DEVOURED
Why is Meta destroying its engineering organization?
Meta’s aggressive pivot to AI has led to internal turmoil, widespread resignations, and a decline in engineering quality.
What: The report details how Meta forcefully reassigned thousands of engineers to menial data labeling for AI training, implemented keystroke tracking, and prioritized token-usage metrics over software stability, resulting in a major security outage.
Why it matters: This demonstrates the danger of 'AI psychosis,' where leadership prioritizes rapid AI deployment at the expense of core engineering culture, leading to the decay of resilient infrastructure.
Takeaway: If your organization is rapidly gutting senior engineering teams to pivot entirely to AI-generated code, anticipate increased production instability and monitor for the loss of institutional knowledge.
Deep dive
- Meta shifted its 'move fast with stable infra' culture to an AI-obsessed, metrics-driven environment.
- Roughly 4,500 software engineers were reassigned to data labeling and RLHF tasks.
- Engineers face surveillance (keystroke and mouse tracking) and hyper-aggressive performance reviews.
- Token-usage quotas incentivized 'tokenmaxxing' over quality.
- Infrastructure teams saw up to 50% headcount reductions, leading to a major Instagram security outage in May 2026.
- Meta's CPO Chris Cox and CTO Andrew Bosworth acknowledged the internal chaos following employee protests.
Decoder
- RLHF: Reinforcement Learning from Human Feedback, a method where humans rank AI outputs to improve model performance.
- SEV0: The highest severity outage incident, indicating a catastrophic system failure requiring immediate, emergency response.
- PSC: Performance Summary Cycle, Meta’s internal system for performance reviews and promotion calibration.
Original article
Full article content is not available for inline reading.
Explore all 53 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi