London is betting £30m that AI makes more jobs than it takes. This year's data says maybe, for some, if the employer is willing.

Start with the number the Mayor put behind the London AI and Jobs Taskforce: £30m, of which £18m is new. Set it against the report's own headline finding, around 600,000 Londoners in jobs highly exposed to AI and least able to adapt, and the arithmetic looks thin: roughly £50 a head if you spread it across the exposed, which the Mayor is not proposing to do. Most of the £18m goes to the young (£10.5m to help around 6,000 people not in work or education) and to small firms adopting AI. Almost none goes directly to the 600,000 the report names most sharply.

So the money is not a retraining fund. It is a bet. And the striking thing about the bet is that this year's data can be read to back it or to sink it, depending where you look.

Look one way and the warning holds. The exposed are concentrated in clerical and administrative work, the tasks software is learning fastest, they are disproportionately women, and the report's chair warns of a quiet drift: fewer entry-level roles, weaker career ladders, a first rung that thins before anyone notices.

Look the other way and it reverses. PwC's 2026 AI Jobs Barometer, published in June and built on more than a billion job ads, found UK demand for specialist AI skills rebounded 61% last year, back to levels last seen in 2022, even as overall vacancies fell 6.6%. The UK wage premium for AI skills tripled, to 34%. PwC's own reading is blunt: the idea that AI simply replaces workers "isn't borne out by the data", and human skills, judgement, creativity, leadership, are worth more, not less. The firms most exposed to AI are hiring and paying more, not fewer and less.

Hold both, because both are true, and the gap between them is the whole story. PwC describes a market splitting in two: professionalised roles, where AI clears the routine and a person's judgement is worth more, growing fast; and democratised roles, where AI makes the job easy enough that it is worth less. The Taskforce's 600,000 sit mostly in the second kind. The £30m is a wager they can be moved into the first.

Can they? Here the caution earns its place. The same PwC data shows entry-level postings flatlining in the most exposed sectors, and junior roles increasingly asked for senior skills: the on-ramp narrowing exactly where London wants to build one. And a survey out this week from the IT firm HCLTech found that while 90% of enterprises say AI is transforming their workflows, just 18% report a significant hit to revenue so far. (HCLTech sells AI transformation, so weigh its framing accordingly, but the direction is telling.) The productivity everyone reports is not yet money most firms are banking.

That fork decides what happens next. If the gains are real and rising, the PwC story wins: firms grow, reskilled workers move up, and £30m of convening and skills money helps grease it. If the gains stay a story firms tell rather than revenue they see, the older temptation returns: take the productivity as a cost cut, not a redesign, and the 600,000 find the ladder gone before the training arrives.

The Taskforce is right that this is a choice, not a forecast. But the economics are what make the choice for most employers, and £30m does not change the economics. It is a signal, a serious one, that a city has read the numbers and wants a say in the answer. Whether it gets one turns on whether the people it named are treated as workers to build around, or costs to remove. This week's data does not settle that. It rarely does.