Meta is facing significant internal turmoil as leadership aggressively reassigns thousands of engineers to AI data-labeling tasks, sparking widespread staff dissatisfaction and a series of high-profile service outages.

Key Points

• Meta has reassigned 30% to 50% of engineers from core product and infrastructure teams to a 6,500-person Agent Data Optimization (ADO) unit.
• The company implemented mandatory, non-opt-out tracking of employee keystrokes and mouse clicks to generate AI training data, though it later introduced limited pause controls.
• A major security breach occurred in May 2024, allowing attackers to hijack high-profile Instagram accounts due to vulnerabilities in AI-managed support flows.
• Internal performance reviews now incentivize "tokenmaxxing," leading employees to prioritize AI-generated output over traditional software engineering quality and stability.
• Meta’s Chief Information and Security Officer, Guy Rosen, resigned following the internal reorganization and subsequent security failures.

Why it Matters

The shift from a high-performance engineering culture to a rigid, AI-focused cost center threatens Meta’s long-term stability and ability to retain top-tier technical talent. This "AI psychosis" risks creating a fragile infrastructure where automated systems prioritize speed and metrics over the security and reliability of core social platforms.