BBC Reports 90-Hour AI Sprints at Major Tech Firms
BBC reporting published on August 10 described current and former workers at OpenAI, Anthropic, Meta, and Google putting in long weeks, including 90-hour release sprints. Separate UC Berkeley research at a 200-person technology company found that generative AI broadened task scope and extended work hours, supporting the work-intensification pattern but not independently verifying the BBC's company-specific accounts.
Current and former employees at several major AI companies described workweeks far beyond the standard 40 hours in an August 10 report from the BBC. The outlet said people it interviewed described at least 70-hour weeks at OpenAI, release sprints at OpenAI and Anthropic that could exceed 90 hours in seven days, and nights, weekends, or on-call expectations around urgent AI work at Meta and Google.
The accounts complicate public claims that AI productivity will quickly translate into shorter workweeks. The BBC noted that OpenAI had encouraged employers to test a four-day, 32-hour week without a pay cut. A former OpenAI technical employee told the outlet that the company had not run that trial while the person worked there and instead described weekend work, recurring crisis meetings, and an intense performance culture.
OpenAI and Anthropic did not respond to the BBC's requests for comment. Meta and Google declined to comment on the reported working conditions, according to the article. The 90-hour figure therefore remains attributable to the workers interviewed by the BBC rather than an independently documented company schedule.
A separate study found broader work intensification
UC Berkeley researchers reported a similar pattern in an eight-month study at a roughly 200-person U.S. technology company. Based on observations and 40 in-depth interviews, the researchers found that workers using generative AI moved faster, took on a wider range of responsibilities, and let work spill further into personal time.
The Berkeley study did not examine the specific companies or verify the 90-hour sprint accounts in the BBC report. It also followed one employer, so its findings should not be generalized to the entire technology sector. Its value is narrower: it documents a mechanism by which time saved on individual tasks can be absorbed by additional work, more task switching, and continued monitoring of AI output.
What deployment teams should measure
For leaders rolling out AI tools, throughput alone can hide whether work is becoming sustainable. A stronger evaluation pairs delivery speed with total hours, after-hours activity, revision effort, defect rates, and worker-reported cognitive load.
The evidence does not show that AI must lengthen every workweek. It does show that faster task completion does not automatically become leisure. Without explicit staffing, workload, and off-hours boundaries, organizations can convert automation gains into a larger queue of work instead of a shorter day.
Key Points
- 1BBC interviewees described 70-hour weeks and release sprints exceeding 90 hours at major AI companies, but the company-specific accounts were not independently verified.
- 2UC Berkeley's separate eight-month study at one technology company found faster work, broader task scope, and work extending into more hours of the day.
- 3AI-deployment evaluations should pair throughput with total hours, after-hours activity, review burden, defects, and cognitive-load measures.
Scoring Rationale
The BBC's multi-company worker accounts and UC Berkeley's field research provide a timely operational warning that AI productivity can expand workload instead of reducing hours. The company-specific claims remain interview-based and the academic study covers one employer, limiting generalization.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
