60 Percent Team Cuts Were The Plan. Meta Stopped.

The Plan Written At A Hawaii Retreat

In January 2026, Mark Zuckerberg gathered his top executives at his Hawaii compound to design what internal documents called Project OT, short for Organization Transformation. The goal was an "AI native" Meta: AI agents handling much of the daily work done by thousands of employees, with small, talent-dense human teams left to oversee them. Scenario planning for the restructuring explored cutting some teams by up to 60 percent, an HR executive projected, a reduction that could match or exceed Meta's 2022-2023 layoffs of roughly 25 percent of its workforce.

Meta proceeded with a first wave in May: about 10 percent of its global workforce was let go and roughly 7,000 employees were reassigned to AI-focused teams. A second, larger wave was planned to follow. It never happened.

The Numbers That Undid It

Zuckerberg called off the second wave on May 19, 2026, just before it was due to begin, and the reasons are visible in Meta's own internal numbers. AI-generated code output rose sharply, but the work that actually reached users did not follow at the same rate, and the systems built to move fast introduced new risk.

One of those incidents was public: hackers exploited Meta's own AI-powered customer support bot to gain access to high-profile Instagram accounts. Employees, meanwhile, feared being replaced outright and flooded internal systems with complaints, and investors began asking harder questions about the return on Meta's AI spending, projected at up to 145 billion dollars for 2026.

Why It Matters Beyond Meta

Meta had every advantage an AI-native restructuring could ask for: its own frontier models, unlimited internal compute, and a CEO personally sponsoring the plan from a January retreat. It still could not close the gap between AI activity and AI output fast enough to justify a second wave of job cuts, and the attempt introduced security incidents it did not have before.

Zuckerberg told employees directly that "the trajectory of the agentic development over at least the last four months hasn't really accelerated in the way that we expected," and that the reorganization's goals "haven't come to fruition yet." That is a rare on-record admission from a company with every incentive to claim otherwise.

The Decision Lesson For Your Own AI Plan

The failure mode here was not trying AI-native restructuring, it was measuring the wrong thing while doing it. Code generated is an activity metric; features shipped and incidents avoided are outcome metrics, and Meta's own numbers show those two categories moving in opposite directions for months before anyone acted on it.

Before committing headcount to an AI-agent plan, set the outcome metrics first, code shipped to production, support tickets resolved, incidents introduced, and require them to move together before scaling the cuts that assume they already have.

Read next: The Man Spending 145 Billion Says It Is Not Working Yet  |  Florida Rejects Meta's Deal, Bets On A Jury Instead