical, and cognitive capacity that makes a particular kind of worker a particular kind of practitioner, rather than of the means of production.

Stiegler distinguishes three kinds of knowledge 1 . Savoir-faire is the practical know-how of a craft or profession. Savoir-vivre is knowing how to live: the cultural and ethical capacity that makes up a way of being in the world. Savoir-penser is knowing how to think, the capacity to engage the world analytically in a particular way. Technical capture, he argues, reaches all three, so what is at stake is cultural and cognitive as well as economic.

A.4 Selective displacement: the qualification this work treats as central

Proletarianisation in Stiegler’s sense does not abolish the practitioner. It displaces specific parts of the work into technical systems and leaves the rest where it was. The line falls where technical mediation can and cannot reach. On one side: routine analysis, the expressive layer of a deliverable, the formative tasks of the early career. On the other: situated judgment, relational intelligence, engagement with consequence, and the contributory expertise of the senior tiers. 💡

Selectivity is why nothing looks wrong from the outside. Practitioners keep working, firms keep trading, the profession keeps functioning. What changes is what practitioners actually do, what firms reproduce, and what gets passed on to the next generation. That is why I read proletarianisation at the level of formation rather than employment: formation is where it shows.

It also marks the distance from substitutionist accounts such as the Susskinds’ The Future of the Professions 3 . They predict wholesale substitution; I predict selective displacement, and the two predictions can be told apart. Mine reaches what theirs cannot: what practitioners are being formed into while parts of their work go elsewhere, and what that does to the profession over a generation.

A.5 How this work uses the term

Proletarianisation is a working category in this work rather than a flourish. Chapter 3 sets it up; Chapters 5 to 9 apply it, from sovereignty and extraction through the adjacent professions, the three Chartered Management Consultant (ChMC) stages, the four scenarios, and the recommendations.

In European infrastructure advisory, it means the selective displacement of contributory expertise into AI systems owned elsewhere, with consequences for formation, sovereignty, extraction, and political economy.

A.6 What this work is not claiming

Three things I am not claiming:

First, that AI mediation will abolish the consulting profession. Because the mechanism is selective, the profession carries on. At stake is what the profession becomes and what the next generation of practitioners is being formed into.

Second, that AI mediation should be refused. The technology is a socio-technical mediator with consequences for how firms, professional bodies, and clients organise the work; the question is how they engage it rather than whether to admit it. Chapter 9’s recommendations assume it is here to stay.

Third, that the trajectory is fixed. Where advisory ends up depends on choices made by four constituencies (consultants, firms, clients, policymakers), and the recommendations exist to make clear what choosing differently would require. Part of the point of the analysis is to make those choices visible at all. 💡

Annex B - Third Box Thinking

B.1 Why this annex exists

💡 The third box is this work’s own coinage. It names the constituency that bears the consequences of infrastructure decisions without holding the contracts under which those decisions are produced: the publics whose lives are shaped by what gets built, including the future generations whose conditions of life are being decided now. Legitimate European infrastructure advisory is constitutively answerable to this constituency, and everything this work argues that advisers owe to those publics and those generations runs through that answerability.

Two things need setting out here that the chapters use without stopping to explain: where the concept comes from (Dewey and Jonas), and the difference between the third box as a constituency and third-box thinking as the habit of asking, at each step, what the analysis means for that constituency.

B.2 The structural problem the concept addresses

An advisory engagement has two visible parties: the firm that produces the work and the client that procures and pays for it. The institutional architecture of the engagement is built around the pair. The contract is between firm and client, the deliverable goes from firm to client, and professional accountability runs between the two.

But what the work does (the decisions it supports, the regulation it shapes, the investments it appraises) lands on people who appear nowhere in that pair. The grid interconnection commissioned by an energy regulator shapes the lives of everyone on the supply. A water tariff reform commissioned by a utility shapes household budgets. A transport extension commissioned by an authority shapes how a population moves. These people bear the consequences without being party to the contract. They are the third box, and advisory practice has historically reached them only at one remove, through regulation, public consultation, and political accountability, never through a contract of their own. 💡

The asymmetry is between consequence-bearing and contract-holding. Advisory work is legitimate to the extent it answers for the consequences it produces, yet nothing in the engagement itself, contract or deliverable, recognises that answerability. I take the asymmetry to be constitutive of the profession’s problem, and the third box is the name that keeps it in view.

B.3 The Deweyan ground: how publics are formed

The first ground is John Dewey. In The Public and Its Problems, he argues that a public has to come into being: it forms when people are exposed together to the consequences of decisions made by others, and when institutions recognise that shared exposure as political standing 6 . Public-interest practice depends on the institutions that do the recognising.

That is what lets me treat the third box as a constituted political reality rather than an abstraction. The people affected by an infrastructure decision become a public through the consequences they bear, and European practice already has the machinery for meeting them: public consultations, regulatory hearings, multi-stakeholder processes. The third box names that public where it meets advisory work.

Dewey also spares the concept the charge of being a normative imposition. On his account the third box is already there, wherever consequences are shared; I am not adding it to the analysis from outside.

B.4 The Jonasian register: responsibility toward futures

The second ground is Hans Jonas’s imperative of responsibility. Jonas extends responsibility beyond the parties present to anyone whose conditions of life a present decision will shape, including future generations who cannot be represented in today’s political processes 7 . They cannot take part in the decisions that bind them, and they will live with the consequences over horizons longer than any life or political cycle.

Jonas is what puts future generations inside the third box. Infrastructure binds futures materially. A fifty-year asset commissioned inside a five-year political cycle sets the conditions of life for people who had no say in authorising it. Advisory work answers across that span too, and the third box is where this work holds that obligation. 💡

Jonas, then, adds time. Without him the third box would be the people at the consultation; with him it includes the people who will inherit the asset.

B.5 Why the three-box framing matters

The conventional view has two boxes, firm and client: the engagement is bilateral, and whatever happens to non-contracting parties is left to politics and regulation to handle, upstream of advisory practice.

Adding the third box moves the constituency inside the practice. Advisory work can then be judged on its answerability to the people it affects, and the contracts, methods, and deliverables of the trade can be examined for how they engage that constituency, or fail to.

The framing also travels. Who sits in the third box changes with the subdomain (the public of a rail extension is not the public of a water tariff or a defence procurement), but the position itself (consequence-bearing without contract-holding) is the same everywhere. That constancy is what makes the concept usable across the field.

B.6 Third box thinking as analytical disposition

Third-box thinking is related to the third box but is not the same thing: the third box is a constituency, whereas third-box thinking is a habit of asking, at every step of the analysis, what this means for the people who bear the consequences without holding the contract.

The habit requires that the question of who answers to whom is never dropped, even where the immediate subject is sovereignty, formation, or firm economics. It does not require that every paragraph name the third box.

The distinction separates the claim from the method. The claim is that advisory work answers to the third box. The method is the discipline of checking, section by section, whether the analysis has kept that answerability in view.

B.7 How this work uses the concept

Chapter 2 introduces the third box and grounds it in Dewey and Jonas. Chapters 4, 5, 8 and 9 apply it (the three-scale system, sovereignty and extraction, the fourth of the seven cross-scenario findings, the recommendations), and Chapter 7 threads it through the Chartered Management Consultant (ChMC) stages, where the constituency recedes at Foundation, is met thinly at Applied, and met directly at Chartered.

It works, by design, as a motif as much as a category: it comes back, and it carries more each time. The third box is where the argument about what advisers owe publics and future generations always returns, and keeping it in sight across the chapters is part of what I take to matter about European infrastructure advisory.

B.8 What the concept does not claim

Three things the concept does not claim.

First, that the third box has a single identity. Who is affected differs by subdomain, jurisdiction, and time horizon. The third box is a structural position, and who occupies it changes with the case.

Second, that procedure is enough. Consultations, hearings, and stakeholder processes reach the third box partially, and they matter. But answerability runs deeper than compliance. A consultation run to satisfy a procurement condition is a performance of engagement, and it is not what legitimacy rests on.

Third, that the concept displaces the democratic, regulatory, and political institutions through which Europe has historically engaged its publics. I work within them rather than against them. The worry is narrower: that AI mediation makes engaging the third box structurally more difficult. The recommendations in Chapter 9 are addressed to those institutions as they stand.

Annex C - The ChMC Framework

C.1 Why this annex exists

The Chartered Management Consultant (ChMC) framework, launched in 2021 by the Chartered Management Institute 8 , is the chartered-status framework for management consultancy in the UK. It anchors this work’s definition of consulting, and its three stages are the scaffold on which Chapter 7’s stage-by-stage analysis is built. This annex sets out what the framework is, how I use it, and where it runs out. 💡

Two chapters lean on it: Chapter 2, where it is introduced, and Chapter 7, where the three stages carry the analysis. For those purposes the framework as published is enough. A fuller treatment of it as an institution would need systematic comparison with adjacent chartering frameworks, which this work does not attempt.

C.2 What the ChMC framework is

The framework has three jobs. It defines what management consultancy is for chartered-status purposes, it sets out the professional competence expected at each stage of a career, and it is the route through which chartered status is awarded and maintained 8 .

Structurally it has two parts: four competency components and three progression stages. Crossed, they give a 4×3 matrix of what competence looks like in each component at each stage, and that matrix is what chartered-status assessment and the professional development pathways are built on.

The CMI publishes and maintains it and applies it through its assessment process, and UK consulting firms whose practitioners pursue chartered status work to it. Adjacent jurisdictions have adapted it in modified forms, but the core (four competencies × three stages) is the same in each.

C.3 The four competency components

Each of the four competency components carries analytical weight in Chapter 7. I gloss them here as the framework names them.

Ethics and professional standards covers the ethical commitments a consultant brings to an engagement, the norms of professional conduct, and how those commitments sit against what the client expects: professional codes, ethical reasoning in advisory settings, conflicts of interest, and the mechanisms of professional accountability. It is engaged at every stage, and it is the competency through which answerability to the third box operates.

Leadership and management covers leading engagements, managing teams, holding client relationships at the leadership level, and the standing through which senior practitioners carry weight in their firms and in the wider profession. It is engaged mainly at Applied and Chartered stages, the stages that supervise the earlier one, and so it bears directly on formation.

Client operating environment covers the consultant’s grasp of the client’s institutional setting, the political and regulatory environment the engagement sits in, and the relational intelligence needed to advise well inside it: regulatory awareness, political context, sectoral knowledge, and how the client’s own organisation works. All three stages draw on it, but it carries most weight at Chartered stage, where contextual mastery is the product of years spent in specific client environments.

Personal and professional development covers how a consultant becomes a consultant: the formation pathway through which contributory expertise develops, the firm and professional structures that support development across a career, and the practitioner’s own reflection on what they are becoming. Of the four, this is the one AI mediation disrupts most uniformly across the stages, because its content is the formation system.

C.4 The three progression stages

The three progression stages are the primary structure of Chapter 7.

Foundation is the first stage. Junior consultants build the basic competencies by doing formative tasks under supervision, and the work lays the ground everything later stands on: tacit pattern recognition, applying formal training, exposure to senior practice, and early contact with consequential work. In European infrastructure advisory this is roughly the first three to five years, depending on the firm and the individual’s trajectory.

Applied is mid-career. Consultants run engagements with real autonomy, lead project teams, hold client relationships at working level, and produce the analytical outputs that have historically shown what the firm is worth. They also supervise Foundation juniors, and it is at this stage that a practitioner’s own professional voice takes shape. In European infrastructure advisory this is roughly years five to twelve.

Chartered is the senior tier. Judgment is now the primary contribution. Chartered consultants carry institutional weight in their firms and engage clients at the level where strategic and consequential decisions get made, drawing on contributory expertise built over decades, relationships with clients and regulators sustained over years, and the judgment that separates senior practice from competent analysis. In European infrastructure advisory this is the second decade of practice and beyond.

The stages are more than seniority markers: each builds on what the previous one established and prepares the ground for the next, and Chapter 7 works on that integrated character. One caution on the year ranges above. They fit UK consulting careers, for which the framework was designed, more neatly than continental ones, where entry often runs through grandes écoles, public-sector planning bodies, or national institutions such as Rijkswaterstaat; the stages should be read as an analytical frame rather than a faithful description of how those careers unfold.

C.5 The framework in comparative context

The ChMC is one chartering framework among several, and I note the others briefly without comparing them systematically. 💡

The closest structural cousin is Chartered Engineer status, awarded through the UK Engineering Council and adapted by other engineering bodies: different competency components, but the same shape, with status awarded through institutional assessment and accountability running through its maintenance. The French grandes écoles tradition forms practitioners through the schools, with no professional body in the chartering role, and career progression runs through their networks rather than through any chartering process. Dutch planning and water-engineering institutions fold chartering, education, and institutional employment together in the service of a public-interest tradition in infrastructure, and Nordic planning culture does something similar in its own idiom.

The content differs in each case; the underlying concern, how a profession forms its practitioners and holds them to account, is shared. I use the ChMC as the institutional anchor because it is the framework written for consulting, not because it is the only scaffold available.

One further scaffold belongs here, the first written for AI-mediated practice. The HEC Alumni white paper (Chapter 2) proposes a pathway for the “consultant junior augmenté” 9 . It runs twelve months, from the final academic year to the end of the first year in a firm. Its four quarters move from understanding generative models, to collaborating with them and deciding which tasks to delegate, to integrating them into engagements, to “une vision stratégique et responsable de l’IA”. Its deliverables are documents: a technology-watch journal, a reflective note, a co-written benchmark, a strategic report. Its treatment of judgment is “validation et relecture humaine” of machine output. The one human teacher it names is the “manager formateur”, and its promised result is “meilleure autonomie et productivité dès la première année”.

Set beside the ChMC’s three stages, the differences are the point. Duration: twelve months against a Foundation stage of several years. Teacher: a trainer against supervised, consequential client work. Deliverable: a report against an outcome someone lives with. Certification: fluency with a tool against the judgment the tool cannot supply. The pathway is competent as training, but as a formation scaffold it is what Chapter 7 predicts an industry would design once it has seen the apprenticeship blow and can no longer afford the apprenticeship: selection at the gate, through “AI interviews” that test for human-in-the-loop reflexes on entry 9 , and acceleration into checking. 💡

C.6 Limits of the framework as an analytical scaffold

The framework has three limits as a scaffold for this work.

First, it is young. Launched in 2021, it has not yet acquired the weight, or the scale of use, that the medical, legal, and engineering chartering frameworks carry. The framework gives this work an analytical apparatus, but the institution behind it is still settling at the same time as the AI mediation this work is about.

Second, it is general. Its four competencies and three stages have to be worked out for European infrastructure advisory, and Chapters 4 and 7 do that work. The framework alone would not carry the argument.

Third, it is UK-anchored. It was written for UK institutions, and carrying it into other European jurisdictions is translation this work does not do systematically. The translation is short to Ireland, the Netherlands, and the Nordic countries; to Germany, France, and the Mediterranean countries, whose professional institutions differ more, it is longer, and I have not made it.

None of this is fatal. The framework is sufficient for what this work asks of it; it is not the complete scaffold the argument points toward, and I would rather say so than pretend otherwise.

Annex D - Individuation

D.1 Why this annex exists

Individuation is the process by which something, a practitioner, a firm, a professional tradition, comes to have a form of its own and goes on holding it: a process that continues for as long as the entity does, never a state it arrives at. The concept is Gilbert Simondon’s and reaches this work through Bernard Stiegler. It is what gives the analytical chapters their unity. I use it at three scales (individual, organisational, and national-cultural), and because it is the same concept at each, the chapters can read contemporary AI mediation as one mechanism showing up at three levels of the integrated system. This annex sets out where the concept comes from, what I take from it, how I use it, and what it does not claim.

The account is pitched at working depth: enough for Chapters 3, 5, and 7, which draw on the concept, with what it leaves open marked in D.6.

D.2 The Simondonian origin

The concept originates with Gilbert Simondon, in L’individuation à la lumière des notions de forme et d’information and adjacent texts 10 . Simondon’s claim is metaphysical: the individual is the product of individuation rather than its starting point. Form emerges, and goes on emerging, out of a field of pre-individual relations, and what we call an individual is a phase of that process.

Simondon’s account runs on a few technical terms. Pre-individual being is what individuation draws on: a field of energetic and informational relations that precedes any individual and out of which form is taken. Metastability is the condition that field is in. A metastable system is charged with tension but holds: stable enough to persist, and ready to reorganise when new information or energy enters it. The first term names the material; the second names its readiness. 💡

Trans-individuation names individuation running across several scales at once, so that what the individual becomes and what the wider field becomes are settled together.

That is as far into Simondon as this work goes. Individuation as ongoing process from a metastable field is what the analytical chapters need; the deeper metaphysics is deferred to the Chapter 3 bis this work points toward.

D.3 Stiegler’s extension

Stiegler takes Simondon’s concept into the technical present 11 . For Stiegler, we individuate through the technical systems we work with: they carry pre-individual material that the individuating entity absorbs and transforms.

From this Stiegler derives proletarianisation as dis-individuation under technical capture 1 . When savoir-faire is captured into technical systems owned elsewhere, the entity’s individuation is interrupted: the material it would have absorbed is taken before it can be absorbed, and what the entity becomes is altered as a result. Dis-individuation is the philosophical face of the proletarianisation mechanism Annex A develops.

This is the hinge for this work. It connects individuation to what contemporary technical mediation actually does, and it licenses the three-scale use of the concept that D.4 sets out.

D.4 How this work deploys the concept

I use individuation at three scales.

Individual individuation is how a consultant becomes a particular kind of practitioner. It runs through the integrated three-scale system Chapter 4 describes: formative engagements, sustained exposure to consequence, and the absorption of senior judgment over the arc of a career. Contemporary AI mediation lands directly on those formative engagements.

Organisational individuation is how a firm becomes a particular kind of firm. It runs through the SECI cycle Annex F describes (socialisation, externalisation, combination, internalisation), through which the firm’s distinctive way of seeing is passed down across generations of practitioners. At this scale AI mediation touches what passes through the cycle.

National-cultural individuation is how the European public-interest tradition keeps its distinctiveness alive. It runs through the institutions of European public-interest practice: regulatory traditions, a public-purpose engineering inheritance, and the wider discursive habits of the tradition. Here AI mediation bears on the discourse through which the tradition reproduces itself.

The chapters give the result so far a name, identity: the practitioner, the firm and the tradition as they now stand (Chapter 3, §3.2). At each scale the process is still under way, which is what makes it vulnerable: something still forming can be deformed. And at each scale it is the same Stieglerian mechanism, dis-individuation under generic technical mediation, doing the work. That is the contribution of the three-scale reading: it treats the three levels as sites of one mechanism rather than as three separate worries.

D.5 What the concept does not claim

Three things the concept does not claim.

First, that individuation is good in itself. It is the process through which distinctive form emerges; nothing says the form is desirable. The weight of this work’s argument sits on what is being individuated (the European public-interest tradition, a firm’s way of seeing, contributory expertise reproduced through formation) rather than on the process as such. Nor is identity, the result of the process, good in itself: the argument is no defence of what the profession happens to be today.

Second, that individuation and technical mediation are at odds. Technical mediation is part of how anyone individuates now, and it shapes what they become. The concern is with the institutional arrangements the mediation runs under, ownership above all, and what those arrangements do across the integrated three-scale system.

Third, that dis-individuation under AI mediation is total. The mechanism is selective: some things are displaced, others are kept. The analytical job is to identify what is at risk as mediation accumulates.

D.6 Where the framework is incomplete

The account above is incomplete in three ways, and I would rather name them than hide them.

The first is Simondon himself. Pre-individual being, metastability, and trans-individuation are glossed above at the depth the argument needs and no further; a complete treatment would develop them, and their ontological commitments, in dialogue with the wider traditions they draw on. That is what the deferred Chapter 3 bis would do. 💡

The second is the neighbours. Stiegler’s relationship to Simondon, and the concept’s relationship to adjacent traditions (Heidegger on Dasein, Foucault on subjectivation, Deleuze on becoming), is noted in passing and not worked through.

The third is the join between the scales. D.4 gives each scale on its own and asserts the common mechanism; how the three actually integrate, and in particular how organisational and national-cultural individuation bear on one another in Chapter 5, deserves more than it gets here.

None of this undoes the analytical chapters, which need only what D.1 to D.4 supply. It marks the edge of what this work has done.

Annex E - Tacit and Contributory Expertise

E.1 Why this annex exists

The most direct contribution this work makes against existing information-systems research on AI in consulting is the distinction between tacit-as-constraint and tacit-as-formation. It rests on two sources: Polanyi on tacit knowledge, and Collins on the difference between contributory and interactional expertise. This annex sets both out in more detail than Chapter 3 has room for.

Chapter 3 assembles the apparatus; Chapters 4, 6, 7 and 9 put it to work, down to the recommendations, where contributory expertise itself is at stake.

E.2 Polanyi and the tacit dimension

From Michael Polanyi I take the account of tacit knowledge developed in Personal Knowledge and The Tacit Dimension 12 13 . Its central claim is the phrase that has become canonical: we know more than we can tell 13 . Professional knowledge includes what cannot be fully written down: the judgment, the pattern recognition, the embodied competence of the practitioner.

Three of Polanyi’s terms matter here. The tacit dimension is what lies beneath explicit articulation: attention to subsidiary particulars that produces focal awareness of the whole. Indwelling is how a practitioner comes to inhabit a practice, so that its expertise becomes available without having to be spelled out. Personal knowledge is expertise as held by a particular practitioner: not subjective in the sense of arbitrary, but personal in the sense that it exists only as held by someone whose engagement makes it operative. 💡

Polanyi’s lesson for this work is that professional knowledge always involves more than can be put into propositions. Codification is therefore not substitution: what can be written down does not exhaust what the practitioner knows, and a practice run on the written-down part alone is a different practice.

E.3 Collins on contributory and interactional expertise

From Harry Collins I take the distinction between contributory and interactional expertise, developed in Rethinking Expertise and adjacent texts 14 .

Contributory expertise is the ability to contribute to a practice through embodied engagement. The contributorily expert practitioner can do the work: produce the analysis, exercise judgment, take part in the practice as a member of it. It is built over years of consequential work, and it cannot be reproduced through articulation alone.

Interactional expertise is the ability to engage with a practice through linguistic competence alone, without the contributory grasp. The interactionally expert practitioner can talk about the practice fluently, follow its arguments, and hold their own in conversation with specialists. What they cannot do is the work itself: the embodied competence is not there, and neither is the contribution. 💡

The distinction makes visible what AI mediation is doing now. It produces interactional fluency, and produces it fast: vocabulary, articulation, a surface familiarity with the practice, all absorbed through AI-mediated work. It does not produce contributory expertise, because the embodied competence built through sustained engagement with consequence is a different thing from interactional fluency, and under current conditions I see no sign that AI mediation can reproduce it.

E.4 This work’s distinction: tacit-as-constraint versus tacit-as-formation

The distinction between tacit-as-constraint and tacit-as-formation puts Polanyi and Collins to a use that existing information-systems research on AI in consulting has not.

Tacit-as-constraint is the dominant reading in that research. It treats tacit knowledge instrumentally, as a limit on what AI can substitute for. The work it supports is the systematic mapping of tasks against substitutability, with tacit knowledge as the boundary that marks where substitution stops. Tuczek et al.’s Task-GenAI Fit framework takes this reading 15 ; they cite Polanyi to explain why some tasks resist codification and therefore resist AI substitution. 💡

Tacit-as-formation is the reading I adopt. It treats tacit knowledge constitutively: as the ground on which contributory expertise is built, and as the substance of how a consultant becomes a consultant. The work it supports is the examination of the formation pathway: what is being formed in practitioners under different mediation conditions, what survives and what is at risk, and what follows for the three-scale system through which professional expertise is reproduced.

The two readings ask different questions about the same situation. Tacit-as-constraint asks which tasks resist AI substitution because of their tacit character. Tacit-as-formation asks what AI substitution for formative tasks does to the formation pathway through which contributory expertise has historically been reproduced. Both are real questions, and they lead to different apparatus and different recommendations.

E.5 How this work uses the concepts

Chapter 3 §3.4 sets out the apparatus and §3.5 draws the constraint/formation distinction, with cognitive-science grounding from Kosmyna et al. and Oakley et al. for the formation reading. From there the distinction runs through Chapters 4–9: individual-scale formation in §4.2, the comparative evidence from software and law in Chapter 6, the three Chartered Management Consultant (ChMC) stages in Chapter 7 (where Foundation-stage work under AI mediation produces interactional fluency while compressing the formation pathway, the Applied stage sits at the intersection of the two, and Chartered-stage advisory is the contributory expertise AI mediation does not replicate), and the recommendations in Chapter 9, which are aimed at protecting that pathway.

E.6 What the concepts do not claim

Three things the concepts do not claim:

First, that all professional knowledge is tacit. Much of it can be written down: formal training, technical method, regulatory knowledge, how institutions work. My concern is with the part that operates tacitly and constitutively.

Second, that interactional expertise is worthless. It is what lets practitioners work across disciplinary boundaries, and no one can be contributorily expert in every domain a large project touches. The concern is that interactional expertise is being substituted for contributory expertise in exactly the places where contributory expertise is what made the practice legitimate.

Third, that tacit knowledge is mystical. It is built through embodied competence, years of consequential work, and professional formation, and it is carried by the ordinary institutions of that formation. Tacit knowledge is real, and it is reproducible through institutional practice.

Annex F - The SECI Model

F.1 Why this annex exists

The SECI model (Socialisation, Externalisation, Combination, Internalisation) is the structural device behind Chapter 4’s organisational scale. I use it to characterise how consulting firms reproduce their distinctive way of seeing across generations of practitioners. It returns in Chapter 5’s extraction argument, where the dark mirror reading takes up the concern that under contemporary AI mediation the firm’s externalised knowledge no longer recirculates within its own SECI cycle but flows outward into vendor systems governed from elsewhere.

This annex sets out the model itself, how this work uses it, and where it runs out.

F.2 What the SECI model is

Ikujiro Nonaka and Hirotaka Takeuchi developed the SECI model in The Knowledge-Creating Company and adjacent texts 16 . It names four modes through which knowledge moves between tacit and explicit forms across an organisation; together they form a continuous spiral through which organisational knowledge is reproduced over time 17 . 💡

Socialisation (tacit to tacit) is knowledge passing between practitioners through shared experience: co-presence, observation, apprenticeship. In a consulting firm this is the senior-junior relationship, the partner who takes a junior along, consequential work done under supervision.

Externalisation (tacit to explicit) is the articulation of what was tacit into a form that can be handed on. Consulting firms do this constantly, usually without calling it that: the methodology written up after a project, the playbook, the case study, the internal note that turns one team’s way of working into something the next team can read.

Combination (explicit to explicit) is recombination: new explicit material built out of existing explicit material. Most analytical work on an engagement is combination, assembling the firm’s methodologies, frameworks and case material into something new for this client.

Internalisation (explicit to tacit) is the reverse movement, by which written material becomes part of a practitioner’s own competence through use. Training programmes are the formal version. The formative version is a junior working with the firm’s material until it stops being a document and becomes how they think.

Run together, the four modes form the spiral: what practitioners learn from each other gets written up, the written material gets recombined in new work, the new work is absorbed by the people who do it, and they in turn pass it on. That loop is how a firm’s knowledge outlives any one cohort of its people.

F.3 The model’s application to consulting

Through the SECI cycle a consulting firm reproduces, above all, itself. Its methodologies bear the marks of the people who wrote them. Its case database is its memory. Apprenticeship passes on the partners’ analytical voice, and internal vocabulary carries the firm’s particular way of approaching the work.

The move I make with the model is to read knowledge management as identity-producing practice, and not only as infrastructure for productivity. The cycle reproduces what makes the firm a particular kind of firm. Disrupt the cycle, and you disrupt the firm’s organisational individuation, not just its output.

F.4 The dark mirror: how SECI becomes extraction

Chapter 5 develops the dark mirror reading of the model, and part of what this work adds to the literature on consulting under AI mediation lies there. The dark mirror is the condition under which the firm’s externalisation feeds outward into channels the firm does not control instead of recirculating within it.

SECI assumes that what externalisation produces stays inside the firm. The methodology document is the firm’s intellectual property, the playbook its competitive asset, the case study its memory. They pass from one generation of practitioners to the next through internalisation, and they are kept from competitors by employment contracts, confidentiality agreements, internal-only access, and the plain difficulty of reproducing tacit firm context from outside.

AI mediation opens a pathway through which that assumption can fail. When a firm sends working material (drafts, frameworks, case documents, internal analyses) through systems run by an external vendor, its externalised knowledge is no longer wholly within its control. Whether it stays there or flows outward then turns on the vendor’s terms of service, its data-handling practices, and whether customer content is used to train later models. I should be clear about the status of this claim. It is a structural argument about where the pathway lies, resting on the shape of the arrangement and on conversations with practitioners, not on documented cases of one firm’s material surfacing in another’s tool. Whether, and how much, such leakage actually occurs is an empirical question this work does not settle. 💡

The dark mirror is the structural form of what practitioners informally call the “digital twin” worry: that the tools will learn how a particular person works and replace them. This version is less personal and larger. The sources of the firm’s value across generations are increasingly mediated through tools whose default outputs are common to every firm using them, and the cumulative weight of that generic mediation produces dis-individuation at the organisational scale, just as it does at the national-cultural scale.

Part of what this work does is make the dark mirror visible. The literature on consulting under AI mediation that I have read is well equipped to see efficiency gains and skill retention; it has less to say about what happens to a firm’s distinctive voice across generations, because organisational individuation is not a concern it has the vocabulary for. The dark mirror reading supplies that vocabulary.

F.5 How this work uses the model

Chapter 4 §4.3 introduces the four modes and what each reproduces in a consulting firm; Chapter 5 §5.4 develops the dark mirror reading in the extraction argument; Chapter 7 draws on the cycle in the Applied-stage analysis, where the supervisory role that mediates Foundation-stage formation runs through socialisation and externalisation. Across all three, the model is the device through which this work reads AI mediation’s consequences for firm-level individuation, and not just a description of knowledge management.

F.6 Limits of the model for this work’s purposes

The model has three limits for my purposes.

First, it was developed in and for Japanese firms. How tacit and explicit knowledge are integrated through institutional practice differs between organisational cultures, and carrying the model over to European consulting requires adjustments that I make without spelling them out.

Second, it describes knowledge management inside the organisation and says nothing about the political-economic conditions under which that management happens. The dark mirror reading extends it into the political economy that the original leaves out.

Third, it describes the institutional architecture and says little about what passes through it. The account of what consulting firms actually reproduce through the cycle, the substance of F.3, is my reading rather than Nonaka and Takeuchi’s. 💡

The model is enough for the work I ask of it here, though the complete apparatus this work points toward would go beyond it.

Annex G - TGAIF and the Tuczek et al. Framework

G.1 Why this annex exists

Tuczek et al.’s Task-GenAI Fit (TGAIF) framework, published in 2026 in Business & Information Systems Engineering, is the most directly comparable contemporary contribution to the conversation about AI in management consultancy 15 . It maps consulting tasks against generative AI substitutability through four task profiles (Adaptive Specialist, Creative Generalist, Precise Specialist, Scalable Generalist), and it does so carefully. I engage it as both foil and complement: a serious contribution that this work runs alongside rather than dismisses. Chapter 7 uses it as an overlay onto the Chartered Management Consultant (ChMC) stages. 💡

G.2 What the TGAIF framework is

TGAIF sorts consulting tasks along two dimensions. The first is uniformity against flexibility: is the task executed the same way every time, or does each instance have to be worked out afresh? The second is generality against specificity: does the task run on generic content, or does it need real domain knowledge?

The two dimensions produce four task profiles.

Adaptive Specialist tasks need both flexibility and domain knowledge. TGAIF’s prescription is high augmentation and low automation: the tool supports a practitioner who is still doing the judging and still bringing the expertise.

Creative Generalist tasks need flexibility but run on generic content, so TGAIF allows moderate automation under high augmentation. The tool can carry significant parts of the work; the practitioner supplies the creative judgment.

Precise Specialist tasks are the mirror image: executed uniformly, but only by someone who knows the domain. Moderate automation with balanced augmentation, the uniform components handed to the tool and the domain expertise kept with the person.

Scalable Generalist tasks are uniform and generic, and TGAIF prescribes high automation with low augmentation: the tool can absorb the work, and little human contribution is required.

The result is a systematic apparatus for mapping consulting tasks against AI substitutability, in the tradition of task-technology fit 18 , with recommendations attached for how AI mediation should be deployed in each context.

G.3 What the paper does and does not do

The paper does several things well. Its recommendations are concrete, its grounding is empirical, and it sits inside an established tradition of information-systems research. It also engages Polanyi explicitly to explain why some tasks resist AI substitution, and does so with care 13 . It reads tacit knowledge as a constraint on substitution, where this work reads it as formation (Chapter 3, §3.5, and Annex E). 💡

Three things central to this work are absent from it. Formation is not a category of the framework. The paper names the risk: it lists “GenAI dependence vs. skill retention” among the tensions its interviewees raised 15 and notes that the automation of junior roles “challenges traditional pathways for skill development” 15 . It leaves the point outside the quadrant recommendations, though, so what AI substitution does to the pathway through which contributory expertise has been reproduced is a question the framework itself does not ask 14 . Political economy is context, not analysis: where the AI is produced and who owns it sits outside the frame. And sovereignty, whether a firm’s voice or a national advisory culture, does not appear.

These are choices rather than failures. TGAIF works at the level of the task; this work operates at the level of formation, political economy, and individuation. The two look at different things and do not compete.

G.4 This work’s TGAIF overlay

Chapter 7 lays TGAIF over the ChMC stages: which quadrant does each stage’s work fall into, and what does that imply about how AI mediation is operating there?

Foundation stage maps to the Scalable Generalist quadrant. Most of what a practitioner does in the first five years (routine analysis, basic deliverable production, the formative tasks of the early years) is uniform and generic, with high automation potential. The irony is sharp: TGAIF is empirically right that this is where AI mediation is most efficient, and this is exactly where I locate the apprenticeship blow. Both readings are true at once; they ask different questions about the same situation. 💡

Chartered stage maps to the Adaptive Specialist quadrant. Situated judgment, relational intelligence, engagement with consequence, contributory expertise: high flexibility, high specificity, high augmentation, low automation. Here the two analyses sit closest together and the Foundation-stage tension eases. Chartered work is irreducibly judgmental; a system with no situated engagement with consequence cannot do it adequately.

Applied stage spans all four quadrants: Scalable Generalist (standard analytical structures), Precise Specialist (regulatory compliance, benchmarking), Creative Generalist (strategic synthesis, cross-engagement framing), and Adaptive Specialist (context-sensitive judgment). The mapping is messier than at Foundation or Chartered, and the mess is itself telling. It is what Applied-stage work looks like.

The overlay lets Chapter 7 do something neither framework does alone. TGAIF shows where AI mediation is most efficient; this work shows what the substitution does to the formation pathway. Together they allow a stage-by-stage reading that holds both in view.

G.5 What the differentiation amounts to

This work parts company with TGAIF at five points.

Unit of analysis. TGAIF asks which tasks are substitutable and what that implies for how work is organised. I ask what happens to the pathway through which contributory expertise has been reproduced when AI mediation takes over the content of professional work. The first is a question about tasks; the second is a question about formation.

Theoretical apparatus. TGAIF’s apparatus comes from information-systems research and is built for systematic empirical work; mine is philosophical and built for normative argument.

What each treats as a good outcome. TGAIF’s good outcome is an efficient one: legitimate task organisation is whatever makes the work productive. Mine is advisory work that stays answerable to the third box, the publics and future generations who bear its consequences, and to European public-interest practice.

Political-economic engagement. For TGAIF, where the AI comes from and who owns it is background. For this work it is the foreground: Chapter 5 argues that these are the conditions under which the mechanism runs.

Domain specificity. TGAIF addresses management consultancy in general. This work takes European infrastructure advisory as its site, and its evidence is shaped by what is distinctive about that domain.

None of this says TGAIF is wrong, and on its own terms it is productive. My claim is narrower: the task level cannot reach the question of what contemporary AI mediation is doing to European infrastructure advisory, and reaching that question needs the apparatus this work builds.

G.6 Where the engagement could be developed further

Two directions remain open.

First, the comparison could be widened. Information-systems research on AI in consulting and adjacent professions (Tronnier et al. on user adaptation to AI tools 19 , the broader literature on technology-organisation fit) is engaged here only briefly. A fuller reading would place this work against that research as a whole, not only against Tuczek et al.

Second, TGAIF’s empirical material could be put to work. Some of my claims, particularly the stage-by-stage reading in Chapter 7 where the two frameworks meet, would be more concretely grounded for it.

Neither is needed for the argument as it stands. I flag them as future work. 💡