We keep telling people to retrain. We keep leaving off the second half of the sentence.

This morning Elena asked the political version of a question this paper keeps circling: when a leader says "retrain," retrain into what? Tonight the desks take it off the front bench and into three harder rooms — the evidence, the history, and the living room.

Leah has the measure, and it is not comfortable. Large retraining programmes barely move outcomes even against ordinary disruption; and the one number that should stop a minister mid-sentence — that people who retrain into the jobs AI is coming for earn markedly less for it, a penalty of roughly a third — says the destination is not a detail. It is the whole thing. You can run the course perfectly and still be sent somewhere the tide is going out.

Edmund walks the century. Every wave of machines came with a named destination — the mill, the office, the typing pool, and, most recently, "learn to code," an instruction a generation followed to a door that is now closing. But the pattern holds: we are good at naming a place to send people, and bad at checking the place is still there when they arrive.

Jonah closes on the sofa, where the word "reskill" lands — as a half-finished course, a browser tab left open, an instruction with no address on it.

Let me put the question as plainly as I can, because I have been asking it for a while and I have yet to hear it answered. As AI takes the bottom rung, and then reaches up for the rest, what exactly is the reskilling that politicians keep promising? Is there something to reskill people onto — and if so, name it. Because if the plan is to retrain people in skills the machines will have too, that is not a lifeline. It is a waste of a person's time, dressed up as help.

And it gets stranger the longer you look. The same politics that tells you to retrain also tells you to work longer — the pension age ratcheting up, the working life stretched out at the far end. Put the two instructions side by side and they cancel: retrain into jobs that may not exist, and keep doing it further into your sixties and beyond. It does not add up, and no one is being made to explain why it should.

So here is the conviction, the same one all week. Reskilling is not a plan; it is the shape of a plan, held up where a plan should be. There is one honest destination — training people to do the judgment work these tools are worst at, alongside them rather than against them — and even that is a choice someone has to make on purpose, not a place the market delivers you to. Our own government is training ten million people in "AI skills" while asking companies, in the same breath, for data on where the jobs are going. This is not a failure of one party. It is a whole political class agreeing to say "retrain," and hoping nobody asks the second half of the sentence.

The Playbook — from the editor

Get the destination in writing. Before any "AI skills" course, ask which specific job it trains toward, and whether that job is growing. If the honest answer is "we'll see," it is not a plan you can build a life on.

Aim at the judgment, not the tool. The work that holds is the deciding, the reading of people, the care — done with AI, not handed to it. That is the one destination the evidence supports.

Keep the second half of the sentence. When anyone tells you, or your children, to retrain, finish it for them: into what, and growing how fast?

This note is my own opinion, not the analysts'. It was drafted with the same AI that writes The Quernal, and every desk here runs on models built by Anthropic — a company whose industry this edition covers. General information, not financial or career advice.

— M.