Meta’s Project OT Hits Agentic Wall

Project OT is a sharp warning that AI-generated activity is not the same as organizational productivity. Meta scaled agentic automation before proving the reliability, supervision, and measurement systems needed to support it.

• –Internal code changes reportedly rose 220% year over year, while user-facing feature changes increased only 36%—a huge gap between output volume and delivered value. [Ars Technica](https://arstechnica.com/ai/2026/08/metas-scrapped-plans-to-go-ai-native-included-slashing-teams-by-60-percent/)
• –Major technical and security incidents rose 40%, while employee time spent resolving them increased as much as 70%, suggesting autonomy shifted work into firefighting. [Ars Technica](https://arstechnica.com/ai/2026/08/metas-scrapped-plans-to-go-ai-native-included-slashing-teams-by-60-percent/)
• –The developer lesson is straightforward: agent deployments need bounded permissions, sandboxing, observability, regression tests, and rollback paths before headcount becomes the optimization target.
• –Meta’s public AI strategy still emphasizes agentic coding and computer-use systems, making Project OT less a rejection of agents than a failure of premature organizational redesign. [Meta AI](https://ai.meta.com/llama/?ss_ad_code=m_FB)