In early June, an internal livestream at Meta erupted. Employees hurled profanities at the company on camera. The clip leaked and spread fast across tech circles. Soon after, Meta CTO Andrew Bosworth sent an internal memo owning up to the mess: management of the new Applied AI division had been 'terrible.'
6,500 engineers, one manager per 50
Launched in March, Applied AI is Meta's bet to turn generative AI models into real consumer products. It pulled roughly 6,500 engineers and product managers from across the company.
The structure was dysfunctional from day one. The initial setup was extremely flat: up to 50 employees per manager. In software, the norm is 6 to 8, with big players like Google and Amazon pushing to 12 or 15. A ratio of 50-to-1 meant no real career conversations and no one tracking individual output.
Worse, the work itself. Many engineers were assigned to generate puzzles and coding problems for training Meta's AI models — data-labeling tasks, not product building. Staff began calling the environment a 'gulag'; others dubbed themselves 'draftees,' emphasizing they had been forced into the unit.
Livestream hijack forces CTO to concede
The livestream incident drew heavy media coverage. In his memo, Bosworth wrote bluntly: 'We clearly failed to explain our vision and how we would support career transitions.' He also said: 'We broke your trust that your expertise and contributions would be valued.'
Fixes followed quickly: manager caps dropped from 50 to 20, and employees could apply to transfer to other roles outside Applied AI.
AI arms race tests management limits
The Applied AI chaos underscores a core contradiction in Big Tech's AI push: recruiting top engineers with the promise of building AI products, then giving them data-labeling work; building a team with a 'move fast' culture while overloading management. Scale can expand overnight; trust cannot.
Bosworth ended his memo with a pointed remark: 'AI won't replace your job, but someone who uses AI might.' It was a veiled message — accept the forced transition, or fall behind.