Tech leaders say AI means less work - their staff say they work up to 90 hours a week

gist

Kali Hays tests tech executives’ promise that AI will shorten work by comparing it with conditions inside OpenAI, Anthropic, Meta, and Google. Workers describe 70-to-90-hour weeks, prolonged release sprints, forced transfers, and failures caused by diverted infrastructure. Research adds a broader mechanism: AI expands task scope and verification work, while employers refill any saved time with more output. Productivity gains can therefore intensify work unless institutions deliberately convert them into leisure.

ideas

• Productivity does not allocate saved time. Employers can turn faster work into a broader workload instead of shorter weeks.
• AI creates implementation overhead. Workers must change processes, check model output, and repair failures before any time savings become available.
• Temporary urgency can become the operating model. Multiweek sprints and weekend work make release pressure continuous inside frontier AI efforts.
• AI priorities affect adjacent teams. Reallocating compute and storage can destabilize internal systems and push non-AI engineers into overnight recovery work.
• The evidence has a defined boundary. Anonymous worker accounts and one eight-month company study document mechanisms and experiences, not sector-wide prevalence.

quotes