The Missing Generation: What Happens When AI Eats the Bottom of the Pyramid?

Every major professional services firm,  consultancy, law firm, accounting practice, systems integrator, has run on the same basic architecture for the best part of a century. A small number of partners at the top, sustained by a much larger base of graduates and juniors underneath, doing the document review, the audit testing, the code, the market research, the first-draft memos. That base was never just cheap labour. It was the training ground. Nobody arrives at a boardroom knowing how to run one; they learn by doing thousands of hours of unglamorous, closely supervised work first.

That base is exactly what generative AI is best at. And across law, accountancy, consulting, and IT services, firms are now visibly hiring fewer graduates to do it.

In Australian law, MinterEllison became the first major firm to openly link a graduate intake cut (down from over 100 places to 72) to AI, rather than hiding behind the usual talk of “market conditions.” Several UK and global firms have pulled back too, even if most won’t say AI out loud yet. In the US, Baker McKenzie cut hundreds of business-services roles in February 2026 citing AI directly. Meanwhile Legal Cheek’s most recent trainee survey found overall UK training contract numbers essentially flat, but with sharp cuts at some firms masked by increases at others. This is  a sign the industry hasn’t agreed on an answer yet, not that the pressure isn’t there.

Consulting shows the same split rather than a uniform collapse. McKinsey has trimmed roughly 10% of its headcount after a hiring spree pushed it past 45,000 staff. BCG and Accenture, by contrast, have grown revenue and headcount, but tilted it hard toward AI engineers and “AI-fluent” hires rather than the traditional generalist analyst. In accountancy, the picture varies by market: some Big Four practices in Asia say they haven’t cut graduate hiring at all and see AI as a retention and productivity tool, while others are quietly restructuring first-year audit and advisory work around AI-assisted review.

Systems integrators face perhaps the sharpest version of the problem, because so much of their entry-level work (writing boilerplate code, configuring standard modules, producing test scripts) is precisely the kind of pattern-matching output large language models now do cheaply and quickly.

In the near term, the effect is mostly a headcount and margin story. Firms do more work with fewer junior staff, entry classes shrink or get more selective, and the juniors who remain are expected to arrive “AI-fluent. They use tools like Harvey or Legora in law, or Copilot-style assistants in audit and SI work, from day one. Client billing models start to shift away from hours-and-bodies toward outcomes, because AI has made the old leverage-based pricing harder to justify. This period feels uncomfortable for the profession but not existential. There are still enough people who came up the traditional way to run client relationships, sign off on judgment calls, and mentor whoever is left.

This is where the more serious problem shows up. The people who would have been mid-level managers, directors, and newly minted partners a decade from now are the ones who never got hired, or who got hired but never did the repetitive, occasionally tedious work that used to build real pattern recognition:  the audit senior who has seen enough dodgy expense claims to smell one, the associate who has drafted enough bad contracts to know exactly where the risk sits, the SI consultant who has debugged enough live systems at 2am to trust their instincts under pressure. Commentators across law, tech, and consulting are converging on the same phrase for this: a “missing generation.” AI can produce a first draft or a first-pass audit file, but it doesn’t sit in the client meeting, absorb the politics of the room, or build the case-by-case judgment that comes from doing the work badly a few times before doing it well. Some firms are trying to head this off now by explicitly redesigning junior roles around supervised AI work rather than simply deleting them. Some are treating juniors more like apprentices than cost centres but by most accounts, that is still the minority approach.

Two decades out, the profession probably splits into two shapes. In the more optimistic version, firms that took the “missing generation” problem seriously in the 2020s have built a smaller but more deliberately trained cohort. There are fewer people go through the pipeline, but each one gets more structured mentorship, faster exposure to real judgment calls, and a clearer path to partnership, because AI has taken the grunt work off their plate rather than their learning off the table. Judgment, trust, and accountability become the scarce, valuable skills, and the firms that cultivated them command a premium.

In the more pessimistic version, the firms that simply cut junior headcount to protect margins arrive at the 2040s with a genuine leadership vacuum. There will be too few people with twenty years of pattern recognition, too much institutional knowledge lost as the last pre-AI generation retires, and a scramble to hire senior talent from wherever it can be found (at a steep premium, and often without the same loyalty or cultural fit). Several analysts already describe this as the biggest quiet risk in professional services: firms optimising for this quarter’s margin while eroding the bench they’ll need in ten years’ time.

It’s whether firms treat the current moment as licence to strip out junior hiring entirely, or as a forcing function to redesign how judgment gets built when the repetitive work that used to teach it is gone. The firms that get that redesign right are likely to be the ones still standing and still able to promote from within in 2046.