Chapter 7 — Three Transformations: What AI Mediation Does at Each Career Stage (plain-language version)

One mechanism, three career stages

Generic AI tools absorb the work through which people become experts. The mechanism lands at three points in a career. At entry, the tasks that used to form juniors are done by the tool before they can form anyone. In mid-career, the polished document stops proving what it used to prove. At the senior end, clients build their own capability and commission less, while the damage done at entry travels slowly towards the senior practitioners of twenty years from now.

The pattern holds in any advisory profession where juniors learn by doing the analysis. I follow it through European infrastructure advisory, using the career ladder of the Chartered Management Consultant qualification (ChMC): Foundation, the first three to five years; Applied, roughly years five to twelve; Chartered, the second decade and beyond. I also use a 2026 framework by Tuczek and colleagues that sorts tasks by how well they suit generative AI 1 : uniform work is to be automated, flexible, context-specific work only assisted.

None of this is fixed in advance. At every stage some firms and partners choose differently, and they are exceptions. Underneath runs the third box: the publics, including future generations, who bear the consequences of infrastructure decisions and to whom advisers answer. At each stage I ask what AI mediation does to whether they are taken into account.

Foundation stage: where the apprenticeship blow lands hardest

Foundation-stage consultants learn by doing consequential work under supervision: drafting, structuring, modelling, presenting.

Paris, 2025. A senior manager uses an AI model to check site photographs and planning documents against the regulations. The task once took a junior a day and a half. It takes him forty minutes, so he no longer hands it down. Three juniors on the same floor have never done this analysis and never watched a trained eye do it. Nobody decided their formation should stop. It simply has.

Tuczek’s framework rightly marks this uniform work for automation. It is also the work that has always turned graduates into consultants: a tariff model that took six months to learn to build now takes days. The junior need not even touch the tool. A manager in Mayer and colleagues’ study of a consultancy in Amsterdam describes the Paris decision exactly: “I will just do that myself because it will take me 2 min” 2 . Productivity rises, but the formation aspect does not.

Two kinds of expertise come apart here: talking fluently in a field’s language, and doing the work itself, which comes from working through one’s own failures. Editing the model’s draft builds the first fast and the second slowly, if at all. Walter Benjamin, a German critic, distinguished a single lived experience from the practical wisdom many such experiences settle into 3 4 . With AI output, they do not settle.

The evidence cannot yet prove the strong claim. A study of more than five thousand customer-support agents found AI raised productivity most for the least experienced 5 , but such studies measure performance with the tool over months, not the ability left without it over a career. The defensible claim is narrower: productivity rises, formation has not been shown to rise with it, and the ways it can fail are documented.

Medicine has a name for the failure. Ke and colleagues call it never-skilling: a basic skill that never develops because AI did the thinking during training 6 . It is still a hypothesis. If it holds, it cannot be repaired after the learning years, only prevented during them, and productivity figures will not show it.

Paoli, writing from corporate management, sees juniors take hard questions to a model instead of the colleague two desks away: “You still pay these people. You no longer form them” 7 . The French profession’s HEC Alumni white paper asks the same question 8 and answers it with a twelve-month pathway of tool fluency in which judgment means checking the machine’s output 8 . That is sound training, but formation takes years of one’s own failures under supervision. The industry has answered the apprenticeship blow with a shorter timetable.

The third box moves further away too. A junior who never learns to read a regulator’s hesitation across the table, the moment the public interest becomes briefly visible, will not recognise it as a senior.

The most thoughtful seniors I know ask less whether juniors produce enough than what they are becoming. That has no metric: the senior cannot prove a loss the output does not show, and the productive junior sees none 9 . One of Mayer’s managers stopped checking on a junior who no longer asked questions; the junior did need help, “because he couldn’t put it into ChatGPT” 2 .

What juniors become is a matter of identity as much as skill. Ibarra, who studied young consultants and bankers, found that juniors build a professional self by watching the seniors in the room with them, trying versions of the role on and keeping what others confirm 10 . The Paris juniors have no one to watch, only a model that supplies how a consultant puts it; Mayer’s juniors even used it to write the story of themselves they presented for promotion 2 . On my reading, the trial self is then supplied by the model: fluent, generic, and with no career behind it that the junior could go on to have.

So the resistance that works is a rule, not a worry: some partners require a junior’s own first draft before any tool is used, and some managers hold dry runs in which the junior explains how they read the task 2 .

Applied stage: where the deliverable stops proving what it used to prove

Applied-stage consultants work with autonomy: they lead teams, handle clients day to day and produce the documents that have long been the firm’s most visible asset.

Frankfurt, 2025. A consultant writes the regulatory-risk section of a bid in under three hours with AI help. Two years ago it took two days, and the partner spent half an hour explaining why two paragraphs had to be reframed, conversations the consultant still refers back to. Today’s review is shorter because the draft starts closer to the answer. Nothing has replaced what the longer distance taught. 💡

The polished report used to prove the firm’s capability and the consultant’s competence. I call this the deliverable-as-performance moat, the advantage that kept competitors out. It is eroding because comparable polish is now available to anyone with the tools, including the client. A firm has to show its worth through what stands behind the document: judgment, grasp, relationships.

Supervision changes too. Reviewing a junior’s AI-assisted draft, often with AI help, is a managerial check; commenting on a memo drafted by hand was a formative act. 💡 A survey by Mizrak and colleagues adds that people hide their AI use for fear it will count against their expertise 11 ; if juniors do the same, the reviewer cannot tell how much of the judgment is theirs and has nothing to correct. Ibarra’s juniors had to “convey a credible image long before they have fully internalized the underlying professional identity” 10 , and closed the gap by trying. A model that supplies the image on demand may remove the occasion to try.

The third box is met more thinly: the consultant who once sat in hearings can work from AI summaries, and the gap shows only when they become the senior expected to bring that feel. Some firms bring Applied consultants into senior engagements they were once kept out of. They remain exceptions.

Chartered stage: shrinking engagements, a thinning pipeline, and judgment that wastes

Chartered-stage consultants contribute judgment above all, and here Tuczek’s framework and my argument agree: situated, consequence-aware judgment is what AI can least replace. The senior tier is exposed all the same, in three ways.

The first is disintermediation: clients with strong technical teams build their own AI capability, take the analysis in-house and call in the senior adviser only for the judgment calls. Each engagement earns less, and a firm that earns less at the top has less to invest in forming people below.

The second is relational erosion. The seniors of 2026 were formed when the old system was intact; the seniors of 2046 are being formed now. Senior advice rests on relationships with clients, regulators and publics, built through direct engagement earlier on. When that engagement thins, so does what reaches the top.

The third is the wasting of judgment people already have. In one study reported in a 2026 review, the detection rate of doctors doing colonoscopies fell from 28.4 to 22.4 per cent when they went back to working without AI after extended use 12 . Today’s seniors may be building a similar deficit, visible only when the tool is unavailable, which is when senior advice counts most. 💡

The third box is met most directly here: the senior adviser across from a regulator whose hesitation may signal a public concern the client has missed. Advisory work is legitimate because that senior answers to the publics it serves.

With AI in the workflow, the HEC Alumni white paper says, risk moves up to whoever signs; the technology “tests the discipline of partners”, and the human moves from producer to guarantor 8 . A signature answers for something only when the signer was formed to know what they are answering for. The partner who signs in 2046 is the Foundation-stage consultant editing drafts in 2026. 💡

Lovett puts it in one line: “effective AI oversight depends on the expertise AI adoption may undermine” 13 . Checking that an output is coherent can be learned from outputs; catching the error only a formed practitioner would see cannot. The chief executive of Secuserve, a French cybersecurity firm interviewed by Pomme, caught a certification norm that “simply didn’t exist” in a polished answer because he knew the field 14 .

The next cohort also inherits a picture of what a senior is. Ibarra’s juniors learned the role by watching the seniors beside them 10 . A junior who watches a partner check the model’s draft, sign and forward it learns the guarantor’s role. That step is mine, not Ibarra’s; for older practitioners the question is whether the adviser they present is still drawn from their own experience. Some partners treat formation as their main contribution. They are still exceptions.

The pattern across all three stages

At Foundation the mechanism closes the path of learning through consequences; at Applied it empties the polished document of its value as proof; at Chartered it shrinks the engagement now and thins the senior tier later. Each break travels forward into the next. Chapter 6’s findings hold throughout: displacement is selective, the pipeline breaks whatever happens to headcount, gains flow up to the firm and out to tool vendors, and the profession must now say out loud what makes its work valuable.

The strongest objection is that AI opens new paths. A senior manager at Capgemini, in the HEC white paper, says juniors rise faster on different foundations, provided judgment is never handed to the model 8 ; Killewald and Haskamp find engagement grows where consultants set the model up to argue back 15 . Every such path depends on something a person or firm does to the tool. None arises from adoption alone, so the objection is the design brief for Chapter 9. 💡

The system that passes this expertise on is also the system through which advisory work answers to the publics who bear its consequences. Chapter 8 carries the pattern into four possible futures.