Earlier this year Meta created a plan to reduce some of its teams by as much as 60 percent in order to make the company "AI native." According to Reuters, citing two people familiar with Meta's internal affairs, the plan was codenamed Project OT.

Meta confirmed the plan's existence, saying that as part of its restructuring it asked some teams to conduct a scenario planning exercise looking at the potential impact of redeployments, open role closures and cuts. The company says this resulted in moving thousands of employees to priority work on newly established teams, and that "we didn't move forward with every scenario from the exercise — and it was never assumed we would."

The scope of the plan

According to documents Reuters reviewed, executives created Project OT in January, exploring the use of AI to perform much of the daily work done by thousands of human employees. Small teams of people would oversee the AI.

  • Headcount effect — one HR executive said the scenarios implied a reduction of about 25 percent or more.
  • Two waves — the plan called for two rounds of layoffs. The first occurred in May; the second was canceled.
  • Use of the savings — some of the money saved would pay high-performing employees, especially those with AI engineering skills.

One document defined "AI native" as a company where AI-ready tools and agents interact, workflows are automated and new builds are AI-first. The definition also covered selling AI agents to third parties, which Meta began doing in June.

The retreat

According to Reuters, immediately after the May layoffs Mark Zuckerberg canceled the November wave. The agency was unable to determine exactly what prompted the shift. But two factors stand out in the reporting.

The first is morale. March and April reports of impending layoffs, along with Meta tracking employees' keyboard and mouse input in order to train AI agents, hurt employee morale. The company has since paused that program.

The second, and more interesting, is uncertainty about whether increased AI use actually boosts productivity. Internal posts contain a striking contradiction: according to a post by CTO Andrew Bosworth in early June, code changes to the internal software platforms and infrastructure employees use were up 220 percent year over year. But changes that led to new or upgraded features reaching Meta users were up only 36 percent.

How the agents behaved

Internal posts point to another problem: AI agents making "large-scale, disruptive actions that humans are unlikely to execute."

That phrase captures the risk autonomous agents carry in a corporate setting. A human employee pauses and asks before making a large change to a system; an agent does not pause. It acts to the extent it is authorized, and at the speed it is authorized.

Why it matters

This case makes concrete the difficulty organizations face in working out where to place AI: deciding whether the technology is a better fit than an employee for a given task is harder than it looks.

Bosworth's numbers show it. Code changes rising 220 percent does not mean output rose; it means there was more activity. If what reached users stayed at 36 percent, the gap is not productivity but noise. Measuring how much a tool gets used is easy; measuring how much it helps is hard — and when the two get confused, headcount decisions get made on the wrong number.