The forty-hour week is not a law of nature. It is a law, passed in 1938, when the argument was about coal dust and child labor and nobody had yet invented the photocopier. It has outlasted the assembly line, the fax machine, the personal computer, the internet and the smartphone, every one of which was supposed to hand us our afternoons back.
Now the pitch is being made again, with artificial intelligence in the role previously played by all of those. Senator Bernie Sanders and Representative Mark Takano reintroduced the Thirty-Two Hour Workweek Act this month, explicitly framing it as the mechanism for making sure AI’s productivity gains land in something other than a shareholder’s pocket.
The evidence that shorter weeks work is stronger than it has ever been. The bill still probably will not pass, and the reason has almost nothing to do with whether the evidence is good.
The short version
- Sanders and Takano reintroduced the Thirty-Two Hour Workweek Act in September 2026. Takano first introduced it in 2021
- It amends the Fair Labor Standards Act of 1938, lowering the overtime threshold from 40 hours to 32 over a four-year phase-in
- It does not mandate a four-day week. It makes hours 33 through 40 more expensive for employers
- Pay and benefits cannot be cut to compensate, which is the provision employers will fight hardest
- A 2025 study in Nature Human Behaviour tracked 2,896 workers at 141 companies across six countries. Burnout fell, satisfaction rose, sleep improved
- Around 90 percent of companies in that trial kept the shorter week after it ended
- A 2023 analysis estimated AI could move about 35 million US workers, roughly 28 percent of the workforce, to 32 hours within a decade without losing output
- The catch: those gains are concentrated in white-collar work. Nurses, builders and line cooks do not get them
What the bill actually does
Almost every headline about this calls it a four-day week bill. It is not, and the distinction is the whole thing.
The Fair Labor Standards Act does not say you must work forty hours. It says that after forty hours, non-exempt employees are owed overtime. That threshold is what makes forty the default: it is the point where an additional hour stops being cheap. Move the threshold to thirty-two and you have not banned the fifth day. You have repriced it.
| What people assume it does | What it actually does |
|---|---|
| Mandates a four-day week | Lowers the overtime threshold to 32 hours. How you arrange those hours is not specified |
| Takes effect on passage | Phases down over four years before reaching 32 |
| Covers everyone | Applies to non-exempt employees. Exempt salaried workers are a separate question |
| Guarantees a three-day weekend | Nothing requires the reduction to be taken as a whole day off |
Read that last row again, because it is where the real-world version tends to land. An employer facing a 32-hour threshold can cut everyone to four days. It can also shave ninety minutes off five days, which satisfies the statute and delivers approximately none of the benefit people are imagining when they picture a long weekend.
The productivity argument, in one chart
The case Sanders makes is not mainly about rest. It is about where the gains from every previous efficiency leap ended up.
American worker productivity climbed 93.2 percent between 1979 and 2026. Hourly compensation for the average worker rose 33.7 percent. Whatever you think of the policy, that divergence is the argument, and it is not a rhetorical one. It is the thing that actually happened the last time output per hour went up sharply.
“At a time when artificial intelligence and robotics will radically transform our economy,” Sanders said alongside the bill, “it is imperative that the financial gains from this new technology benefit working families, not just a handful of billionaires and corporate CEOs.”
The counter-case is not that the gap is fake. It is that a national hours rule is a blunt instrument for closing it, and that the same technology being invoked as the source of the surplus is also being invoked, by the same companies, as the reason headcount is falling. Those two stories are difficult to hold at once, which is roughly the position a lot of workers are already in as they watch layoffs continue even where AI costs more than the humans it replaces.
The trial data is unusually good
This is the part that has changed since 2021. The four-day week used to be an argument. It is now a reasonably large body of evidence.
The strongest single piece is a study published in Nature Human Behaviour in July 2025, run by researchers at Boston College. It followed 2,896 employees at 141 organizations across Australia, Canada, Ireland, New Zealand, the United Kingdom and the United States, over six months, with everyone keeping 100 percent of their pay.
| Measure | Change over six months |
|---|---|
| Burnout | Down 0.44 points on a 1 to 5 scale |
| Job satisfaction | Up 0.52 points on a 0 to 10 scale |
| Mental health | Up 0.39 points |
| Physical health | Up 0.28 points |
| Sleep | 16 percent more than before |
| Companies that kept it | About 90 percent continued after the trial ended |
Boston College researchers, published in Nature Human Behaviour, July 2025. Six countries, 141 organizations, 2,896 employees, full pay retained.
That continuation rate is the number employers should look at hardest. Companies in these trials are not charities, and they are under no obligation to keep a policy after the researchers pack up. Nine in ten did anyway, which is a stronger signal about business outcomes than any self-reported wellbeing score.
The honest caveat is selection. Firms that volunteer for a four-day week trial are disproportionately white-collar, disproportionately small, and disproportionately run by someone who already liked the idea. Nobody has run this experiment on a hospital system or a regional trucking firm, and the results from a 60-person software consultancy do not automatically transfer.
The AI number underneath the bill
The specific claim linking AI to hours comes from a 2023 analysis by the think tank Autonomy, which estimated that productivity gains from large language models could let about 35 million US workers, roughly 28 percent of the workforce, move to a 32-hour week within a decade without losing pay or output.
Twenty-eight percent is a striking figure, and it is also the problem. It means 72 percent of the workforce is not in the group where the technology pays for the shorter week. The gains cluster in exactly the jobs that involve producing and processing documents, which is why any serious version of this conversation eventually runs into the same wall that a tool scoring individual jobs for AI exposure using real chat logs keeps illustrating: exposure is wildly uneven by occupation.
The sectors where the math does not work yet
A hospital cannot deliver the same care in 32 hours because a language model got better at summarizing. Neither can a construction site, a kitchen or a warehouse floor. In those sectors a 32-hour threshold does not convert productivity into time off, it converts directly into either overtime costs or more hires.
That is not a reason the policy is wrong. It is the reason it is hard, and it is why the bill’s opponents will lead with healthcare and construction rather than with software companies.
Why it stalls anyway
Set aside the merits. The obstacles are structural, and they are the same ones that killed it in 2021 and 2024.
The pay provision is the core of it. The bill requires that hours fall without compensation falling, which means employers absorb the cost rather than passing it to workers. Every business lobby in Washington will treat that as a straightforward labor cost increase, because that is what it is. Support in Congress for anything that raises labor costs during a period of active white-collar layoffs is thin regardless of party.
Then come the second-order interests, and they are larger than they look. Commercial real estate is priced on occupancy. Office development, transit, and the entire economy of lunch places that exist because a building fills up five days a week all have a stake in the fifth day. That coalition is exactly the one that fought remote work to a standstill, and it is better organized now than it was in 2020.
The realistic path is not federal. It is what has been happening anyway: individual employers adopting it because it helps them hire, a few states experimenting at the margins, and the standard drifting over a decade or two rather than changing by statute. That drift is already visible in places where the productivity dividend shows up first, which tends to be smaller firms rather than the large enterprises that get the headlines.
What to actually watch
- Whether the bill gets a hearing. Not a vote. A hearing. It has never had a serious one, and that is the tell for whether anything has changed
- Whether any large employer with hourly staff adopts it voluntarily. The evidence base is all knowledge work. One retailer or hospital system publishing real numbers would matter more than another software trial
- Whether the reduction gets taken as a day. If early adopters shave hours across five days instead, the policy delivers the statutory win and none of the thing people wanted
- Whether AI productivity claims survive contact with measurement. The 28 percent figure is a projection. The next few years will produce actual numbers, and they will either strengthen this argument or quietly end it
Nearly ninety years is a long time for a number to go unchallenged, and the case for changing it has never rested on better data than it does right now. That still may not be the relevant variable. The forty-hour week did not arrive because someone proved it was optimal. It arrived because a coalition got large enough to make it law, and then everything else arranged itself around the number. The question is not whether thirty-two would work. Most of the evidence says it would. The question is who has to be convinced, and right now the answer is the people paying for the fifth day.
