The AI four-day work week has become the tech industry’s favourite promissory note: endlessly issued, never quite cashed. OpenAI, Anthropic, Meta and Google have all, in one form or another, told the world that artificial intelligence will compress human labour and give people their time back. Their own employees tell a different story.
The AI Four-Day Work Week That Never Was
In June 2026, OpenAI published a formal policy paper titled ‘Industrial Policy for the Intelligence Age’, urging employers and unions to run ‘time-bound 32-hour/four-day workweek pilots with no loss in pay that hold output and service levels constant.’ It also proposed converting AI efficiency gains into better retirement matches, broader healthcare coverage, and subsidised childcare and eldercare.
Fine words. A former OpenAI technical employee who left the company last year told the BBC that during their tenure, the firm never actually trialled the four-day week it was recommending to others. What they experienced instead was frequent ‘crisis meetings’, weekend work, and ‘super cut-throat’ performance reviews that would see colleagues abruptly let go. ‘You go in on Saturday or Sunday just to catch up or make sure things aren’t broken,’ they said.
That gap between prescription and practice is the sharpest indictment here. OpenAI received more than 400 responses to its policy paper via a dedicated inbox before closing submissions to review potential grant recipients. Whatever the merits of the proposal, the company appears to have been advocating externally for conditions it had not established internally.
Sprints, Drafts, and the Reality Inside AI Teams
The former OpenAI employee said they routinely put in at least 70 hours a week, well above their hours in prior tech roles. Since moving to an AI start-up, they said their workload has eased to 50–60 hours a week outside of product sprints. At OpenAI and Anthropic, those sprints can extend for many weeks and top 90 hours in a seven-day period, according to tech workers who spoke to the BBC. Neither company responded to requests for comment.
At Meta, the pressure has taken a different shape. Workers this year described being abruptly reassigned to AI teams, a process they called being ‘drafted’ because refusal was not a realistic option. ‘You can’t say no, or if you do, you have to quit,’ one former employee said. Those teams work nights, weekends, and report feeling perpetually on call.
The scale of Meta’s AI push helps explain the urgency. The New York Times reported that alongside 8,000 layoffs announced for May 2026, Meta told employees that a further 7,000 would be reassigned to AI initiatives, and cancelled plans to fill 6,000 open roles. The company has forecast capital expenditure of between $125 billion and $145 billion for 2026 in connection with its AI infrastructure build. A Meta spokesperson declined to comment.
Meta had already cut approximately 10% of its workforce in February 2026, with roughly 7,500 roles eliminated, concentrated in recruiting, business operations, and engineering teams not directly building AI products. The logic is transparent: strip out labour not pointed at AI and redeploy the budget towards teams that are. Those teams then work around the clock to justify the reallocation.
Time Saved Gets Spent Again
My read is that the four-day work week argument misunderstands how productivity gains actually move through organisations. A UC Berkeley study tracking hundreds of workers over eight months at a tech company with broad access to generative AI found that employees ‘worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day.’ The research, published by UC Berkeley Haas and summarised in the Harvard Business Review in February 2026 under the title ‘AI Doesn’t Reduce Work, It Intensifies It’, found that constant checking of AI output was itself a new and expanding workload.
Neil Thompson, an innovation scholar at MIT, put the mechanism plainly. ‘People assume that 20% less work means four-day weeks,’ he said. ‘But new work emerges.’ Even where workers had genuinely automated parts of their jobs, Thompson noted, they tended to fill the recovered time with more tasks, either by choice or because they felt they needed to demonstrate value.
Former Google engineer Amin Shali left the company in May because, he told the BBC, AI projects were cannibalising the processing and memory resources that internal engineering functions depended on, forcing him to work through the night when systems failed. Since leaving, his sleep has improved. His conclusion: heavy reliance on AI ‘creates a bad culture with excess pressure on engineers.’ A Google spokesperson declined to comment.
The honest version of the AI work-reduction thesis is not that it is wrong in principle. It is that every efficiency gain is immediately absorbed by expanded scope, greater output expectations, and the overhead of managing the tools themselves. Until organisations explicitly choose to bank time savings rather than convert them into output, the four-day work week will remain exactly what it is today: a policy paper, not a timetable.
