The shrinking apprenticeship

The first version of every client proposal at my desk has been generated for several months now. I still edit, scope, restructure, push back. I no longer compose from a blank page. Most weeks I forget I used to.

The personal observation is uninteresting on its own. Every consultant I know is now somewhere on the same curve. The interesting question is what the rest of the consulting firm does about the fact that one of its central training rituals just stopped being necessary.

What first drafts were actually doing

The first draft of a client document — a proposal, a discovery memo, a steering deck — was never primarily a delivery artefact. It was the place where a junior consultant learned to think.

You took the brief, the meeting notes, the half-decisions from the client conversation, and you tried to render them into a document somebody would pay for. You discovered, while writing, what you did not understand. You noticed where the scope was incoherent. You felt the weak joints in the argument. You came back to your manager with questions you had not had at the start, because the act of writing had surfaced them.

That was the apprenticeship. The deliverable was almost incidental. The manager’s editing pass was where the real teaching happened, because the junior could now see, with their own draft in front of them, what they had missed.

Generative AI does not replace the document. It replaces the friction. And the friction is where the apprenticeship lived.

The junior consultant now opens with a draft that is structurally complete, grammatically correct, mostly defensible, and produced in two minutes. The manager’s editing pass becomes about taste and judgment, not about teaching the basics of structuring a thought. The thing that used to teach is gone, and the teaching has not migrated anywhere obvious.

Production and deliberation are not the same thing

The simplest way to describe what happened is this: AI compresses production, but not deliberation.

The work that used to take a day now takes two hours of writing and the same six hours of thinking. The drafting got faster. The judgment the drafting was teaching did not get faster — and crucially, it no longer gets taught by the drafting itself, because the drafting no longer requires the apprentice to do the work that produces the judgment.

This is not a technology problem to be solved by a better prompt or a more obedient model. It is a structural change in what the work of producing first drafts was secretly accomplishing.

The steelman

There is a strong version of the opposing view, and any partner who has been in consulting for twenty years will have made some version of it to me by now.

It goes like this. Every generation of consulting tools has displaced some part of the craft, and the apprenticeship has migrated to the next layer up. Word processors displaced typing pools and we worried that juniors would never learn document discipline. PowerPoint displaced hand-drawn storyboards and we worried that juniors would never learn to structure an argument. Excel displaced manual financial modelling and we worried about analysts who could not derive a discount rate by hand. Each time the apprenticeship moved up. The juniors became fluent in the new tooling and learned judgment from a higher starting point. Why should this be different?

It is a serious argument. I half-believe it. The reason I only half-believe it is this: in each of the previous shifts, the displaced layer was visible. The junior could still see what the word processor or the spreadsheet was doing, and could in principle have done it by hand. The displaced craft was apprehensible.

Generative model output is not apprehensible in that way. A junior consultant who accepts a generated first draft of a discovery memo cannot reconstruct how the model decided to organise it. They cannot inspect the chain of small judgments that produced the structure. They can only accept it, edit it on taste, and hand it on. The apprenticeship that used to happen between brief and draft — the part where you learned how to organise — has been compressed into a black box that the junior never opens.

What might replace it

The answer is not to ban juniors from using the tools. That position has the structure of every previous wrong answer to a technology shift, and it loses for the same reason: the next firm down the street will not ban it.

What I think is starting to happen, in the firms that have taken this seriously, is a redesign with three rough shapes.

The first is judgment-first staffing. Juniors are moved much earlier into the rooms where the trade-offs get made — scoping conversations, client steering meetings, internal review of competing proposals — because that is now where the learning happens that the drafting used to provide. They watch a senior decide what to refuse. They listen to a partner re-frame a request. The apprenticeship moves from the document to the room.

The second is the deliberate slow draft. A small number of pieces — perhaps one a quarter for each junior — are still written from scratch, by hand, deliberately. Not because the firm cannot afford the AI-assisted version, but because the junior cannot afford not to have done it. This is the consulting equivalent of medical schools that still teach physical examination in the age of imaging: you teach the slow craft because it is what trains the eye, not because it is what you do on most Tuesdays.

The third is scoping as the apprenticeship craft. The new first job of a junior consultant is not to write the proposal but to construct the scope: to listen, to ask, to articulate the ambiguity, to draft the question that the proposal will answer. The model can produce the proposal once the scope is right. It cannot produce the scope. Teaching scoping as a craft — the way drafting was taught as a craft — is the work in front of consulting leaders right now, and almost nobody has a curriculum for it yet.

What is unresolved

None of the three shapes above are mature. The firms that have started them are not certain they will work. I am not certain.

What I am certain of is that the cohort of consultants now in their first three years will either be the first generation trained on the redesigned apprenticeship or the last generation whose seniors trusted their judgment. The choice is being made at the staffing level inside firms right now — mostly unconsciously, mostly without anyone noticing that it is being made.

The firms that do not have this conversation in the next twelve months will mis-price their own work, and the buyers who keep paying for decks instead of judgment will be the ones who overpay. That part I am willing to defend.

The part where I am less willing to defend is the redesign itself. I would like to be wrong about which of the three shapes will turn out to matter most. Tell me where.