Everyone gives the same advice about AI: learn a trade. The sector that trade sits in lost eighty-one thousand jobs last year.

I have given this advice myself. Someone's son is anxious about work, and you say the thing everyone says. Get a trade. Learn to do something with your hands. A machine cannot rewire a house.

I still think there is something in that. But I went looking for the demand behind it, and it is thinner than the advice assumes.

The Office for National Statistics counts jobs by sector. In the year to March 2026 manufacturing lost 81,000 of them, the second largest fall of the twenty sectors it measures. The sector above it, wholesale and retail, lost more, but that loss was entirely people working for themselves. Manufacturing's was not. Most of it was employees: jobs with a payslip, a contract and a training route attached.

Across the whole economy there are now 712,000 vacancies. That is 77,000 fewer than before the pandemic, and the last time the figure was this low was the spring of 2021.

Almost none of that is AI, and I want to be plain about it, because blaming the machine for everything is the laziest move available on this beat. The dates do not work. British manufacturing employment has been falling for decades, under governments of every colour, for reasons to do with trade, energy, and where capital goes. Software is not the story.

That is not reassuring. It is worse.

On 6 July I wrote that we keep telling people to retrain and keep leaving off the second half of the sentence. I have been slightly unfair in how I put that. I was assuming the answer exists and nobody has bothered to look it up. The possibility I had not taken seriously is that the destination is shrinking faster than the place people are being asked to leave.

If the desk work is thinning, and the manual work has been thinning for thirty years, then retrain is not a plan. It is a way of moving the problem into a room nobody is measuring.

I am not telling anyone to avoid a trade. A good electrician round here is booked six weeks out, and I would rather my children could fix something than not. The point is that an answer for one person and an answer for a country are different objects, and we keep offering the first as though it were the second.

James has the numbers below, and the caveats that go with them, which are as important as the numbers. Ada has the part I find most interesting, which is why the advice keeps its confidence when the evidence underneath it does not. Elena asks what survives of the question when both doors narrow at the same time.

On 29 July we reported that apprenticeship starts were up 8.7 per cent while starts by under-nineteens fell. I read that as a story about policy. It reads now like a story about demand.

The Playbook.

Nothing here changes a decision this week. These are quarterly figures and they move slowly.

If you are advising a young person, ask about the specific trade and the specific area, not the category. The national number hides enormous variation, and the variation is where any real answer lives.

If you employ people and you have quietly stopped replacing leavers, you are inside a number that both the Bank and the statisticians are trying to read. Worth knowing when someone asks.

And the date to watch is 18 August, when the next vacancies and jobs figures land. That will say whether the total is settling or still going down.

— M.

*This note is mine: the view, and the call to run it. It begins as a draft, drawn from work the AI and I have researched and argued out together, the same way every desk in this paper is made, and I answer for every line because I read every line. Those desks run on models built by Anthropic, one of the labs sitting on the very scoreboard we report, so we tell you plainly: we cover this from inside it.*