The four-day week has spent years as a slogan in search of evidence. Then, in 2022, a coordinated trial gave it something closer to data. Around sixty companies in the United Kingdom, employing roughly three thousand people, ran a six-month pilot organised by 4 Day Week Global and the think tank Autonomy, with researchers from Cambridge and Boston College measuring the results. When the findings were reported in early 2023, the headline was hard to ignore: the large majority of participating companies chose to keep the four-day week after the trial ended.

That is a striking result, and it deserves to be read carefully rather than cheered. The interesting question is not whether the trial was a success on its own terms. It plainly was. The interesting question is why it worked, because the mechanism behind the result is the same one that makes a shorter week realistic for an individual working alongside AI.

What the trial actually measured

The 2022 UK pilot followed a simple model, often called 100-80-100: full pay, eighty percent of the hours, on the condition that workers maintained one hundred percent of output. Over six months, companies tracked revenue, retention, sick days, and a battery of wellbeing measures.

The reported outcomes were consistent and, in places, large:

  • Burnout fell, and self-reported measures of anxiety, fatigue, and sleep problems improved.
  • Staff turnover dropped, and the number of employees leaving fell sharply against the prior period.
  • Revenue, on average, broadly held over the trial window, with many firms reporting it roughly flat or modestly up.
  • At the end, most companies continued. The large majority kept the four-day week, and a substantial portion said the change was permanent.

Those are good numbers. They are not, however, proof of a universal law, and it matters to say so plainly.

The honest caveats

The companies in the pilot were self-selected. They volunteered, which means they were already sympathetic to the idea and probably already suspected their weeks contained slack. They skewed smaller, and they skewed toward knowledge work and services: marketing agencies, software, professional firms, nonprofits. The sample under-represents manufacturing, shift-based care, hospitality, and the kinds of work where output is tightly coupled to hours on a line or hours at a bedside.

The measurement also leaned heavily on self-report. Wellbeing surveys are valuable, but they are not the same as an objective productivity audit, and revenue held over six months is a shorter horizon than a sceptic would want before declaring the effect durable. There was no randomised control group running the old five-day pattern alongside the trial, which means we cannot fully separate the effect of fewer hours from the effect of being a motivated company that volunteered for a closely watched experiment. The Hawthorne effect, where people work differently simply because they know they are being observed, is a real possibility here, and the researchers were appropriately cautious about it.

The four-day-week trials are strong evidence that a shorter week can work for a particular kind of company. They are not evidence that any company, anywhere, can simply drop a day. Treating promising results as settled proof is how good ideas get discredited.

None of this undermines the finding. It locates it. The pilot demonstrates that for a large class of knowledge-and-service organisations, a substantial cut in hours is compatible with steady output and better-off people. That is a meaningful claim, and it is the one worth building on.

The mechanism nobody put on the poster

Here is the part that gets lost in the headline. The companies that succeeded did not, for the most part, take their existing five days of activity and compress it into four. Compression is exhausting and it does not last. What they did first was cut.

Reports from the trial describe the same pattern again and again. Meetings were the first casualty: shortened, made optional, or deleted outright. Standing meetings that existed mostly to reassure people that work was happening were the easiest to lose and the least missed. Status updates moved to writing. Interruptions were corralled into windows. Focus time was protected because, with a day gone, there was no longer room to waste an afternoon.

In other words, the constraint did the work. Once a day disappeared from the calendar, low-value activity could no longer hide inside the slack. The week had to be examined, and examination is precisely what most of us never do when time feels abundant. The four-day week functioned less as a wellbeing intervention and more as a forcing function for elimination.

Why the constraint matters more than the count

If the gains came from cramming, the four-day week would be a story about heroic effort, and effort does not scale. But the gains came from removal, and removal is durable. You only have to delete a recurring meeting once. You only have to kill a report nobody reads once. The saved time keeps returning, week after week, without further willpower.

This is the single most important thing the trials proved, and it is rarely stated this directly: a hard limit on available hours is what finally makes an organisation confront the work that was never worth doing.

From the company to the individual

This is where the four-day-week evidence connects to the larger argument of this book. The trials proved, at company scale, the move I call Shed: the deliberate elimination of low-value work before any attempt to do the remaining work faster. A whole organisation, given a constraint, found that a meaningful fraction of its week was not load-bearing.

AI now extends the same logic to the individual. The trials needed an external constraint, a missing Friday, to force the audit. AI supplies a different lever: it can absorb a large share of the drafting, summarising, formatting, searching, and first-pass thinking that fills a knowledge worker’s day. The result is the same kind of slack the pilot companies discovered they had, except it is created by capability rather than by decree.

And it carries the same risk. The pilot companies that benefited defended the freed time. The danger, for an individual, is identical to the danger of any efficiency gain: time that is freed and not defended is simply refilled. Work expands to consume the hours available, and the gain evaporates into more meetings, more channels, more output that nobody asked for.

That is why the book’s framework does not stop at Shed. The five moves are See, Shed, Shift, Shield, Spend: see where the time actually goes, shed what does not earn its place, shift the rest to AI and to better systems, shield the recovered time so it is not reabsorbed, and spend it deliberately. The four-day-week trials are a real-world demonstration of the first two moves working together at scale. You can read the full sequence in the method.

What the evidence supports, and what it does not

Read soberly, the four-day-week experiments support a precise and useful conclusion. For knowledge-and-service work, a substantial reduction in hours is achievable without loss of output, provided the reduction is paired with genuine elimination of low-value work and the recovered time is protected. The trials do not support the fantasy of doing the same volume of busywork in less time through sheer compression. They support the opposite: that much of the busywork was never necessary, and a constraint is what reveals it.

This is exactly the bet the book is making about AI. The constraint that the pilot companies imposed by removing a day, the individual can now impose by handing genuine work to a capable tool and refusing to let the calendar reclaim the difference. The four-day week is not a curiosity from a few sympathetic agencies. It is a proof of concept for a defended, deliberately shortened week, and it points directly at the discipline that makes one last: not working harder in the time you have, but being honest about how little of it the work actually needs.