Chapter 7 - Three Transformations across the ChMC Stages

7.1 The three-stage analytical apparatus

The proletarianisation mechanism strikes the formation pipeline of any advisory profession at three points, and in the ChMC vocabulary each has its own blow. At Foundation it is the apprenticeship blow: the formative analytical work through which juniors develop contributory expertise is absorbed by AI mediation before it can form them, because the model produces an adequate output faster than the struggle that used to produce grasp. At Applied it is the erosion of the deliverable-as-performance moat, as AI-mediated outputs reach parity with what once distinguished the firm’s analysis from generic alternatives. At Chartered it is disintermediation now, as clients build in-house capability, and relational erosion later, as the disruption at earlier stages reaches the senior practitioners of two decades hence. A third Chartered dynamic, the atrophy of established judgment under extended AI reliance, is beginning to be documented in adjacent professions.

The stages hold wherever juniors do the analysis, mid-career practitioners show distinctiveness and seniors exercise judgment built over years (§7.5 gives analogues). This chapter traces them through European infrastructure advisory; the content is the domain’s, the mechanism general, as the Introduction’s two-register framing sets out.

Three devices organise the analysis. The ChMC framework supplies the architecture: three stages and four competency components (ethics and professional standards, leadership and management, client operating environment, personal and professional development). The Task-GenAI Fit (TGAIF) framework of Tuczek et al. (2026) lays its four quadrants over the stages and shows where AI mediation fits the work 1 . Chapter 6’s four findings (selective displacement, pipeline rupture regardless of headcount, uneven upward value flow, explicit narration of professional legitimacy) and the three vectors of transformation (commoditisation of outputs, extraction of inputs, apprenticeship rupture) then serve as lenses within each stage.

Is the trajectory deterministic? It’d be sad if it were. At every stage some firms, partners and practice areas recognise the formation question and choose differently. They are exceptions, and I note them without inflating them; Chapter 9 sets out what choosing differently would require institutionally.

The third box (the publics who bear the consequences of infrastructure decisions, including the future generations whose conditions of life are being shaped now) runs through the analysis as the constituency to which advisers answer. At each stage I ask whether AI mediation leaves it engaged, and how. 💡

7.2 Foundation stage: where the apprenticeship blow lands hardest

The Foundation stage covers junior consultants in their first roles, learning the firm and the domain through formative tasks. In European infrastructure advisory it typically spans the first three to five years of a career, in which juniors lay the foundations of all four ChMC competencies through consequential work under supervision.

Paris, 2025. A senior manager at an infrastructure advisory firm has learned to use a multimodal LLM to analyse site photographs, satellite images, and planning document scans against regulatory compliance norms: identifying what does not conform to current standards, generating a structured findings list, and producing a first-draft analysis that previously would have taken a Foundation-stage consultant a day and a half. The analysis now takes him forty minutes including time to verify the model’s reasoning against the relevant norms. He does not delegate it. He has always preferred working with people who have begun to develop their professional eye; with the tool, he does not need to wait for that development. The output is more consistent than a junior’s, and faster; what the juniors see of his eye is the finished list. On the same floor, three Foundation-stage consultants are working on other sections of the same project. None of them have done image analysis against regulatory norms. None of them have been in a room where that professional eye is being exercised. The formation it would have delivered is no longer happening, though no one has decided that it should stop. 💡

Where the blow lands

Mapped onto the TGAIF framework, Foundation-stage work falls mostly in the Scalable Generalist quadrant: high uniformity, high generality, high automation potential. TGAIF recommends it for automation with little augmentation, and on its own terms it is right. It is also the work that has historically formed Foundation-stage consultants, and I read the same correspondence as the point where formation takes its heaviest blow. The two readings ask different questions about the same work.

The Foundation stage is the site of the deepest disruption because the formative tasks here (drafting, structuring, modelling, presenting, the basic analytical movements that build into Applied-stage work) are exactly the tasks AI mediation is most apt to absorb. The tariff model and the failing memo of Chapter 4 are the plainest cases: six months of learning to build the model from scratch become days with AI assistance, and the memo reads as competent at the first attempt. The junior does not even have to use the tool. In the Paris example it is the senior who uses it, and the delegation simply stops. A manager interviewed by Mayer, Baygi and Buwalda, who spent eight months of 2024 in the technology-strategy department of a large consultancy in Amsterdam, describes the same decision in two sentences: “I might not task an analyst to go and turn those 60 words paragraph into 40 words. I will just do that myself because it will take me 2 min” 2 . The authors draw the consequence themselves: the juniors “may miss out on valuable (tacit) insights embedded within those tasks” 2 . 💡 Productivity rises, but the formation aspect does not.

Chapter 3 named this substitution twice over. In Debord’s terms it is spectacular time: the output of formative engagement without its duration, so that the junior has the model and not the formation that building it would have produced 3 . In Benjamin’s, the encounters stay Erlebnis and never sediment into the Erfahrung that makes counsel possible 4 5 .

Chapter 6’s first two findings appear here in their starkest form. Selective displacement acts on formation rather than on headcount: juniors are still hired, staffed on engagements and doing recognisably junior work, but the formative content of that work has changed. The junior who edits an AI-generated draft builds interactional fluency quickly; the contributory expertise that comes from working through one’s own failures develops more slowly, or not at all (Annex E develops the distinction). The pipeline ruptures regardless of headcount.

What the productivity evidence shows

The claim that productivity rises while formation does not goes beyond what the evidence can yet show, and the best evidence against it deserves a hearing. Brynjolfsson, Li and Raymond’s study of more than five thousand customer-support agents found that AI assistance raised productivity by around fifteen per cent on average, most for the less experienced and lower-skilled, and appeared to help workers learn 6 . Dell’Acqua and colleagues’ field experiment with 758 Boston Consulting Group consultants found large gains in the quantity, speed and quality of work on tasks inside the model’s capability frontier 7 . If the tool carries the know-how of the best practitioners to the least experienced, the argument runs, it is itself a formation technology: it shortens the experience curve rather than removing it.

Three things limit what this evidence can carry. It measures performance with the tool, over months, and not the capability that remains without it over a career; the HEC Alumni white paper (Chapter 2) makes exactly this objection to the MIT cognitive-debt study discussed below, that such work measures “un instant d’usage, pas une trajectoire d’apprentissage” 8 , and it holds as well against the productivity studies. It measures tasks inside the frontier, while the same BCG experiment found consultants using the model less likely to reach a correct answer on a task just outside it, which is where formed judgment is supposed to operate 7 . And a field study of consultancies that looks at learning directly finds both effects at once: Killewald and Haskamp (Chapter 4) call it skill polarity, in which “skill expansion and erosion mutually affect each other” 9 . The defensible claim is narrower: productivity demonstrably rises, formation has not been shown to rise with it, and the ways it can fail are documented. That is enough for a chapter whose subject is the pathway, not the output.

Why the deficit does not show

The EEG study by Kosmyna et al. introduced in Chapter 3 names what accumulates across repeated sessions of writing with AI assistance: cognitive debt, down to a reduced sense of authorship 10 . 💡 The effects persist after the assistance is withdrawn; Oakley et al.’s memory paradox (Chapter 3) gives the broader frame 11 . A consultant whose formative engagements are routed through AI is not building the cognitive ground contributory expertise depends on, even while the outputs look competent.

In Ke et al.’s vocabulary (Chapter 6), this is never-skilling, not deskilling 12 : a hypothesis grounded in learning theory, for which the authors find no direct evidence yet. It calls for a different response, since never-skilling cannot be remedied after the formation window has closed, only prevented while it is open. False proficiency, in the same paper’s terms, is why productivity accounting misses it: the deficit surfaces only when the support is withdrawn, or when the practitioner meets a consequence their fluency cannot handle, and by then the window has closed.

Two further accounts describe the same thing from other sides. Xu et al. call it cognitive agency surrender, the cession of epistemic agency that frictionless interface design invites 13 . Quattrociocchi, Capraro and Perc (Chapter 3) call it Epistemia, ‘the possession of an answer without having traversed the cognitive labour of judgment’, and find that scale intensifies it, so that better output makes the formation deficit harder to see 14 .

The diagnosis from practice

The mechanism has also been named from inside practice. Tarki and Raczynski’s Harvard Business Review diagnosis, cited in Chapter 4, puts it at sector level 15 . Paoli (2026), writing from inside corporate management, sees it through the channel I treat as primary, transmission between peers: when junior practitioners take hard questions to a language model rather than to the colleague two desks away, “that small question was how tacit knowledge moved and how a team became more than a list of names”.

He states the consequence with a directness it has taken me several chapters to earn. Judgment, he writes, is “built the slow way: by doing the work badly, then less badly, then well, until you can feel when an answer is off before you can explain why.” The outcome is a cohort of “fluent operators who present well, get along, and cannot tell when the confident output in front of them is wrong”, which is the interactional/contributory distinction in practitioner vocabulary. His closing formulation is exact: “You still pay these people. You no longer form them” 16 . Testimony does not settle the question, but it shows the mechanism is visible, and named, from inside management.

The industry analysis whose vendor-convergence figures Chapter 5 used (§5.3) locates the casualty in the same place: where the next cohort acquires judgment “is an open question once a model does the analytical middle of the work and the pyramid beneath the partner thins every quarter” 17 .

The French profession has reached that diagnosis about itself, and shown what its answer looks like. The HEC Alumni white paper, whose “chaos structurant” Chapter 4 adopted 8 , also observes that agentic tools accelerate experienced consultants “qui voient les erreurs faites par les IA, ce que ne peuvent pas voir les plus jeunes” 8 , and that the building blocks historically given to juniors (diagnostics, financial modelling, scenario analysis, recommendations) are now “de 60 à 80% pris en charge par l’IA” 8 . It then poses the formation question verbatim: “comment les juniors vont-ils apprendre à devenir de bons consultants si l’IA réalise les tâches d’apprentissage basiques ?” 8 .

Its answer is to reinvest the saved time in critical thinking, with the model as sparring partner, so that learning “monte plus vite en cadrage/validation” 8 ; its pathway for the “consultant junior augmenté” fits entry into twelve months of tool fluency, in which judgment appears as “validation et relecture humaine” of machine output 8 (Annex C sets the pathway beside the ChMC stages). As training, nothing in it is wrong. Formation, though, is the passage through one’s own failures, over years, under supervision, which the same document has just said no longer happens. The industry has seen the apprenticeship blow, named it, and answered it with a shorter timetable: the compression the temporal-structure argument predicts, in its clearest institutional form.

Outside the profession, Magnifica Humanitas (Leo XIV, 2026) gives the concern doctrinal form: work is ‘a crucial sphere in which identity is formed’, and ‘excessive reliance and the search for ready-made answers’ ‘weaken personal creativity and judgment’ 18 .

What economic theory adds

Economics noticed this knowledge long before the technology arrived. Hayek asked his readers to remember ‘how much we have to learn in any occupation after we have completed our theoretical training’ 19 : knowledge of people and of particular circumstances, built in practice, which a generic model’s statistical aggregation cannot carry.

Formal models now reach a compatible result by their own route. In Ide’s (2025) overlapping-generations model, novices acquire tacit knowledge by working alongside experts, under contracts left incomplete because the knowledge is embodied and cannot be verified. Entry-level automation raises output on adoption but “can reduce growth and welfare, even without reducing entry-level employment”, when it moves novices away from the most productive experts 20 . The rupture runs through who learns from whom, which headcount does not show. Acemoglu, Kong and Ozdaglar’s model (Chapter 5) reaches it again from different premises, through the depletion of the collective knowledge stock 21 . I have not stipulated the Foundation-stage pipeline rupture: it appears in each place this chapter has looked.

Garicano’s (2000) model of knowledge hierarchies shows why the architecture fails. Read through it, the Foundation, Applied and Chartered stages are layers of problem-solving difficulty, each defined by the range of knowledge its problems require 22 . Practitioners move up through repeated exposure to the escalation boundary: problems they cannot yet solve but learn to recognise as belonging to the layer above. AI substitution at the Foundation layer removes that exposure, and with it the boundary encounters through which knowledge ranges expand, before the expansion can happen.

That ladder is visible even to those who are not looking for it. Mathéo Pomme’s HEC Paris master’s thesis on leaders’ autonomy under AI 23 reproduces, in its appendix, an episode of the podcast Beyond the Prompt in which its co-host Jeremy Utley is asked whether people should use AI for routine work or for the strategic twenty per cent. Most people, he answers, are not good partners to AI, and “probably they can build their chops in the 80 percent stuff and it earns them the right to deal with the 20 percent stuff”; it is, he adds, about building skills and then using skills: “you want to build skills in the less important, but … higher volume stuff. And then you want to use skills in the more important … lower volume stuff” 23 . Offered as advice on prompting, the rule is also an exact account of how a consultant has always learned to consult, and the higher-volume work it sends people to practise on is the work Foundation-stage AI mediation now absorbs. 💡

What is at stake at this stage

The four ChMC competencies show what is at stake. Personal and professional development is the most directly disrupted, since its content is how a consultant becomes a consultant: the development continues, but it becomes interactional rather than contributory. Client operating environment is affected ambiguously: juniors pick up the vocabulary of client engagement fast with AI assistance, but not the grasp of what the client is doing, why, and with what consequences. Leadership and management belongs mostly to later stages, though its early signals (taking ownership, making calls under uncertainty, defending a position) are exactly what AI mediation can paper over. Ethics and professional standards is caught in a bind: ethical formation comes from living with the consequences of one’s own analytical choices, and that is harder when the choices run through a generic system the junior does not own.

The third box becomes further removed at this stage under AI mediation. The junior who works mainly through AI-mediated drafts engages less directly with what the client is doing, which is where the third box’s interests are eventually met at the senior tier. A junior who has not learned to read the regulator’s hesitation across the table, the small motif of European infrastructure advisory in which the third box becomes briefly visible, will not, as a senior, recognise the hesitation when it counts. The loss shows years later, in what the formed practitioner does and fails to do.

In my own practice and in firms I know well, the most thoughtful seniors are increasingly aware of this. Their concern is less the productivity question, “are juniors producing enough”, than the formation question, “what are juniors becoming through what they do.” The first has clear metrics; the second has none, and that absence is itself a structural feature of the situation.

The “perceptual asymmetry” Yu and Moon report among software engineers in South Korea (Chapter 6) gives that structural feature a mechanism: it “prevents either side from correcting these dynamics on their own” 24 . The senior who worries cannot demonstrate a loss the junior’s outputs do not show. The junior who is productive has no reason to believe there is one.

Inside a consultancy the asymmetry takes a plain form. One of Mayer and colleagues’ managers had stopped checking on a junior who no longer brought him questions: “I didn’t check in with him for two days because he didn’t tell me he needed help, so I assumed it was going well. He did need help because he couldn’t put it into ChatGPT” 2 . Another put it in one line: “They went to GenAI and got unstuck” 2 . This is Paoli’s small question seen from the senior’s desk. It carried tacit knowledge one way, and the other way it told the senior where the junior stood. The asymmetry is the formation question’s missing metric seen from both ends of the pipeline at once. It is also why the resistant practices described below are institutional rather than personal: a partner’s standing rule works where a partner’s worry does not, because the rule does not depend on the junior sharing the perception.

The seniors’ question, what juniors are becoming, is about identity as much as skill. The HEC Alumni white paper says so in the summary that opens its Part 4, on skills, roles and firm organisation, where the change it describes “touche au cœur même de l’identité professionnelle du consultant, de sa légitimité et de sa trajectoire de carrière” 8 . Carried over to advisory, Ke et al.’s hypothesis for medicine (Chapter 6) describes a junior whose first drafts come from the model, who learns to treat the job as checking and passing on, and who settles into being that practitioner. The risk is highest at this stage if, as Ibarra’s study assumes (Chapter 3), professional identity “is more adaptable and mutable early in one’s career” 25 .

Ibarra’s three tasks show where the model comes in. Her juniors assembled their possible selves by watching the seniors they worked with most closely, “in particular, those seniors who attended client meetings with them” 25 , then tried versions of the role on and kept what others’ reactions confirmed. The three Foundation-stage consultants in the Paris example have no one to watch: the senior’s eye is exercised alone, with the model. What they have instead is a model that will supply, on request, how a consultant puts it. Mayer and colleagues found juniors using it as a career mentor as well, to prepare appraisals, to write the reflection paper their firm requires through the firm’s own reflection tool, and to “provide a better story of yourself to show that you’re ready for the promotion” 2 . Ibarra’s provisional selves were borrowed from people. On our reading, when much of what a junior tries on comes from the model, the provisional self is supplied by it: fluent, generic, and with no career behind it that the junior could go on to have.

Resistant practice

Bristol, 2024. A senior partner who leads graduate development at a water sector regulatory practice has a standing rule: every Foundation-stage consultant on her team produces a first draft of the analytical section themselves before any AI-assisted tools are used on that section. The draft does not need to be good, only to exist. She is explicit about the reason: I need to see what you think the problem is before I see what the model thinks the problem is. If I skip that step, I am reviewing the model’s thinking and you are reviewing the model’s thinking and neither of us is forming you. Several colleagues regard this as inefficiency that clients are noticing. Her 2021 and 2022 intakes, now in their third and fourth years and taking on Applied-stage work, perform better in client meetings, in her view, than peers from other practices. The correlation is imperfect. She is not sure what else accounts for it. 💡 💡

Mayer and colleagues found a documented cousin of the Bristol rule. Their managers had introduced dry runs before client presentations, in which the junior walks the manager through how they understood the task, how they approached it and where the model came in 2 . The purpose was quality control: to stop unchecked model output reaching clients. It still puts the junior’s own reading of the problem back in front of the senior, after the work rather than before it, and that reading is what the Bristol partner insists on seeing first.

Beyond such rules, some partners route work for formative reasons, giving juniors tasks that could be AI-mediated but whose formative value depends on doing them unaided. Some firms run deliberate non-AI-mediated formation tracks, and whole practice areas, mostly in heavily regulated subdomains where formation failure shows soonest, have agreed to handle certain engagements without AI mediation.

These practices run against the dominant trajectory, and they are strategically intelligent in a way productivity framings cannot capture: the firms that keep formation pathways are betting that their senior practitioners in two decades will be distinguishable from competitors whose pathways have eroded, and that the difference will be worth more than current metrics can show. Whether the bet pays out depends on how the wider profession resolves, which is what Chapter 8’s scenarios examine.

7.3 Applied stage: where the deliverable-as-performance moat erodes

The Applied stage covers mid-career consultants who now work with autonomy: they structure the analysis, lead project teams, manage clients at working level, and produce the work whose output has long been the firm’s most visible asset. In European infrastructure advisory this is roughly years five to twelve, depending on the firm and the consultant.

Frankfurt, 2025. An Applied-stage consultant at a German infrastructure advisory firm is working on a regulatory risk section for an energy client bid response. He produces it in just under three hours: LLM-structured regulatory landscape, firm-specific framing applied in two editing passes, partner review pass of forty minutes. Two years ago, producing that section took him two days. The partner would have reframed two paragraphs and spent thirty minutes explaining why, conversations the consultant still refers back to. Today’s output is as good, possibly better edited. The partner’s review still happens, now as a check at the end rather than a conversation along the way, and it is shorter because the starting point is closer to the answer. The workflow no longer produces the formation that the longer distance used to create, and nothing has replaced it. 💡 💡

Applied work does not map cleanly onto TGAIF. Routine modelling and formatted deliverables are Scalable Generalist work; compliance and benchmarking belong to the Precise Specialist; synthesis across analytical strands sits in Creative Generalist territory; and the judgment calls that mark a consultant as more than a competent analyst are Adaptive Specialist. The messiness is informative: at this stage the deliverable-as-performance moat, the polished deliverable as proof of the firm’s capability and the consultant’s competence, is being compressed beyond what the Foundation-stage mapping captures.

Here selective displacement, Chapter 6’s first finding, reaches the expressive layer of consulting work as well as the formative one. The polished deck and the elegantly framed synthesis were the artefacts through which a firm’s way of seeing became visible to clients, and the Applied consultant who produced them was adding to the firm’s competitive position. AI mediation now produces that layer quickly and competently, so the deck’s craft no longer carries the weight it did when comparable craft was scarce, and that is how the moat erodes. The deliverables are not getting worse; the comparators, including what a sophisticated client can produce in-house, are getting better at what they appear to demonstrate. The firm’s distinctiveness has to be shown through what stands behind the deliverable: the judgment, the grasp of the client, the relational intelligence.

This compression has secondary consequences for the Applied consultant’s role within the firm. Applied consultants have been the formative layer for Foundation-stage juniors, the supervisory tier through which their formation passes. When Applied work is itself substantially AI-mediated, Applied consultants review juniors’ AI-mediated outputs, often with AI assistance of their own. Reviewing a junior’s AI-mediated draft with an AI-mediated review is not the formative supervision that reviewing a hand-drafted memo with analytical commentary was. Supervising AI-assisted production calls for managerial judgment; supervising original production was a formative engagement, and the difference propagates downward to the Foundation-stage practitioners whose formation depends on it. 💡

The deliverable’s loss of evidential value reaches the supervisory relationship from a second direction. A 2026 survey of 564 consulting and professional-service employees by Mizrak and colleagues finds that once a model can supply the polish, “surface quality alone provides less information about the employee’s independent reasoning or professional contribution”, and that the strongest predictors of hiding generative AI use are “AI disclosure anxiety” and “competence penalty fear”, the expectation that acknowledged assistance will count against one’s expertise 26 . The survey does not separate juniors from seniors. If the pattern holds at Foundation and Applied stages, concealment hides the formation deficit from the people best placed to see it: the reviewer reads a polished draft, cannot tell how much of its judgment is the junior’s, and the correction that would have formed the junior has nothing to correct. 💡

Mizrak and colleagues do not measure identity, but it frames how they read concealment. Hidden use, they suggest, may reflect “tensions involving professional identity, autonomy, legitimacy, and the perceived value of human contribution”, in a profession where identity “is closely tied to expertise, analytical judgment, and the ability to produce credible recommendations” 26 . Ibarra’s study suggests why the tension gathers at this stage. Its subjects, moving from analytical into client-facing roles, “must convey a credible image long before they have fully internalized the underlying professional identity” 25 . In that account the image is the persona shown to clients, and the gap is closed by trial (Chapter 4, §4.2). Extended from the persona to the written deliverable, it implies that a model supplying a credible image on demand can remove the occasion to try. The deliverable was the Applied consultant’s way of showing, and of finding out, what kind of adviser they were becoming.

The four competencies shift accordingly. Client operating environment is hit hardest, since the deliverable that demonstrated the firm’s grasp of the client is the artefact AI compresses most readily. Personal and professional development continues through different engagements: learning to add value beyond what the model produces. Leadership and management gains weight, because the Applied consultant now leads juniors formed differently and carries a formative responsibility the previous generation did not. Ethics and professional standards meets the question of attribution: when to acknowledge that work is AI-mediated, and what that implies for the client and for the consultant’s own self-understanding.

The third box is encountered at Applied stage, but more thinly under AI mediation. The Applied consultant who would have spent more time engaging with the regulatory environment (reading consultations, attending hearings, talking with regulators) is increasingly able to operate on AI-mediated summaries of those activities. The summaries are competent. The engagement that would build a contributory grasp of how the third box shows up in a hearing room is thinner. The consequence shows only later, when the Applied consultant who never built that engagement becomes the Chartered consultant who is supposed to bring it.

Stockholm, 2025. An infrastructure advisory firm has introduced a “depth lead” model: one Applied-stage consultant per major engagement takes primary responsibility for the analytical sections, produced without AI-mediated scaffolding, with explicit partner time allocated for joint development review. The managing partner describes it in an internal note as a formation investment, not a quality control mechanism, and specifically not as distrust of AI tools, which the firm uses extensively for other workflow stages. The engagements take longer, though the clients have not raised this. The consultants in the depth-lead rotation, tracked informally over eighteen months, produce, by the firm’s own account, stronger strategic narratives in year two than peers who have not been through it. The firm’s partner track data is too thin to be confident. The managing partner has stopped worrying about whether her depth-lead alumni will be ready to sign. 💡

Firms that think hardest about this give Applied consultants formative engagements to replace the ones deliverable production no longer provides. Some rotate them through practice areas, betting that breadth now shows more than the depth of any single deliverable can. Others bring them into senior work they were once kept out of, the hearings and consultations where the third box shows up, since senior judgment has to be watched before it can be developed. A few partners cultivate Applied consultants whose analytical voice is recognisably their own rather than the firm’s standard output. Such practice is easier to see here than at Foundation, because Applied consultants speak for themselves in client meetings. The dominant movement is still the compression described above: the Applied consultant’s contribution is harder to make legible through the artefacts that once did that work.

7.4 Chartered stage: disintermediation, relational erosion, and judgment atrophy

The Chartered stage covers the senior tier: partners, principals and their equivalents, usually in their second decade of practice, whose primary contribution is professional judgment and who carry institutional weight in their firms.

Amsterdam, 2024. A regional Dutch energy distribution operator has deployed an AI-mediated regulatory monitoring and analysis platform. Two years ago, the operator commissioned a quarterly regulatory intelligence briefing from an advisory firm: forty pages, produced by a four-person Applied-stage team, delivered in a half-day session with the operator’s regulatory director and two senior engineers. The operator now produces its own twenty-page regulatory summary monthly using the internal platform; no adviser signs it. The quarterly briefing has compressed to a ninety-minute strategic conversation with a single senior partner: what the platform cannot tell the client, which is what the evolving regulatory environment means for capital allocation decisions over the next regulatory period. The senior partner’s day rate has not changed, but the total engagement revenue has halved. The Applied-stage team who produced the briefings is working on a different client. The client is, by any measure, analytically more capable than it was eighteen months ago. The practice is smaller, and the practice has started asking what it will produce next year that a platform cannot. 💡 💡

Chartered-stage work falls mostly into TGAIF’s Adaptive Specialist quadrant, where Tuczek and colleagues recommend high augmentation and low automation. Here the two analyses agree most closely: this is the situated, consequence-aware judgment that contributory expertise produces, and a system with no engagement in consequence cannot supply it. The mapping still overstates the protection. Two dynamics act on this stage even though the judgment itself holds, and a third is beginning to show in adjacent professions.

The first dynamic is disintermediation. Clients with strong analytical functions are building in-house AI capability and taking the analytical part of the traditional senior engagement (strategy review, regulatory analysis, market positioning) inside. The engagement shrinks to what that capability cannot perform: the specific judgment calls, the political readings, the relational assessments. It moves fastest where clients already have deep technical teams, such as energy utilities and transport authorities with developed planning departments, and slowest among municipal authorities, smaller water utilities and some social-infrastructure clients. Where it advances, analysis leaves the advisory relationship for in-house platforms with no external adviser answerable for it, which narrows one of the channels through which the third box has been heard; Chapter 8 (§8.6) follows the consequence through. Commercially, a smaller engagement means less revenue per engagement, and the firm must either take on more engagements, straining the senior tier, or build new offerings it may not fund. The individual partner can stay busy while the firm feels the squeeze, and the squeeze travels down: a firm whose senior revenue is compressed has less to spend on forming people at the earlier stages.

The second dynamic, relational erosion, acts on the pipeline that produces Chartered consultants rather than on those now in practice. The Chartered consultant of 2026 was formed in the 2000s and 2010s, through the intact three-scale system of Chapter 4. The Chartered consultant of 2046 is being formed now, with AI mediation running through the Foundation and Applied years. Senior advisory is relational throughout (with clients across engagements, regulators across cases, peers across firms, publics through consultation), and that capacity is built earlier: by the junior who attends regulator meetings, by the Applied consultant who sits in stakeholder engagements. When those earlier engagements thin, the relational capacity that reaches the top is thinner too.

A third dynamic is beginning to be documented in adjacent professions: established judgment wasting under extended AI reliance. A 2026 scoping review of deskilling among physicians reports that erroneous AI prompts raised false-positive recall rates by up to 12 per cent among experienced radiologists, and that in a multicentre study of AI-assisted colonoscopy the adenoma detection rate fell from 28.4 per cent to 22.4 per cent when endoscopists went back to working without AI after extended use 27 .

For Chartered-stage advisory, this brings the pipeline concern from the long horizon into the present. The 2046 cohort may reach the senior tier with thinner formation; the 2026 cohort, after years of AI-mediated practice, may already be building a judgment deficit that shows only when the support is unavailable, which is when senior advice counts most. 💡 The regulator’s hesitation across the table, the ambiguous figure in the client’s financial model, the precedent the AI has no record of: these are the encounters that define senior value, and the ones where extended reliance is most likely to leave a practitioner without resources of their own. Pomme’s study of executive decision-making treats autonomy as “a muscle that atrophies if not regularly exercised” 23 and proposes a “meta-autonomy” in which the leader chooses when to delegate and when to take back control 23 . It does not ask how the judgment that would make that choice is formed.

Of the four competencies, leadership and management weighs most here, since it decides whether the firm’s capacity is reproduced across the generational shift. Ethics and professional standards weighs more than before, because senior decisions bear most directly on the third box. Client operating environment is engaged at its highest resolution and resists substitution best, since contextual mastery is built over years with particular clients. Personal and professional development becomes the cultivation of judgment and the formation of the next generation, and AI mediation alters both.

The third box is encountered most directly at Chartered stage. The senior advisor who sits across from a regulator, whose hesitation might signal a public concern the client has not noticed, is engaging with it in the form constitutive of European infrastructure advisory. My argument depends on this engagement: the legitimacy of advisory work rests on the senior practitioner’s answerability to the publics the work serves, exercised through the relational and judgmental capacities of Chartered-stage practice.

The HEC Alumni white paper locates the future of the senior tier in responsibility rather than expertise. It closes on four shifts, the last of them “de l’expertise à la responsabilité (assumer la gouvernance, la traçabilité et les conséquences des décisions)” 8 . Earlier it observes that with AI in the workflow “le risque ne disparaît pas, il remonte vers celui qui signe”, and that the technology “ne remplace pas les juniors. Elle teste la discipline des partners” 8 ; the human’s role, it says, moves “du producteur au garant de la fiabilité et de l’éthique” 8 . Seen from the client’s side, the same figure comes with a warning: Pomme cautions leaders against becoming “a mere symbolic guarantor of decision already made by a model” 23 . I agree with the relocation and dispute what it leaves out. Answerability is concentrating at the signature at the same moment the formation of future signatories is thinning at the base. A signature answers for something only when the person signing was formed to know what they are answering for; without that formation, the guarantor guarantees little beyond the signature. The partner who signs in 2046 is the Foundation-stage consultant editing drafts in 2026. The white paper’s move is this chapter’s move with the time dimension removed, and the time dimension is the argument. 💡

Lovett’s 2026 “Cognitive Commons” paper in Human Resource Development Review, which treats a profession’s expertise as a shared stock that has to regenerate, supplies the mechanism tying the signature to the formation behind it. He calls it the Validation Tether: “effective AI oversight depends on the expertise AI adoption may undermine” 28 . Surface validation, checking coherence, plausibility and obvious error, can be learned from the outputs themselves. Substantive validation, “recognizing domain-specific errors, identifying contextually inappropriate recommendations despite technical correctness”, “depends not only on what practitioners know but on how they regard knowing” and is built only through the developmental pathway that adoption erodes. Surface validation is what a guarantor can still perform once formation has thinned; substantive validation is what the guarantee is supposed to mean.

The chief executive of Secuserve, a French cybersecurity firm, interviewed by Pomme, gives the distinction a field instance. He asked the model for an audit of the cloud certifications in Europe and received a typology that “looked really exhaustive”; one of the norms it named “simply didn’t exist”. He caught it because, knowing the field, he had never heard of that norm and checked, and he drew the right conclusion: the cases “where you don’t know all the details on something … are really the most dangerous ones” 23 . His check began as substantive validation, a suspicion resting on what he already knew.

The adoption that makes a guarantor necessary spends down the formation the guarantee draws on. Lovett adds a detail this chapter has not stressed: the depleted pathways “also remove the apprenticeship contexts in which practitioners learn to question authority and test claims against evidence”, so the habit of contradiction has to be formed as well. 💡

Formation is not the only thing the next cohort inherits from the senior tier; it also inherits a picture of what a senior is. Ibarra’s juniors learned what a senior adviser is by watching the ones they worked beside, and changed models when feedback told them to 25 . In the Amsterdam example, the ninety-minute conversation that is now the whole engagement has no junior in the room. Where a junior does watch, what they increasingly see is a senior who checks the model’s draft, signs and forwards it; if the partner’s own framings and client notes come through the model, the junior is learning from the model at one remove, and the partner is drawing on the same source for the self they present. The role shown at the top narrows to the guarantor’s, and with it what the junior can aim to become. The step is ours: Ibarra studied juniors, not partners, and assumed identity to be most mutable early 25 . For the older practitioner the question is not formation but whether the adviser they present to the client is still drawn from their own experience.

For the cohort of 2046 the exposure comes earlier and runs deeper than for today’s seniors. The risk is gradual, and nothing in current practice shows it. It shows only if one looks at the formation system as a whole, across decades, and asks whether the conditions under which contributory expertise is reproduced are being preserved.

Edinburgh, 2025. A senior partner at a Scottish infrastructure advisory firm has spent three years thinking publicly about what advisory practices owe their Foundation-stage staff in an AI-mediated environment: in firm strategy presentations, in a contribution to a professional-body discussion on development, in a paper submitted to a built environment journal that was politely rejected and that she intends to revise. She is not opposed to AI tools; her own practice uses them extensively for research and document processing. She has invested in a structured rotation ensuring every Foundation-stage consultant spends at least one substantial engagement in a work type that resists AI mediation: stakeholder facilitation for a contested public infrastructure inquiry, preparation for a formal regulatory hearing, negotiation support for a cross-border infrastructure agreement. She also takes a Foundation-stage consultant into every hearing she leads, so that the juniors see what the signature stands on. She calls it a formation commitment, not an AI policy. It costs the firm approximately twelve percent more per Foundation-year cohort than she estimates the market standard to be. She has not, in three years, lost a single Foundation-stage consultant to a competitor who cited better development opportunities. 💡

Resistant practice at Chartered stage means treating formation as one’s main contribution to the profession’s long-term health: what a partner teaches, what they refuse to delegate to the model, how they route work so that juniors get the engagements they need, and the supervision time they protect, even where the short-term productivity sums do not favour it. These choices show most at the senior tier, because partners have the institutional weight to make them openly, in how a team is configured and what the firm’s training looks like. Peers, juniors and clients can see them, and that visibility is itself a strategic asset for a firm seen to invest in formation.

However public these choices are, the trajectory running beneath them is the one I have traced: disintermediation contracting the engagement now, relational erosion arriving at the senior tier only when the formation cohorts of the 2020s complete their journey to it.

7.5 The cross-stage pattern

Across the three stages the effects differ and the cause does not: in each case generic AI mediation alters the integrated three-scale system through which European infrastructure advisory has materialised its expertise. 💡

The four findings from Chapter 6 hold at every stage.

1. Selective displacement alters formative content at Foundation, commoditises the expressive layer at Applied, and compresses engagement scope at Chartered.
2. Pipeline rupture regardless of headcount acts on formation rather than employment, so it shows whatever headcount does. Paoli (2026) sees it moving forward through the career structure as the altered cohort enters Applied and Chartered practice 16 .
3. Uneven upward value flow: Foundation-stage productivity gains accrue partly to the firm and partly to the platform vendors, while at Applied and Chartered firms compete, and earn, increasingly on the senior judgment AI cannot replicate.
4. Explicit narration of professional legitimacy is required at every stage (what makes a junior’s work valuable, an Applied contribution distinctive, a Chartered judgment authoritative), because the implicit legitimacy that sustained the profession no longer suffices. 💡

The three vectors of transformation run across the stages too.

• Commoditisation of outputs is most visible at Applied, in the eroding moat, but reaches Foundation, where competent outputs no longer signal grasp, and Chartered, where engagement scope is disintermediated.
• Extraction of inputs weighs most at Applied and Chartered, where firm and senior knowledge is most exposed to what Chapter 5 reads as the dark mirror of SECI; at Foundation, juniors’ formative work increasingly trains the next generation of tools as much as the next generation of consultants.
• Apprenticeship rupture is most direct at Foundation and propagates forward: Applied formation depends on Foundation grounding, and Chartered substance on Applied formation.

Across the four ChMC competencies the pattern is one dependency seen four times. Personal and professional development is disrupted most uniformly, because its content is the formation system itself. The other three take their full weight at Applied and Chartered and depend on a formation the earlier stages are thinning: interactional fluency arrives quickly, while contextual mastery, leadership judgment and answerability do not arrive with it. Ethics weighs most of the four, since its exercise is the answerability to the third box, and the productivity framings do not engage it.

Read for identity, the stages ask the same question in turn: who the junior becomes when the tool does the formative work; what kind of adviser the Applied consultant is once the deliverable no longer shows it; what a Chartered signature answers for once the formation behind it has thinned. Killewald and Haskamp close their interview-based case study of two European consultancies by pointing the same way: the frictions consultants report about their own roles “point to broader reservations about how GenAI may reshape professional identities and boundaries” 9 , and they call for closer engagement with research on occupational identity. This work comes at that question through individuation (Chapter 4, §4.5). One of the identities at stake is the adviser who answers to the third box, so the question reaches beyond the profession’s regard for itself.

The strongest objection, beyond the productivity evidence weighed in §7.2, grants that formation changes and holds that AI opens pathways the old pyramid never offered. It has support inside the profession. In the HEC Alumni white paper a Capgemini senior manager argues that the technology does not remove junior work but “transforme le sens même de ce travail”: juniors “montent plus vite en responsabilité, mais sur des bases différentes”, interpreting and checking more than producing, on condition that the judgment is never delegated to the model 8 . Killewald and Haskamp find the constructive use in the field: where consultants set the model up to argue back, “engagement with content and its underlying substance is strengthened rather than weakened” 9 , and, as Chapter 4 noted, juniors take on complex tasks earlier 9 .

I take this seriously; it specifies the argument’s conditions more than it contradicts them. Every reported new pathway depends on something a practitioner or a firm does to the tool: a partner who withholds it until a first draft exists, a mentor who uses it to provoke rather than to answer, a prompt that makes it contest rather than comply. None arises from adoption itself. Yu and Moon (Chapter 6) reach the same point from the other side: the pathway can be preserved only by “deliberate institutional design”, because collective pressure overrides individual disposition 24 . The objection leaves the formation concern standing. It is the design brief for the resistant practices described at each stage above, and for the recommendations of Chapter 9. 💡

A single mechanism produces the three transformations: generic AI mediation absorbs the formative components of the work, erodes the visibility of contributory expertise, and structurally distances the third box from the advisory relationship, acting differently at each stage. The integrated system through which contributory expertise has historically been reproduced is the same system through which advisory work has been answerable to the publics bearing its consequences.

The transformations have analogues elsewhere. In strategy consulting the Foundation blow lands on market sizing, financial modelling and structured problem decomposition under pressure; in financial advisory the Applied moat erodes in the modelling through which firms set their analytical voice apart from generic market analysis; in management consulting for large organisations, Chartered disintermediation follows wherever clients build in-house AI capability. In each, an analogous competency framework would do the work the ChMC does here.

Chapter 8 carries the cross-stage pattern into futures, constructing four scenarios as philosophical thought-experiments and identifying the structural trends that hold regardless of which trajectory resolves.