Inside the revolt at Meta's Applied AI unit

Meta's 6,500-person Applied AI unit is in open revolt after thousands of engineers were involuntarily reassigned to data-labeling and RLHF work. Engineers dubbed themselves 'draftees,' a surveillance system tracked keystrokes for AI training data, and a perverse token-usage incentive called 'tokenmaxxing' led to sloppy AI-generated code. The dysfunction culminated in a major Instagram account-takeover outage, the departure of Meta's CISO, and CTO Andrew Bosworth calling the reorg 'atrocious' with morale at a 20-year low. Zuckerberg has since promised no further layoffs in 2026 and capped manager-to-report ratios, but a surge in engineers signing up for interview-prep services signals a talent exodus. The broader context: Meta's $14.3bn bet on Scale AI and founder Alexandr Wang reflects a strategic wager that training data and distribution matter more than model benchmarks — even as the human cost inside the engineering org mounts.