AI companies push four-day weeks while staff work 90-hour sprints

Technology executives have spent years promising that artificial intelligence will reduce working hours, with some predicting four-day work weeks by 2025. OpenAI formally urged companies to test shortened schedules earlier this year, claiming AI would soon handle enough human labor to make it feasible.

Yet workers inside the companies building these tools describe a starkly different reality, according to reporting first published by the BBC.

A former OpenAI technical employee told the BBC the company never trialed the four-day work week it recommended to others. Instead, they described a work culture marked by frequent crisis meetings, weekend work, and aggressive performance reviews that resulted in sudden terminations. The employee reported working at least 70 hours weekly—far more than in previous tech positions.

The sprint that never ends

At OpenAI and Anthropic, development sprints—intense periods leading up to product releases—can stretch for weeks and exceed 90 hours of work in seven days, according to tech workers interviewed by the BBC. Neither company responded to requests for comment.

Meta employees reported being "drafted" onto urgent AI teams this year without choice. "You can't say no—or if you do, you have to quit," one former employee said. Workers on these teams regularly work nights and weekends while remaining on call during off hours.

The pressure extends beyond AI development teams. Amin Shali, a former Google employee, left in May after experiencing frequent overnight work caused by internal engineering failures—problems he attributed to Google redirecting crucial resources toward AI projects. Google declined to comment.

Research contradicts efficiency promises

New research from UC Berkeley tracked hundreds of workers at a U.S. tech company over eight months as they adopted AI tools. The study found employees "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, explained that even genuine time savings get absorbed by implementation demands and output verification. The Berkeley research confirmed that workload expansion stems partly from constant checking of AI tool outputs.

When processes do become more efficient, workers typically fill saved time with additional tasks—either voluntarily or to demonstrate value to employers. "People assume that 20% less work means four-day weeks," Thompson said. "But new work emerges."

Why it matters

The disconnect between AI companies' public messaging and internal work culture reveals a fundamental tension in the technology industry's transformation. As these firms race to develop tools they claim will liberate workers from excessive hours, their own employees are experiencing the opposite—raising questions about whether AI deployment will genuinely improve work-life balance or simply intensify productivity demands across industries. For business leaders evaluating AI adoption, the experience inside leading tech companies suggests implementation costs and oversight requirements may offset efficiency gains.

These details were first reported by the BBC.