The four-day week meets AI: productivity data worth trusting

I distrust ideas that arrive pre-loved by everyone on LinkedIn, and by that standard the "AI will give us all a four-day week" take deserves quarantine. But underneath the posting, there are two genuinely decent bodies of evidence here — and they're more interesting read together than apart. So let's do the unfashionable thing and read them.

What the four-day week trials actually found

The landmark is the UK's 2022 pilot: 61 companies, around 2,900 workers, six months of 100% pay for 80% time coordinated by 4 Day Week Global with the think tank Autonomy and academics including Boston College's Juliet Schor. The results were strikingly positive — most firms reported revenue roughly stable, sizeable falls in burnout and turnover, and 56 of the 61 firms carried on after the trial; follow-ups a year later found the overwhelming majority still at four days. A larger second British pilot reported similar findings in 2025, and the pattern echoes earlier Icelandic public-sector trials.

Now the caveats, which the press releases skip. These firms volunteered — self-selected organisations with leadership already convinced and workflows amenable to compression. Output was largely self-defined and self-reported. Small firms and desk-heavy sectors dominate the samples. None of that makes the results wrong; it makes them results about willing, suitable firms. A hospital rota is not a marketing agency, and no trial yet says otherwise.

What the AI productivity studies actually found

The controlled evidence is more specific than the discourse suggests. A well-known MIT experiment found professionals completed writing tasks around 40% faster with AI assistance, with quality holding or improving. A large call-centre study measured roughly 14% more issues resolved per hour, with gains concentrated among newer staff. GitHub's controlled Copilot experiment clocked developers finishing a coding task about 55% faster. As of early 2026, the honest summary across studies is: double-digit gains on drafting-heavy, language-heavy tasks; smaller or murkier effects elsewhere; the largest boosts going to the least experienced — and a persistent gap between what people report saving and what output data shows, partly because checking AI output eats into the winnings.

Time saved by AI doesn't pool politely in a corner waiting to be redeemed. It evaporates.

Where the two datasets meet

Here's the arithmetic the enthusiasts are gesturing at. A four-day week asks you to produce five days of output in four — a 25% productivity lift. The AI studies show gains in exactly that range for drafting-heavy roles. So in principle, for a subset of jobs, the machine really can buy back the Friday.

The catch is the word "banked". Time saved by AI doesn't pool politely in a corner waiting to be redeemed; it evaporates into extra meetings, more polish, longer documents nobody asked for. Firms that got a shorter week out of AI did it deliberately: they measured the baseline, deployed the tools against specific processes, and converted the surplus into hours off rather than inflation of the work. That's an organisational design job — the same discipline we described in what AI-first companies actually change — and it's also where agents that take over whole task-chains matter more than chat windows.

Who this works for, honestly

Reading both evidence bases together, the four-day-plus-AI combination is most credible where three conditions hold: the work is drafting- and admin-heavy (the AI gains are real), output is measurable enough to prove nothing slipped (the trial design is possible), and management is willing to fix the meeting culture (the gains survive contact with the calendar). Professional services, marketing, software and chunks of the public-sector back office qualify. Frontline care, logistics and hospitality mostly don't yet — their productivity story runs through physical automation, on a slower clock.

The bottom line

Beware anyone offering the universal version of this story in either direction. The evidence, as of early 2026, supports something narrower and more useful: for a meaningful minority of workplaces, AI-driven gains are now large enough to fund a shorter week without heroics — provided the firm treats it as a measured operational change, not a vibe. Two overhyped ideas, one modest, well-supported conclusion. That's about as good as evidence about work ever gets.