Meta's AI agents wrote 220% more code. Only 36% more features shipped. Then outages spiked 40%.

Meta's Project OT was supposed to be the blueprint.

Cut up to 60% of some teams. Replace them with AI agents. Go "AI native."

It imploded.

Internal data showed AI agents produced 220% more code changes year-over-year.

But features reaching users? Only up 36%.

The gap isn't a rounding error. It's the entire thesis collapsing.

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

Major technical and security incidents spiked 40%.

Time spent firefighting those incidents jumped 70%.

Meta CTO Andrew Bosworth admitted internally the company did an "atrocious job explaining the vision."

Zuckerberg told staff he got "the timing wrong" on agentic development.

The $1.18 billion in severance from May's 8,000-person layoff is already booked.

The second wave of cuts? Cancelled hours before it was supposed to start.

Meta still plans to spend $130 billion on AI infrastructure this year.

That's not a bet on AI working. That's a bet on AI eventually working after it already failed.

If your enterprise AI strategy resembles Project OT — automate first, measure later — audit your implementation metrics today.

Volume of AI output is not value. Code changes are not features shipped.

The companies that survive AI transformation will be the ones that measure outcomes, not activity.