The four-day work week that never came: what OpenAI, Anthropic, and Meta insiders say about AI and working hours

Four years ago, a Google engineering director publicly predicted that by 2025, AI would push companies to a four-day work week. Earlier this year, OpenAI picked up that thread and formally urged enterprises to trial a four-day week (with no change in pay), arguing that AI would soon accelerate enough human labor to a single corporate day that the rest of the corporate world should get ready.

A long-form report published by BBC on August 10 places that promise next to the reality: current and former employees at OpenAI, Anthropic, Meta, and Google almost unanimously give the opposite answer — AI has not reduced working hours, it has stretched them.

OpenAI / Anthropic: 90-hour weeks during sprint are normal

A former OpenAI technical employee who left last year told the BBC he worked "at least 70 hours a week" at OpenAI, far more than at any of his previous tech jobs. He now works at another AI startup and puts in "closer to 50-60 hours a week, outside of sprints." He described OpenAI's culture as "frequent crisis meetings, working weekends, super cut-throat performance reviews that would see colleagues suddenly let go," and was blunt: the company never actually trialled the four-day work week it asked others to try.

OpenAI and Anthropic push "sprints" to an extreme. In the weeks before a new feature or product release, hours can run "over 90 hours of work in a seven-day period." Neither company responded to BBC requests for comment.

Meta: engineers get "drafted" into AI teams with no choice

At Meta, employees describe a phenomenon they call being "drafted" — being abruptly reassigned to AI projects treated as urgent, without being given a choice. "They just move you over," a current employee and a former employee told the BBC. "You can't say no — or if you do, you have to quit."

Meta's internal AI work includes a software-engineering AI tool and infrastructure for measuring how well models replicate human tasks. "You're literally working in teams of people trying to replicate humans doing jobs," the former employee said. "It seems endless." Meta has laid off roughly 1 in 10 of its employees this year.

Google: former employee Amin Shali

Amin Shali, a former Google employee, left the company in May. Among the reasons: AI changed how his job worked. Internal engineering systems kept failing at night because critical resources — compute, memory — were being redirected to AI projects. He told the BBC that since leaving, his sleep and overall health had improved, and that heavy reliance on AI in big tech "creates a bad culture with excess pressure on engineers."

UC Berkeley eight-month study: time saved is time consumed

A UC Berkeley study tracked hundreds of employees at a US tech company for eight months. It found that after using AI, workers "worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day." Neil Thompson, an innovation scholar at MIT, gave a clear framing: "Even if there were real time savings, it would be sucked up by the changes, and implementing them, and making sure they worked." In other words, AI does not make work disappear — it makes "work that needs to be checked" multiply.

Thompson's summary is worth remembering: "People assume that 20% less work means four-day weeks. But new work emerges."

Editorial: this is a product problem, not just a labor problem

If you read this as the same old "tech overwork culture" story, you miss the point. The sharp part is:

  • The people who tell the story are the people working the hours. Publicly advocating for "AI lets humans work four days a week," while internally running 90-hour sprint weeks, is not a contradiction — it is a dual KPI: external productivity revolution, internal endurance test.
  • "Time saved" is structurally consumed. Berkeley's findings do not point to individual cruelty; they point to the integration shape of AI tools. As long as AI output still needs human verification, validation, and debugging, every new token is generating new work. This is not a software bug — it is a deployment form.
  • The four-day-week promise was never a promise, it was sales copy. Convincing enterprises that they will not have to pay for as much human labor is the AI giants' customer-acquisition playbook; making their own engineers work 70–90 hours a week is the other side of the same budget.

If the AI industry wants to keep "improving human wellbeing" on its homepage, the contrast at the heart of this BBC story — a four-day work week on stage, a 90-hour sprint week backstage — is the question it has to answer.

Sources: BBC News report "Tech leaders say AI means less work - their staff say they work up to 90 hours a week" by Kali Hays; Chinese reprint on Solidot "科技公司高管称 AI 将会减少工作时间，实际上 AI 增加了工作时间".