In 1930, John Maynard Keynes sat down to write a reassuring essay for a generation staring into the teeth of the Great Depression. His argument, in “Economic Possibilities for our Grandchildren”, was that the real problem of the future wouldn’t be scarcity. It would be leisure. By the time his grandchildren were grown, he reckoned, the fifteen-hour working week would be the norm, and the harder question would be what on earth to do with all the free time.

That was ninety-six years ago. Keynes’ grandchildren are now grandparents themselves. The fifteen-hour week remains conspicuously absent.

Tim Harford’s column this month asked the obvious question about that historical disappointment. Every generation has had a technology that was supposed to liberate us. Every generation has used it to do more work instead. Is AI going to be any different?

Harford, being Harford, did the respectful thing and reached back further. Jevons, 1865, the coal paradox: make steam engines more efficient and you don’t reduce coal consumption, you increase it. The pattern has been consistent ever since. Farm mechanisation, the washing machine, the computer, the email, the smartphone. Each one promised to give us time back. Each one raised expectations of output until the time got spent somewhere else.

His answer, to nobody’s surprise, was that AI is probably not going to be the exception. I think he’s right. But I also think the interesting question isn’t whether the pattern will repeat at a societal level. It will. The more useful question — the one you can actually do something about — is whether it’s already repeating in your own working week.

The individual version of Jevons

I wrote about this at the organisational level back in February. The acceleration trap is a real and well-documented phenomenon, and most of the conversation around it has been about what managers and executives should do differently to prevent it.

That’s the right question. It isn’t the only one.

There’s an individual version of the same trap, and it runs on slightly different mechanics. Your company can have all the right policies — reasonable expectations, protected focus time, no meeting culture — and you can still end up at the end of an AI-augmented week more tired than you were before. Because the most persistent pressure to fill reclaimed time doesn’t come from management. It comes from you.

This is where I’ll stop pretending it’s other people. For the past year, I’ve been running an unusual amount of my working week through AI. Drafting, research, coding, property-management admin, trading-system prototypes. On paper, the time savings are ludicrous. Things that used to take a Sunday afternoon take twenty minutes. Things that used to take a week of calendar time happen in an evening.

I’m not working fewer hours. I’m working different hours, on harder problems, and on more of them simultaneously. The AI hasn’t given me my weekends back. It’s let me cram more of the interesting work into them.

And nobody’s making me do that. That’s the bit that keeps nagging at me.

What the data is starting to say

Mercer’s 2026 workforce research — their large-scale look at how people are actually experiencing the AI transition, rather than how executives think they’re experiencing it — put out two findings that caught my eye.

The first is that employee concern about AI-driven job loss has climbed from 28% in 2024 to 40% in 2026. That’s not a productivity story, but it’s the substrate everything else sits on. If you’re worried AI is coming for your job, you don’t respond to an AI-accelerated workflow by working less. You respond by proving you’re still indispensable.

The second is that 62% of employees feel leaders underestimate the emotional and psychological impact of AI on work. Translation: the people actually using these tools don’t feel the time savings in their lives, and they don’t feel their organisations understand why.

Put those two together and you get the shape of the individual trap. The pressure to do more with your AI-augmented capability is partly organisational — expectations creep, scope creeps, meeting density creeps — but it’s also personal. Fear of redundancy. The compulsion to justify yourself with visible output. The very old, very human impulse to fill a vacuum with activity.

Jevons figured out the economic version in 1865. He didn’t need to figure out the psychological version. We’ve all been running it ourselves for years.

The income/substitution frame

One of the things I liked about Harford’s piece was that he didn’t paint this as a moral failing. He reached for the standard economist’s toolkit: two effects, pulling in opposite directions.

The income effect says that when AI makes each hour of your work more valuable, you can produce the same output in less time, so rationally you should work less and take the gains as leisure. The substitution effect pulls the other way. When each hour of work is more valuable, the opportunity cost of not working goes up, so rationally you should work more.

Economists have known for a long time that the second one tends to win in the early phase of a productivity shock, and honestly I’m not sure that’s settled science so much as a rough empirical regularity. Still, the historical data on working hours broadly supports it. Average weekly hours did fall over the twentieth century, but a lot less than Keynes predicted, and mostly because of legislation and collective bargaining rather than individual choice.

AI is a productivity shock with one unusual feature. The efficiency gains don’t arrive as “your employer gives you a faster tool.” They arrive as “you, privately, found a way to do something in twenty minutes that used to take three hours.” The negotiation isn’t happening at the company level. It’s happening in your head.

That’s part of why this is personal in a way previous waves weren’t. You’re the one deciding whether the saved time gets captured or reinvested. If the default is “reinvested”, you will be measurably better at your job a year from now — and measurably more tired.

A rough audit

I don’t have a seven-step plan, and I’m suspicious of anyone who does. What I do have is a handful of questions I’ve started asking myself at the end of AI-heavy weeks. They aren’t rigorous. But they surface the pattern quickly enough to act on it.

One is the volume question. When AI halves the time for a particular task — drafting an email, summarising a document, coding a prototype — what happened to the saved time? Did it go into fewer total hours worked, or into more drafts, more summaries, more prototypes? In my own case, the honest answer is almost always the latter, and I have to keep reminding myself that’s a choice.

Another is the shape question. Have my days got smoother or choppier? AI makes it easy to run several threads at once: a draft being reviewed here, a script being tested there, a research summary coming back from an agent in the background. The output goes up. The cognitive load goes up more. If my days feel more like air-traffic control than focused work, that’s a signal, not a feature.

And then there’s the depth question. The creative dividend of AI, if you want it, is the space it creates for the slow, harder-to-automate thinking. Strategy. Reading widely. Sitting with a problem long enough to see it from a different angle. If I can’t remember the last time I did any of that, my reclaimed time isn’t being reclaimed. It’s being topped up with more of the same.

These questions don’t give you an answer. They give you a diagnostic. For me, the act of asking them is usually enough to reveal that I’ve drifted back into the Jevons pattern without noticing.

The question underneath the question

Most of the coverage around AI and work still assumes the central question is “will AI take my job?” I can see why. That’s the one that hits the emotional circuitry hardest.

But for most knowledge workers, that’s probably not the question that’s going to define the next few years. The more consequential one is this: now that AI has made huge chunks of your job faster, what are you doing with the time?

Harford’s instinct is that collectively, we’ll mostly do what we’ve always done. Absorb the gains into higher output and carry on.

The individual version of the answer can be different. But only if you’re deliberate about it. The default path is the historical path, and the historical path hasn’t changed in two hundred years.

Keynes’ grandchildren didn’t get their fifteen-hour week. There’s a version of the story where ours don’t either. And there’s a version where a few people, quietly, work out how to actually take back some of the time while everyone else is busy filling theirs.

Which version you end up in isn’t, for once, mostly up to your employer.