tech
Inside the revolt at Meta's Applied AI unit
It takes a particular kind of dysfunction for an employee to hijack a company-wide livestream and demand the hosts pass a message to a senior executive: that he is, in not so many words, garbage.
TL;DR
- Meta's Applied AI unit is suffering from severe dysfunction, with employees calling it a "gulag" due to mandatory reassignments for data labeling.
- A significant portion of Meta's engineers have been moved to data labeling roles against their will, impacting core product, infrastructure, and security teams.
- Employees protested surveillance systems tracking their activity for AI data, though some opt-outs were later introduced.
- A culture of "tokenmaxxing," where engineers burn AI tokens for performance reviews, has led to wasteful spending and perverse incentives, potentially prioritizing AI-generated code over human-written code.
- The internal chaos has contributed to major outages, including Instagram account takeovers, and a decline in security and product reliability.
- Meta's CTO described the AI reorganization as "atrocious" and morale as near the worst in 20 years, comparable to the Cambridge Analytica scandal.
- Mark Zuckerberg has acknowledged mistakes, promised no further company-wide layoffs for 2026, and is attempting to address issues like manager-to-report ratios and team morale, but an exodus of engineers is reportedly underway.
- Despite internal turmoil, Meta's advertising business is booming, driven by AI-powered tools, leading some to question if the market is overlooking the company's strategic bet on data and distribution over models.
- Industry figures warn of a wider trend of "AI psychosis," where companies prioritize rapid AI development at the expense of engineering culture and reliability, with Meta as an extreme example.