The tools to run on far fewer people already exist. What's holding the cuts back is a choice — and it won't hold everywhere

Start with the claim in its cleanest form: the cost of paying someone to think is dropping faster than almost anything else in the economy. If that is true, the jobs that sell thinking should be going.

The comforting version of what happens next goes like this — adoption is up, but the payroll data is calm, so relax, it has not really happened. That version is real, and it is also the wrong way round. The calm is the lag, not the all-clear.

Here is what is actually happening in 2026, and it is not slow. One big employer after another has cut staff this year and said, on the record, that AI is part of the reason. Amazon cut 16,000 corporate jobs in January, on top of 14,000 last autumn; its chief executive had already said the plain thing — as the agents roll out, "we will need fewer people." Salesforce shrank its support team from about 9,000 to 5,000 and said it needed "less heads," because the agents handle the work. In Britain, the law firm Baker McKenzie said in February it would cut up to a tenth of its business-services staff — the researchers, marketers and secretaries who keep a firm running — and named its use of AI as one of the reasons. These are not rumours. They are the companies' own words.

The sharper signal is smaller. The firms moving fastest are not the giants; they are small companies, which can rebuild around the tools in weeks. TIME reported this spring on a founder who took his 48-person business down to 30 without losing any revenue — "we actually get slightly better results," he said — after AI agents took over the software he used to pay for and much of the work his staff used to do. Another small firm now has agents writing 90 per cent of its code and answering 70 per cent of its customer questions. A Harvard economist put it plainly: AI spreads fastest in small firms. They are the canary in the mine.

And this is more Britain's story than almost anywhere, because of what the British economy actually is. Strip away the household names and the country is a nation of very small firms: of 5.7 million private-sector businesses, three-quarters have no employees at all, and 99 in every 100 have fewer than fifty people. That base is not a footnote — it is around 60 per cent of private-sector jobs and half of all turnover. It is also the part of the economy that can turn on a sixpence. A thirty-person firm can remap itself around the tools in a fortnight; a bank cannot. So the disruption does not land first in the towers. It lands first in the small firms that are the real backbone of the country — and they are already moving.

So why does the payroll data still look calm? Because most firms have not pulled the trigger yet — and "yet" is the whole story. The Office for National Statistics found about a quarter of British firms using AI by late March 2026, nearly half among the big ones, but only five per cent saying it had cut their headcount. That five per cent is not proof the jobs are safe. It is a count of how many employers have so far chosen to act on a power that almost all of them now have. Anthropic — whose models write this paper, the sharpest conflict we carry — reported this March that its AI is being used for only a fraction of the work it can already do. The gap between "can" and "have" is not a gap in the technology. It is a decision.

And decisions do not hold across a whole market. A firm can look at the map, see that it could run on a third of the people, and choose not to — for the staff, for the quality of the work, for the plain nerve of it. Klarna did something close to the reverse: it swapped customer-service workers for AI, watched satisfaction drop, and hired some back. That restraint is real, and it is sometimes just good sense. But it is a choice each company makes alone, and the one competitor who chooses differently sets the price the others have to match. Restraint is not where a market settles. It is a favour, and favours end.

This is Britain's problem more than most, because Britain sells what it knows — law, finance, consultancy, analysis — and that is exactly the work these tools reach first. The government's own assessment in January found hiring already falling fastest where AI exposure is highest: adverts for the most exposed roles down by more than a third since 2022. A hiring freeze comes before the redundancies. The front edge is already here; the headline payroll number simply has not caught up.

Why it matters here: do not read the calm figures as an all-clear. The power to run your industry on far fewer people already exists — the only reason it has not reached you is that, so far, someone chose not to use it. The useful question is not "will the machines take the work." It is "how long will the people around me keep choosing not to let them, once one competitor decides they will not."