In response to “Can America retrain workers before AI leaves them behind?”, The Economist, 2 August 2026 (https://www.economist.com/united-states/2026/08/02/can-america-retrain-workers-before-ai-leaves-them-behind)
Invest in workers as seriously as America invests in chips. That suggestion appeared in The Economist this week, and the striking thing is that it had to be written at all. Capital knows exactly how to fund a data centre. It doesnât quite know how to fund a career. I say this as someone whose investment work moves through the first channel and rarely finds the second, because the second barely exists.
This asymmetry isn’t a failure of conscience. It’s a pricing problem. A server farm generates returns on a schedule a spreadsheet can hold. A retrained worker does so over a decade, onto someone else’s balance sheet, in a currency the market doesn’t measure. So capital easily floods toward the compute and thins to almost nothing around the people the compute displaces. The Economistâs own reporting makes the gap legible: OpenAI plans to spend on computing power through 2030 something on the order of three thousand times what it and Anthropic have so far committed to helping workers adapt. This ratio isn’t cruelty. It’s what happens when one side of a ledger is visible, and the other pays out too far downstream to price.
I should be honest about where I sit, because the alternative is to write this from a grandstand I don’t occupy. I back the companies building these exact tools. This gives me an obligation and a self-interest, and I’d rather name both than pretend to only the first. The obligation is obvious. The self-interest is the part investors tend to leave unsaid: Gina Raimondo, quoted in the same piece, put it as clearly as anyone has. Get this wrong, and the regulatory backlash is going to land on the companies. She’s right. The businesses deploying capital into AI have every reason to want the adaptation problem solved, and almost none of them are funding it at the scale of the disruption they’re financing. I include my own conviction in that count.
So, what do you actually do about a problem no one is able to specify? The sharpest admission in the original piece is that nobody knows what to train these workers for. Connecticut is exploring whether back-office staff might move into health care, while conceding that insurance agents may not have the bedside manner for it. Washington reads that uncertainty as a planning problem to be solved with better forecasts. I read it as the ordinary condition of a conviction call, because it’s the condition I work in every week.
You don’t resolve this kind of uncertainty by predicting the destination. You fund optionality and stay close to real demand. It’s why the interventions that work look nothing like the ones that don’t. The sectoral programmes, where intermediaries build courses around actual vacancies and coach people into them, show durable earnings gains of between eleven and forty percent across randomised trials. The apprenticeships, which put training inside a paying job rather than ahead of a hoped-for one, hold. The federal voucher system, which funds attendance rather than outcomes and hands a displaced worker a modest cheque and a list of approved courses, mostly doesn’t. This isnât a mystery – it’s the difference between building a portfolio against an unknown and placing a single order and hoping.
Europe understood part of this a long time ago. Germany embeds training inside the employer, so the worker and the skill move together. America bolts it on afterwards, once the job has already gone and the worker is standing outside the building. I don’t raise that to lecture. I raise it because the structure of the thing determines the outcome, and one structure keeps people attached to work while the other catches them, if it catches them at all, after they’ve fallen.
Here is what I think we’re actually failing to price. Capital now moves at machine speed. People still move at human speed. A business can redirect a billion dollars of compute in a quarter; a worker in her fifties cannot rebuild a career on the same clock, and the gap between those two velocities is where the damage collects. The data centres are going to get built either way. What remains unsettled is whether the workforce is treated as a line item or a foundation, and I already know which of those compounds. The market will find out later, on the side of the ledger it never learned to read.