Meta's AI agents wrote 220% more code. Only 36% shipped. Zuckerberg killed the second layoff wave.

Meta planned to replace thousands of workers with AI agents.

The plan was called Project OT.

Some teams faced 60% headcount reductions.

Then the internal data landed.

Code changes surged 220% year over year.

User-facing features rose 36%.

That is not productivity. That is AI generating code that nobody can ship.

But the damage went deeper.

Technical incidents climbed 40%.

Time engineers spent fixing AI-created problems rose 70%.

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

Zuckerberg cancelled the November layoff wave the night before it was supposed to start.

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

Meta is the world's most valuable company.

It committed $130 billion to AI infrastructure in 2026.

And its own internal data proved the agents cannot do the job.

Audit your AI workforce plan today.

If your ROI model assumes AI output equals shipped product, run the numbers again.

220% code changes. 36% features shipped. 40% more incidents.

That is the real productivity equation.