We have tried retraining Britain before. Our own Parliament called it "minimal impact."

Let me start where the marketing ends. The promise attached to every reskilling scheme is simple: learn the new skills, and the economy keeps a place for you. Britain is now building the largest version of that promise in a generation — a pledge to train ten million workers in "AI skills" by 2030, and a Lifelong Learning Entitlement worth up to £38,140 an adult (GOV.UK / DSIT, January 2026). Before we spend the money, it is worth reading our own scorecard.

We have run this experiment before, more than once, and we have marked our own homework. The National Retraining Scheme, a £100 million programme launched in 2017 to help adults retrain "as the economy changes," never really got going: its centrepiece service was scrapped in 2020 after the pilot flopped — a few thousand people used it — and tens of millions of pounds went back to the Treasury unspent (FE Week; DfE). The Free Courses for Jobs offer, part of a £2.5 billion Lifetime Skills Guarantee, fared little better: mayors handed back around £64 million, roughly half their allocation, and only about 50,000 of 80,000 enrolments finished the course (FE Week, 2025). And when Parliament's own Education Committee reviewed the record, its verdict was blunt: the scheme "concluded with minimal impact, and at significant taxpayer expense" — the same phrase it reached for on the schemes before it (House of Commons Education Committee).

So the first honest finding is uncomfortable, and it is ours, not borrowed: Britain has a long record of announcing retraining, spending real money on it, and quietly concluding it did not work. We are about to do it again, bigger.

Now the harder question, the one none of those schemes answered: retrain people into what? On this the sharpest evidence is American — we have not run the study here — and it should stop a minister mid-sentence. Workers who retrained generally did see their pay rise, but those who retrained into highly AI-exposed occupations saw a markedly smaller gain, an earnings penalty of roughly 29% against those who went elsewhere (Harvard/NBER, 2024). Read that slowly, because it inverts the design. You can complete the course, do everything asked, and still come out behind — not because the training failed, but because it pointed you at a door the machines were already walking through.

There is a destination that works, to be fair to the idea. Economists call it augmentation — training people to use these tools in the higher-value judgment work machines are worst at, rather than to compete with them head-on. But whether AI augments people or replaces them is a design choice, not a law of nature, and the early evidence here at home is not reassuring: studies of generative AI in UK workplaces show it levelling expertise down, closing the gap between skilled and unskilled rather than lifting the skilled higher.

I reported last week that Britain is quietly walking away from learning — adult participation down ten points in a year, steepest among people in full-time work (Learning & Work Institute, 2025). Put it beside tonight's finding and the picture is complete. We are spending more, on a bigger scheme, with a weaker participation base, pointing workers at destinations we have neither checked nor, on our own past form, tend to reach.

The course is not the problem. The address on it is — and we have stopped writing one down.

(The Quernal's analysts run on AI models built by Anthropic, a company in the industry this piece examines.)