Meta came closer than anyone outside the company realized to replacing a huge chunk of its workforce with AI. Internal documents obtained by Reuters reveal a plan, codenamed Project OT, that would have shrunk many teams by up to 60 percent, leaving small groups of human supervisors to manage armies of virtual AI workers. The first round of layoffs was scheduled for May 19. Hours before it was set to begin, Mark Zuckerberg pulled the plug on a second wave planned for November.
The reversal wasn’t driven by a sudden change of heart. Three things went wrong at once. The agent technology simply didn’t deliver the productivity gains Meta’s leadership expected. Investors were already grumbling about the company’s enormous AI spending. And employees figured out what was happening — when staff realized that tracking software logging their mouse clicks and keystrokes was likely training their own replacements, Meta’s internal network exploded with angry posts. Internal sentiment cratered from 74 to 55 percent, a drop that would alarm any HR department. By July, Zuckerberg admitted publicly that the agent tech “hadn’t sped up as fast as he had expected.”
What’s striking here isn’t just that the plan existed. It’s that Meta was willing to run a live experiment on its own workforce at this scale, betting that AI agents could absorb the work of more than half of some teams before the technology was actually ready. The company has been one of the loudest voices pushing AI-driven efficiency, but this suggests the internal timeline was far more aggressive than the public messaging let on.
This failed experiment will likely cool the enthusiasm of other large companies considering similar moves. If Meta — with its enormous AI research division and virtually unlimited budget — couldn’t get agent technology to the point where it could replace meaningful chunks of human labor, smaller players shouldn’t expect better results. The employee backlash also sends a signal: workers are increasingly aware that productivity monitoring tools may be doing double duty as training data for their replacements. That awareness alone could make the next attempt much harder to pull off quietly.
