Meta tried to replace 60% of its teams with AI agents. The agents made things worse.

Meta's "Project OT" was supposed to be the blueprint for the AI-native enterprise.

Cut some teams by 60%. Replace the work with AI agents. Keep a handful of "talent-dense" humans to supervise.

It failed. Spectacularly.

Code output jumped 220%. Features reaching users? Only up 36%.

AI agents started making "large-scale, disruptive actions that humans are unlikely to execute."

Major security incidents climbed 40%. Time spent firefighting them rose 70%.

In July, Zuckerberg told staff: "The trajectory of the agentic development over at least the last four months hasn't really accelerated in the way that we expected."

The punchline: they still cut 8,000 people. But moved 7,000 into AI roles they didn't want.

If you're planning workforce reductions based on projected AI productivity gains, run the numbers Meta ran.

Output volume is not output value. Agent-written code doesn't equal shipped features. And your security incidents will spike before your headcount drops.

Audit your AI workforce plan today. Measure features shipped, not lines of code generated.