Don't Automate the Apprenticeship: The Real Risk in Cutting Junior Consultants

There’s a quiet restructuring happening across professional services, and most of the people affected by it aren’t in the room where it’s decided.

The bottom of the pyramid is being pulled upward. In the last eighteen months, entry-level job postings have fallen by roughly a third; junior openings dropped while senior ones rose. In the professional firms specifically, the signals are unmistakable: KPMG cut its graduate intake by nearly a third in one year, and PwC’s UK leadership openly cut around two hundred entry-level roles and named generative AI as the reason, with fewer first-year tasks now handed to humans.

I want to make an argument that runs against the grain of that trend. Not a sentimental one — a strategic one. I think cutting the junior rung to save money on work AI can now do is one of the most expensive mistakes a firm can make, and that the firms doing it are confusing a cost centre with a training ground.

Let me first give the other side its due, because it isn’t stupid.

The seductive logic

Here is the case for shrinking the junior ranks, stated as strongly as I can make it.

A huge amount of what first- and second-year consultants have always done is exactly the work AI is now good at: the first-draft research, the data reconciliation, the formatting, the summarising of a data room, the initial cut of a model. It was never glamorous, and clients never really wanted to pay for it. If a capable agent can do it in minutes, connected to the actual source systems, then paying a graduate salary to do it slower looks like sentimentality dressed up as tradition. Why staff a pyramid of juniors to produce raw analysis when the analysis produces itself?

And this isn’t a fringe view. Anthropic’s own chief executive has warned that AI could wipe out as much as half of entry-level white-collar jobs within a few years. Plenty of serious people look at the same trend and conclude the junior consultant is simply being automated out, the way the typing pool was.

If you stop the analysis there, cutting juniors is the obvious, rational move. The reason it’s a trap is that it mistakes what junior work is for.

The thing you can’t download

In my last piece I argued that in the age of agents, judgment is the scarce and valuable layer — the ability to decide what matters, to frame the problem, to pressure-test an answer, to carry the risk of a recommendation. Tools are commoditised; judgment isn’t.

Now ask the obvious follow-up: where does that judgment come from?

It is not downloaded. It is not in the model. It is not conferred by an MBA. It is apprenticed — accumulated slowly, over years, by doing real work under the eye of someone more experienced, getting it wrong, seeing why, and gradually developing the instinct that lets a partner glance at a deck and say “the second option doesn’t survive contact with the client.” Every senior consultant you admire was, once, a junior doing unglamorous work while quietly absorbing how the senior people thought. The drudgery was the tuition. The output was almost beside the point.

This is the sleight of hand in the cost-cutting logic. It looks at junior work and sees only the output — the research, the model, the formatting — and reasons that if AI can produce the output, the role is redundant. But the output was never the point of the junior role. The point was the apprenticeship that happened around the output. Automate the output and, if you’re not careful, you don’t just remove a cost — you remove the training ground where your next generation of partners was supposed to learn to think.

Do that at scale and the arithmetic is brutal on a delay. You save money this year on salaries. In ten years you discover you have no senior people, because you stopped growing them — and senior judgment, the one thing clients will still pay a premium for, is precisely the thing you can’t hire in a hurry or generate on demand. This is what a growing number of firm leaders have started, nervously, to call the apprenticeship crisis: by cutting the entry rung, you saw off the bottom of your own leadership ladder.

You cannot automate your way to a wise forty-year-old.

Where the cost-cutters and the romantics are both wrong

Here’s where I part company with the obvious defence of juniors, too — because “so keep hiring lots of juniors” is not the answer either.

If your defence of the junior consultant is that someone has to do the drudgery, you’ve already lost the argument, and you should lose it. Making a bright graduate spend two years reconciling spreadsheets by hand because that’s how it’s always been done is not apprenticeship; it’s waste, and it teaches the wrong lesson. That work should be automated. The evidence even suggests the graduates themselves know it: the ones who thrive are those who arrived with real work experience, hired at roughly twice the rate of those without. They want to do work that matters, not busywork a machine could do.

So both instincts are wrong. The cost-cutter automates the apprenticeship along with the task and severs the pipeline. The romantic protects the task as make-work and wastes the talent. The way through is to separate the two things that junior work used to bundle together — the low-value output, and the high-value learning — and treat them completely differently: automate the output, and deliberately redesign the learning.

Redesigning the early career around judgment

What does that actually look like? It means the junior role stops being “produce the raw analysis” and becomes “direct and govern the machine that produces it, and start exercising judgment on day one.”

Concretely, a well-designed early-career job in 2026 looks less like the old first-year and more like a compressed version of what used to be mid-level work. The junior briefs the agents, supervises their output, sanity-checks the numbers, spots where the model has confidently gone wrong, and translates raw AI analysis into something a client can act on. That is a more demanding job than the old one, not a lesser one — and the data reflects it: roles most exposed to AI are now several times more likely to demand traditionally senior skills like judgment and strategic thinking, and “seniorised” entry-level roles have grown even as generic ones have flatlined.

This is genuinely good news, if firms have the nerve to design for it. It means juniors can start building the valuable muscle — judgment — years earlier than they used to, because the machine has cleared away the years of drudgery that used to sit in front of it. A first-year who spends their time pressure-testing AI output and learning to frame problems is developing partner-grade instincts far faster than one who spent the same two years formatting slides.

But it only works if the apprenticeship is protected on purpose. Automate the repetitive task, not the learning environment. Keep the mentoring, the rotations, the sitting-in on the hard client conversation, the deliberate feedback — the things that were always the real curriculum. The firms that win the next decade won’t be the ones that cut the fastest. They’ll be the ones that used AI to upgrade the apprenticeship instead of abolishing it.

The frameworks are an accelerant here

This is, quietly, what the frameworks in my book are for in the new era — and why I think they matter more for early-career people than anyone else.

A framework is a portable piece of senior judgment. When a junior learns to brief with structure — a role, the missing context, a precise ask, a clear standard — and to govern the output with a sceptic’s eye, they are practising, from week one, the exact discipline that used to take years to absorb by osmosis. GOAL and RCAS aren’t just prompt shapes; they’re a way of thinking that used to be transmitted slowly through apprenticeship, made explicit and learnable. Give a first-year that discipline and a set of agents to direct, and you have compressed a chunk of the old apprenticeship into something they can start doing brilliantly on day one. (If you want the mechanics, the cornerstone piece on working with AI now lays out the operating model, and the GOAL framework is the fastest place to start.)

What this means for you

If you’re early in your career, the lesson is not to panic about the shrinking junior market — it’s to make yourself the kind of junior firms are actually still hiring, and hiring more of. Don’t compete with the agent on producing output; you’ll lose. Compete on the thing the agent can’t do: judgment, context, the ability to catch its mistakes and turn its raw work into something a client trusts. Learn to direct and pressure-test AI, not just use it. The single most valuable habit you can build is the one that used to take a decade of apprenticeship — and it is now learnable deliberately, and early.

If you’re building or running a firm, the question is simpler and sharper than the cost model makes it look. You are not choosing between “expensive juniors” and “cheap AI.” You are choosing between growing your future partners and quietly deciding not to. Automate the toil with enthusiasm. But protect the apprenticeship as if your firm’s next twenty years depend on it — because, unglamorously, they do.

Cut the tasks. Keep the teaching. Don’t automate the apprenticeship.

---

If this was useful, the six frameworks I mention are free — you can get them here — and the full set of a hundred ready-to-run prompts built on them comes with the book, as the companion Prompt Vault for readers.