Tech leaders say AI means less work – their staff say they work up to 90 hours a week
Major AI companies including OpenAI, Anthropic and Meta have publicly claimed that artificial intelligence will reduce how much people need to work, with some executives suggesting a four-day week could soon become standard. However, a BBC investigation has found that employees at these same firms, including those building and using the AI tools in question, are actually working far longer hours than the typical 40-hour week, often exceeding 70 or even 90 hours during intense periods.
A former OpenAI employee said the company never trialled the four-day week it publicly urged other businesses to test, and instead described a culture of weekend work, "crisis meetings" and harsh performance reviews. Workers at OpenAI and Anthropic reported "sprints" lasting many weeks with over 90 hours worked in a seven-day period, while Meta staff described being involuntarily "drafted" onto urgent AI projects with no option to refuse. Neither OpenAI nor Anthropic responded to the BBC's requests for comment, and the contrast comes even as some of these firms, including Meta, have simultaneously cut jobs.
- AI firms predict shorter working weeks, but staff report the opposite
- Some OpenAI and Anthropic staff report sprints topping 90 hours weekly
- Meta employees describe being "drafted" onto urgent AI teams
New here? Start with this
Tech companies at the forefront of building artificial intelligence, such as OpenAI, Anthropic and Meta, have suggested publicly that AI will cut down how much people need to work, with talk of a four-day week becoming normal in future. This has drawn attention to how these firms treat their own staff, since they are the ones designing and using the technology said to be capable of easing everyone else's workload.
The BBC looked into working conditions inside these companies and spoke to current and former employees. It found accounts of long, intense working patterns, including extended "sprints" and periods where staff worked far beyond a standard 40-hour week, alongside descriptions of pressure to take on urgent projects and tough performance reviews.
The story matters because it highlights a gap between how AI companies talk about the technology's effects on work generally and the working lives of the people building it. It also comes at a time when some of these same firms have been cutting jobs, adding another layer to the debate about AI's impact on employment.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Critics argue this exposes a genuine contradiction: executives are publicly promoting shorter working weeks and framing AI as liberating humanity from toil, while privately extracting extraordinarily long hours from the very staff building that technology. They point out that some of these firms have simultaneously made redundancies, meaning fewer people are being asked to do more, which sits uneasily with talk of reduced workloads. From this perspective, the mismatch between public messaging and internal culture raises fair questions about whether such promises are genuine or simply favourable positioning, and workers describing involuntary "drafting" onto projects or punishing performance reviews suggest real personal cost behind the rhetoric.
The case against
A more sympathetic reading holds that building genuinely transformative technology at the frontier of a fiercely competitive industry is, almost by definition, a period of intense effort, much as early-stage start-ups or major scientific breakthroughs have historically demanded short bursts of extreme commitment from a relatively small team. Advocates of this view would distinguish between a temporary, self-selected sprint by highly paid specialists racing to ship a product, and the executives' broader, longer-term claim about how AI might eventually reduce hours across the wider economy once the technology matures and diffuses. On this account, there is no necessary hypocrisy in working intensely now to build tools that could plausibly reduce work for others later, even if communication about that distinction could have been clearer.