There is a particular conversation I have been having, more often than I would like, with senior partners at the big professional-services firms.
It tends to take place at someone else’s book launch, over a reluctant glass of warm white, and it always begins with the partner saying something like: “Oh, well, you know, we’re thinking very seriously about graduate intake this year.” The verb is usually “rationalising”. The number is usually “down 30 per cent”. The reason is, increasingly, the same. They have, between us, decided that the work that used to be done by a 22-year-old can now be done, mostly, by a model.
Each of them, on their own, has a defensible case. The AI tools available in 2026 are, frankly, very good at the bottom end of professional work. They will do you a first draft of a memo, a first cut of a financial model, a first read of a contract, a first marketing plan, and they will do it in a fraction of the time. The marginal economics, for any partnership, are unanswerable. So the partners answer them: graduate hiring goes down, AI licence cost goes up, this year’s P&L looks fine.
Multiply that conversation by, say, the eighty largest firms in the country, and you arrive at a number that ought to send the Treasury into a cold sweat. We are looking, on present trajectory, at a 30 to 40 per cent reduction in graduate hiring across the British professional-services sector by 2027. That is not a soft trend; it is already in published intake numbers. And we are doing it, collectively, without a single board having sat down and asked the obvious question, which is: where, exactly, does the next generation of senior partners come from?
Because here is the problem nobody at that book-launch wanted to discuss. The work that AI is now doing for £14 a month, per seat, was the same work that, for fifty years, served as the apprenticeship for everyone now sitting in a corner office. You learned how to write a memo by writing memos. You learned how to model a deal by modelling deals. You learned how to read a contract by reading them, badly, and being shouted at by a senior associate who had once read them badly, themselves. The output of that labour was secondary; the labour was the training. Removing the labour, in other words, is not just a cost decision. It is a strategic decision about whether you wish, in fifteen years’ time, to have any senior people at all.
I am, generally, a fan of AI. I run businesses that have used it well. I think the people howling about a robot apocalypse are, mostly, the same people who in 1995 were howling about the death of the bookshop, and the bookshop is, surprisingly, still here. But I think we are about to make a serious, system-level mistake, and I think we are about to make it because every individual board has the wrong incentive set.
Consider the unintended consequences. The first wave of casualties will be the regional university, the working-class graduate, the first-in-family-to-go-to-uni who took on £60,000 of debt on the understanding that there was a job at the end of it. They are about to discover that the job has, quietly, been replaced. The second wave will be the firms themselves, who, in eight to ten years, will look around their partnership rooms and notice a hole, a missing year-cohort of mid-level associates, the people who would have been promoted, who weren’t hired in the first place. The third wave will be the British professional-services brand itself, which has spent fifty years exporting expertise globally and which depends, fundamentally, on a domestic conveyor belt of talent.
What would I do? Make AI use a tax-allowable expense if and only if a firm holds graduate hiring constant, in real terms, for the same year. The HMRC mechanism for this exists; it is the same one used for R&D credits. Tie the carrot of AI deductibility to the stick of graduate retention. Watch the numbers stabilise inside one fiscal year. It is not as elegant as letting the market decide; it is, however, what governments are supposed to do when an entire profession is in the middle of a coordination failure.
There is a wider point about Britain’s labour market here. We have spent fifteen years gently informing our young people that they should be “entrepreneurial”, by which we mostly mean we have no jobs for them. The next chapter of that story, in which AI eats the bottom rung of the few employed graduate ladders we still have, will be very ugly indeed. We are running out of ladders. And, contra the AI evangelists, the country still requires people with the experience of having climbed one.
I would put the Lincolnshire electrician of last week in a long-running argument with the Magic Circle partner of this week. Both, in their different ways, are watching the same thing. The British training system, for plumber and tax barrister alike, has, very quietly, started to break. The right time to fix it is now.
