We Built the Machines to Reach the Seventh Day. Then We Cancelled It.
Keynes promised a fifteen-hour week by 2030. It's 2026 and the hours haven't moved. The harder question isn't whether the machines will let us rest — it's whether we'd know how.
I need to own something before I make an argument out of it.
For about two years I have been telling anyone who would sit still that the machines were building me a day off. Not in a utopian way — I am not a person who thinks technology arrives bearing gifts. I meant it operationally. I automated my drafting. I automated my research triage. I automated the small administrative sludge that used to eat a good chunk of my evenings. I could show you the hours I saved on a spreadsheet (I did make the spreadsheet, which in hindsight tells you something).
Then I looked at what I did with them.
I filled every single one. Not with rest — with more work. Better work, sometimes, work I chose, but work. The tools handed me back my evenings and I immediately spent them. And when I finally noticed, the thing that unsettled me wasn't the busyness. It was that I couldn't clearly say what I had been planning to do with the time instead. I had spent two years arguing for a prize I had never once described.
That's the essay. Not "will the machines take our work" — we have argued that one to death, and I have taken my swings at it. The question underneath is older and stranger, and almost nobody in the AI debate is asking it: if the machines really do take the work, do we rest? Or do we come apart?
The oldest story we have ends with a day off
The creation story most familiar in the West runs six days of making, then a seventh of rest — and the interesting thing about it, for our purposes, has nothing to do with theology. It's a structural observation, and you don't need to believe a word of the story to notice it.
The seventh day is not the reward for the six. It's defined by the six. Take away the making and the resting doesn't become infinite — it becomes meaningless. A day off is a boundary, and boundaries need something on both sides.
Hold onto that, because the secular version of the same promise is the one we actually organized our economy around.
In 1930, with the world falling apart around him, John Maynard Keynes sat down and wrote something almost absurdly hopeful. In Economic Possibilities for our Grandchildren, he argued that the accumulation of capital and the compounding of technical efficiency meant "the economic problem may be solved, or be at least within sight of solution, within a hundred years." A hundred years from 1930 is 2030. That's four years from now.
And what would we do with all that solved-ness? Keynes was specific: "Three-hour shifts or a fifteen-hour week may put off the problem for a great while."
Fifteen hours. By 2030.
So — how are we doing?
A team of psychologists led by Seth Kaplan went back and checked the prediction against the record in Industrial and Organizational Psychology last year. Their finding is blunt: "full-time workers in high-income countries still generally work 35–42 hours per week on average," which is "more than double what Keynes projected." Their summary of the leisure half of his forecast is one of the flattest sentences I have read in an academic journal: the abundance of leisure time Keynes foresaw "seemingly has not materialized."
Here is what makes that a story about AI rather than a story about economics. Keynes wasn't wrong about the productivity. He was, if anything, conservative. The wealth arrived. The output per hour arrived. The thing that never arrived was the hour.
The dividend is real, and it is small, and we spend it instantly
Let's get precise, because precision is where this argument either survives or dies.
Economists Alexander Bick, Adam Blandin and David Deming ran one of the first nationally representative surveys of who actually uses generative AI at work, and they measured the time dividend directly. Their estimate: "a mean time savings of 5.4 percent among all genAI work users, which implies a mean time savings of 1.4 percent among all workers (including non-users)."
Read those two numbers carefully, because they get mangled constantly. 5.4 percent is not the workforce. It is the people who actually use the tools for work. Spread across everybody — the users and the plumbers and the nurses and the people whose jobs the chatbot cannot touch — it's 1.4 percent. That's the honest version. On a forty-hour week, for the users, we're talking about roughly two hours.
Two hours a week is not nothing. Two hours a week is a Thursday afternoon. So where does it go?
Researchers at UC Berkeley's Haas School of Business went and looked. Xingqi Maggie Ye, working with Aruna Ranganathan, spent eight months embedded in a 200-person technology company, watching how people actually worked and running more than forty interviews, and what they found was the opposite of liberation. (It's still described as in-progress research, so hold it a little loosely — but it is the closest thing we have to someone actually watching where the hours go.) Their sentence deserves to be quoted whole: "Employees worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day, often without being asked to do so."
Often without being asked to do so. Nobody mandated it. There was no cruel manager in this story. The company didn't even require AI use. People simply found that doing more had become possible, and so they did more, and the day got longer. When they wrote it up for Harvard Business Review in February, they gave it a title that saves me a paragraph: AI doesn't reduce work — it intensifies it.
I recognized myself in that study, which is an uncomfortable way to read research.
The man selling the machine says don't wait up
Now here is the part that moved this from an interesting pattern to something I wanted to write about.
You would expect the people building these systems to sell the dream. Abundance, liberation, the four-hour week, the end of drudgery — that's the pitch, and it has been the pitch since the first loom. So when the CEO of OpenAI was asked directly whether his technology would shorten the workweek, I expected the usual (some version of "the best is yet to come," delivered warmly, meaning nothing).
Instead, Sam Altman said the quiet part out loud on the Relentless podcast, in remarks published on 28 July: "somehow we never get the promise of the four-hour work week at mass scale in society." And then, removing any ambiguity about whether his product is the exception: "I don't expect AI to change that."
He went further. In a world past superintelligence — the thing his company exists to build — his forecast is that "we're all going to be much busier than we thought we were supposed to be."
Sit with that for a second. The person with the most commercial incentive on Earth to promise you a day off is telling you, unprompted, that there isn't one coming.
To his credit, he has been consistent about the awkward part. Speaking at a BlackRock summit in Washington back in March, Altman acknowledged the structural problem in a line I keep returning to: "if it's hard in many of our current jobs to outwork a GPU, then that changes." And on what to do about it, he was refreshingly honest — "I don't think anyone knows what to do".
So that's settled, then. The machines won't give us rest, we'd only fill it anyway, and we should all go back to work.
Right?
So let's say we get it. Would we actually be okay?
Here's where I have to argue against my own instinct, because my instinct — and I suspect yours — is that of course we'd be fine. Give me the fifteen-hour week and watch me flourish. I have books. I have people I don't see enough. I have a list.
The evidence is not that kind.
Start in a small Austrian village. The textile factory in Marienthal closed in 1930, and over the winter of 1931–32 a research team moved in to document what happens to a town that loses its work; Marie Jahoda, Paul Lazarsfeld and Hans Zeisel published the result in 1933. Everyone expected radicalization — an idle, impoverished workforce was supposed to organize and revolt. What they found instead was a "tired community" marked by resignation. As the University of Vienna puts it, Jahoda concluded that long-standing unemployment "did not, as had been assumed, lead to revolt but rather to resignation and political apathy."
Not rage. Not liberation. Apathy. The time was there, and it did not turn into leisure.
Jahoda spent much of her career explaining why, and her answer is the most useful idea in this entire debate. She argued that a job delivers one obvious thing — money, the manifest function — and quietly delivers five others that nobody puts in the contract. As a large German study lays them out: work provides "(1) collective purpose, the opportunity to contribute to a higher collective purpose… (2) status and identity… (3) social contacts, in the sense of regularly shared experiences and contacts with people outside the nuclear family… (4) required activity; and (5) a time structure for the waking day, the week, and the year."
Read that list again and ask a hard question: which of those does a paycheck replace?
Because that's the entire policy conversation about automation. Retraining, wage insurance, universal basic income — all of it is aimed at the manifest function. The money. And the money, it turns out, is the part we know how to fix.
Is this just theory? No — and this is the row in my notes that I trust most. A meta-analysis pooling 106 primary studies tested whether those latent functions actually predict mental health once you control for people's financial situation. They do. "All six manifest and latent functions emerged as significant predictors of mental health in this analysis." Status showed the largest employed-versus-unemployed gap (d = 0.94); time structure came in at d = 0.59. Together the manifest and latent functions explained roughly 19% of the variance in mental health.
To be clear about what that does and doesn't say: money still matters enormously, and nobody in this literature claims otherwise. The finding is that the other five things matter in addition — that if you replace the income and nothing else, you have solved part of the problem and left the rest standing.
And there's a harder data point. Researchers Alan Adelman and Plamen Nikolov studied China's New Rural Pension Scheme — a program that let rural workers stop working — and found something nobody wants to hear about a comfortable retirement: over roughly a decade, the program "led to a decline in cognitive performance by as high as almost a fifth of a standard deviation on the memory measures we examine." That is one specific pension program in one country, and I'm not going to inflate it into a universal law of retirement. But it points the same direction as everything else here. Unstructured time is not automatically restorative. Sometimes it costs you something.
Keynes, to his enormous credit, saw this coming in 1930. Buried in the same essay that promised the fifteen-hour week is a warning almost nobody quotes: "there is no country and no people, I think, who can look forward to the age of leisure and of abundance without a dread." His diagnosis of why still lands like a slap: "We have been trained too long to strive and not to enjoy."
He wasn't worried we wouldn't get the time. He was worried we wouldn't be any good at it.
But rest itself is not the problem — and here is the proof
If I stopped there, I'd be writing an argument for working forever, and that is not remotely where the evidence lands. So let me take my own counter-case apart.
Everything above describes people who had time imposed on them: a factory closed, a pension arrived, a job vanished. Time without structure, without status, without colleagues, without choosing it. Of course that hurts. That isn't a seventh day — it's a cancelled week.
What happens when you take the same reduction in hours and design it?
In 2025, Nature Human Behaviour published the largest trial of a four-day week run so far. Led by sociologists at Boston College, it reduced the hours of 2,896 workers across 141 organizations in six countries — Australia, Canada, Ireland, New Zealand, the UK and the US — for six months, with no cut in pay. Burnout fell by 0.44 on a five-point scale. Job satisfaction rose 0.52 on a ten-point scale. Mental health improved by 0.39 and physical health by 0.28, both also on five-point scales.
And afterwards? More than 90% of the companies kept it.
One caveat I have to hand you, because the rest of this piece has been strict about them: the companies volunteered. That is not a randomized trial, and as the write-up itself concedes, results from firms enthusiastic enough to sign up "might overestimate the true effect." The authors say plainly that randomized studies are still needed. So read the direction of travel, not the decimal places.
Same direction from a different experiment entirely — and this one comes with a disclosure I'd rather make than have you find. When OpenResearch gave a group of people $1,000 a month for three years against a control group receiving a nominal amount, the fear was that they'd stop working. They didn't. Recipients "worked an average of 1.3 fewer hours per week compared to control participants" — about eight fewer days a year — and were "10% more likely to be actively searching for a job." On the family side, the researchers report that "parents improved their parenting practices and made meaningful investments in their children and themselves" — while noting in the same breath that those changes "did not translate into measurable improvements in children's academic or parent-reported behavioral outcomes in the short run."
(The disclosure: OpenResearch was founded and funded by Sam Altman — yes, the same one. I'm citing the study of the man whose forecast I spent three sections arguing with. Weigh it accordingly; I'm using it for a finding that cuts against the pessimistic read, which is the direction where his funding gives him the least to gain.)
Be careful with that study, because it gets over-claimed in both directions. It is not evidence that people given money go off and enjoy abundant leisure. The hours went into job searching and caregiving. But that's precisely the point worth making: handed unconditional time and money, people did not rot. They reallocated toward things that look an awful lot like Jahoda's list — activity, purpose, people.
So the variable was never the number of hours. The variable is whether the structure survives the reduction. Fewer hours inside a job, a team, a rhythm and a role: better on every measure we know how to take. The same hours removed with nothing underneath: Marienthal.
Which brings me back to the seventh day, and to the thing I think we have genuinely misunderstood. That day was never a gift that fell out of an abundance curve. It was an institution — a rule, a boundary, socially enforced, applying to everyone at once so that nobody was resting alone. That's the piece we're not building.
Iceland didn't wait for the dividend. It went and took it.
Here is what I find bracing about Iceland: nobody there waited for the productivity fairy.
Between 2015 and 2019, Iceland ran two large-scale trials of a reduced working week, analysed jointly by the Icelandic research organisation Alda and the UK think tank Autonomy — which, in fairness, campaigns for shorter working weeks, so read them as an advocate with good data rather than a neutral referee. Public-sector workers moved from 40 hours to 35–36 hours "with no reduction in pay." The trials covered about 2,500 workers — over 1% of Iceland's entire working population.
The productivity result is the one that ends most arguments: "Productivity and service provision remained the same or improved across the majority of trial workplaces."
Then came the part that matters more than any trial. The unions took the results to the bargaining table, and since completion, "86% of the country's workforce are now working shorter hours or gaining the right to shorten their hours." Note the precision there, because it's often flattened in the retelling: that figure covers people who now work shorter hours and people who won the contractual right to. Either way, it wasn't a gift from a machine. It was negotiated.
Belgium took a different route to a related idea. In February 2022, the government of Prime Minister Alexander de Croo announced a four-day option and a right to disconnect, with the legislation itself following later that year. Workers can compress their week — though I should be honest that this is not an unconditional right; as labour minister Pierre-Yves Dermagne put it, the request comes from the employee "with the employer giving solid reasons for any refusal." Alongside it came the right "to turn off work devices and ignore work-related messages after hours without fear of reprisal."
Dermagne's justification is the most seventh-day sentence uttered by a modern politician: "The boundary between work and private life is becoming increasingly porous."
That's the whole thing, isn't it? Not how many hours. Where the edge is.
Neither country did this because AI arrived. They did it because someone decided rest was a thing you legislate and bargain for, like a wage or a safety rail — not a dividend you wait for.
Four ways the next decade could go
Let me get speculative, because I think the near future is stranger than either the doomers or the utopians are describing.
Imagine productivity abundance with no distribution mechanism. The output arrives. The economy is spectacular. And because there is no institution converting productivity into time — no seventh day, no bargaining table, no statute — the gains land as they always have: as capital returns and as competitive pressure. Your firm's AI makes your team twice as effective, so your firm's competitor's AI does too, and the equilibrium is not two days off. It's the same week at double the intensity. Altman has essentially told you this is his base case.
Imagine the two-tier seventh day. A slice of knowledge workers — senior, leveraged, hard to replace — negotiate genuinely shorter weeks and take them. Everyone else absorbs the intensification. Rest becomes a class marker, the way a private garden did in an earlier century. We'll have a word for it (something aspirational, probably borrowed from Scandinavian). It will be a nice word.
Imagine the loneliest version. Automation does reach deep, the political system responds with the thing it knows how to do — a transfer, a cheque, an income floor — and it works, materially. Nobody is hungry. And we discover, at scale, what Marienthal discovered in a village: that we replaced one of the six functions and left five vacant. A generation with money, time, and no time structure, no collective purpose, no colleagues, no status, no required activity. The policy will be declared a success by every metric we thought to measure.
Now imagine the good one, because it is just as plausible and nobody spends enough time on it. The gains show up and we treat them the way Iceland treated theirs — as something to be claimed rather than received. Shorter weeks bargained sector by sector. A right to disconnect that has teeth. Public institutions — libraries, clubs, sports leagues, choirs, workshops, civic bodies — funded as seriously as retraining programs, because we finally understood that they are the infrastructure of the seventh day. Not entertainment. Infrastructure.
The difference between the third scenario and the fourth is not the technology. It's identical technology. The difference is whether anyone built the institution.
What the smart people are saying
The encouraging thing is that this has stopped being a fringe conversation, and it's converging from opposite ends of the political spectrum.
From the center-left, the Brookings Institution published a workforce framework for the AI transition in June by Xavier de Souza Briggs, organizing the response into brakes, steers, buffers and shifts. The "shifts" category is where it gets interesting: rethinking "the role of paid work, length of the work week, and who shares in the wealth AI produces." Briggs is explicit that a cheque alone won't cover it — noting that "some proposals would redistribute wealth and not address access to good work, at least not directly."
From the right, Arthur C. Brooks — writing from the American Enterprise Institute on what he calls the dignity deficit — has been making a version of this argument for years, and his formulation is the tightest I've found.
"We feel a sense of dignity when our own lives produce value for ourselves and others," he wrote. "Put simply, to feel dignified, one must be needed by others."
A Brookings scholar and an AEI scholar agree on the diagnosis: the paycheck is not the payload. When those two institutions land in the same place, the finding is usually real and the disagreement is about instruments, not facts.
And from the academic side, Kaplan and his colleagues put the puzzle plainly — a century on, the abundance of leisure "seemingly has not materialized," and the reasons are psychological and social as much as economic.
What does this mean for you?
I'm wary of ending a structural argument with self-help. Most of this needs bargaining tables and statutes, not personal grit. But there are things genuinely inside your control, and they are not trivial:
Audit where your dividend went. For one week, note what you actually did with the time your tools saved you. Not what you meant to do. If you can't find the hours, that's the finding — and it's the same one I got.
Give the time a destination before you save it. "I'll have more time" is not a plan; it's a vacancy, and work is extremely good at filling vacancies. Name the thing first. Then automate.
Defend an edge, not a quantity. Belgium's insight is better than a target number. A hard boundary — a device off at a time, a day nobody schedules over — protects rest more reliably than hours ever will.
Bank the five things, not just the money. If your work winds down — by automation, retirement, or choice — deliberately replace the latent functions. A standing weekly commitment (time structure). Something with other people in it (social contact). Something where you're needed (status). Something bigger than you (collective purpose). Something that demands doing (activity). Do this before you need it.
Treat "rest" as a civic question, not a lifestyle one. When your union, your professional body or your representative discusses AI, notice whether anyone is talking about hours and boundaries — or only about retraining and income. Ask the other question.
Be suspicious of the abundance pitch — including mine. The most credible people in this industry aren't promising you a day off. Believe them over the marketing.
The lesson, as I see it
We told ourselves a story where the machines do the work and we inherit the rest, and we have been telling it since the first spinning jenny. It has never once come true. Not because the productivity failed — it arrived exactly as promised, and then some — but because nobody ever built the thing that turns productivity into time. Rest doesn't fall out of an economy. It gets legislated, bargained, defended, and practiced.
So: will we rest when the automata can do everything? On the current trajectory, no. We'll do what the Berkeley researchers watched people do — take the saved hour and spend it, faster, on more.
And would we suffer if we got it anyway, handed down as a transfer with nothing underneath? The Marienthal evidence says some of us would, and it says so with a century of corroboration behind it.
But that's a verdict on our institutions, not on human nature. Given fewer hours inside a structure, people got measurably healthier and 90% of their employers never went back. Given money and time with no conditions, people didn't collapse — they went looking for better work and invested more in their children. We're not fragile. We're just badly scaffolded.
The old story got the architecture right and we've been reading it backwards for two centuries. The seventh day isn't what you get once the work is finished. It's a wall you build on purpose, while the work is still going, because you decided the week should have an edge in it.
My vote? Stop waiting for the machines to hand us the day. They won't — the man selling them just told us so, on the record. Build it ourselves, the way Reykjavík did: on purpose, in writing, for everyone at once.
Otherwise we'll automate all six days, arrive at the seventh with nothing planned, and call the emptiness progress.
If you've been meaning to take an afternoon back, consider this your nudge — and then tell someone else why you did. The HAIA Foundation is here to argue that the human part of this is worth designing for, not just budgeting around.