AI Was Supposed to Give Us a Four-Day Week. Inside Big Tech, Some Staff Are Logging 90 Hours Instead.

Key points

• OpenAI urged other companies to trial a four-day work week in 2024, but a former employee told the BBC the company never ran that trial internally.
• Workers at OpenAI and Anthropic described sprint periods, intense crunch phases before a product ships, that can reach 90 hours in a single week.
• A UC Berkeley study tracking hundreds of workers over eight months found that AI tools made employees work faster, take on more tasks, and extend work further into evenings.
• Meta employees said they were "drafted" onto AI teams without a choice, then worked nights and weekends on call.
• Anthropic CEO Dario Amodei has warned that more productive AI could cause broad job losses across the wider economy.

The pitch has been consistent. Build better AI tools, and humans work less. In 2021 a Google engineering director predicted that AI would deliver a four-day working week by 2025. OpenAI went further, formally urging companies to trial that shorter week, with no pay cut, arguing that AI would soon handle enough human labour to make it possible.

A former OpenAI technical employee, who left the company last year, told the BBC that OpenAI itself never ran that experiment. What they described instead was 70-hour weeks, weekend catch-up sessions, crisis meetings, and performance reviews sharp enough that colleagues vanished overnight.

"You go in on Saturday or Sunday just to catch up or make sure things aren't broken," the person said.

What is actually happening inside these companies?

Workers at OpenAI, Anthropic and Meta say the reality is long hours, not short ones. Sprint periods, intense crunch phases in the weeks before a product ships, can stretch to many weeks and top 90 hours in a seven-day period at both OpenAI and Anthropic, according to workers who spoke to the BBC. Neither company responded to requests for comment.

At Meta, staff described being "drafted" onto urgent AI projects this year without any real choice. One current and one former employee said the message was blunt: move to the AI team, or leave.

"You can't say no," the former employee said. "Or if you do, you have to quit."

Meta has since softened that policy, but many workers had already been reassigned by the time it did. A Meta spokesperson declined to comment.

Does the research back any of this up?

Yes, and it cuts against the shorter-week promise. A UC Berkeley study that followed hundreds of workers at a US tech company for eight months found that AI tools caused employees to work at a faster pace, take on a wider range of tasks, and push work further into the evening. The tools saved time in narrow spots, but that time filled back up.

Neil Thompson, an innovation scholar at the Massachusetts Institute of Technology, explained the pattern plainly. Even when AI creates genuine time savings, those savings tend to get absorbed by the work of deploying the tools, checking their output, and fixing what breaks.

"People assume that 20% less work means four-day weeks," Thompson said. "But new work emerges."

That checking burden matters. AI tools, including large language models (the technology behind chatbots like ChatGPT and Claude), can produce plausible-sounding but wrong outputs. Someone still has to verify the result.

What does this mean for people outside big tech?

For workers beyond Silicon Valley, two things are worth watching. First, Anthropic CEO Dario Amodei has warned publicly that as AI becomes more capable, broader job losses are likely across the economy. Second, the US has no legal cap on working hours for adults; the UK and Europe limit the working week to 48 hours including overtime.

Former Google engineer Amin Shali left the company in May after AI projects consumed resources that other internal systems depended on, forcing him to work through the night when those systems failed. Since leaving, he said his sleep and health have improved. His conclusion: heavy reliance on AI in large companies "creates a bad culture with excess pressure on engineers".

Common questions

Does using AI at work actually save time?

Sometimes in the short term, but research suggests those savings tend to fill with new tasks. The UC Berkeley study found that workers using AI tools took on broader workloads and longer hours, not shorter ones.

Could laws limit how long AI workers are made to work?

In the US, no federal law caps working hours for adults over 16. Workers in the UK and EU have a legal 48-hour weekly limit including overtime, which gives them at least some floor that American workers currently lack.

Should I worry that AI will take my job?

Anthropists and economists are genuinely split. Anthropic's CEO has warned of significant job displacement as tools improve. Other researchers argue new kinds of work tend to appear. What is clear is that this is an active debate with real stakes, not a settled question.