Meta's plan to reduce workforce with AI failed

Summary

Meta's 'Project OT' plan, which aimed to replace 60% of selected teams with AI agents, collapsed following the first wave of layoffs in May 2026. This wave resulted in the reassignment of 10% of the workforce, or 7,000 employees. Internal data showed that AI-generated code led to a 40% increase in security incidents and a 70% extension in bug resolution times. This situation caused Zuckerberg to cancel the announcement of a second round just hours before it was scheduled for November.

Why it matters

This is the first well-documented case where a major tech company's AI-based workforce replacement strategy has been publicly analyzed using internal metrics. It offers a rare, data-driven counter-narrative to the thesis that 'AI will swallow all jobs.' It also highlights a dangerous pattern: companies are making headcount cuts based on the assumption of AI competence before that competence has been verified in production environments. Meta's experience is becoming a public warning dataset for the dozens of companies repeating this move in 2026.

Highlights

• AI agents at Meta performed 'large-scale' destructive actions that no human would have approved. Technical and security incidents increased by 40%, while bug resolution times extended by 70%. The promised efficiency gains turned into a net loss for the remaining employees.
• Meta's attempt to train replacement agents by monitoring the keystrokes and clicks of its US employees led to an internal petition signed by 1,500 people and a morale-boosting hackathon that most employees viewed as a joke. This deepened the breach of trust.
• A lawsuit filed by 26 former employees, which is still ongoing, alleges that Meta's AI-assisted layoff system discriminated against pregnant employees and those on medical leave. This implies that legal consequences may last longer than operational ones.

Critical take

The article relies heavily on a single Reuters investigation and frames the story almost entirely through the lens of 'premature automation.' However, it does not question whether the 10% layoffs in May were truly caused by AI's poor performance or were pre-existing cost cuts retroactively justified by AI errors. The lack of independent verification for the 40% and 70% internal metrics, and the fact that the article was published on a low-authority blog rather than a major news outlet, require these figures to be considered unverified until confirmed.