Meta's attempt to reorganize work around AI agents moved further than had previously been known, according to Reuters, before the company pulled back. Internal documents described a plan called "Project OT" that would have reduced many teams sharply and shifted more responsibility to smaller groups of people overseeing virtual AI workers.
The plan did not survive its own first stage. On the evening of May 19, only hours before the first round of layoffs, CEO Mark Zuckerberg stopped a second wave that had been planned for November.
What Project OT Was Designed To Do
At the center of the plan was a major assumption: AI agents could take on enough work to let Meta operate with far smaller teams. Internal documents showed that many teams were expected to shrink by up to 60 percent.
The remaining employees would not simply continue the old structure with fewer people. The documents described compact, "talent-dense" groups of human staff who would supervise virtual AI workers. In that model, the human role would become more concentrated, with people directing and checking AI-driven work rather than performing every task themselves.
That design made the performance of the agents critical. If the agents did not produce the expected productivity gains, the staffing model would become difficult to justify. According to Reuters, that is what happened: the agent technology did not deliver the productivity improvement Meta had expected.
Why The Plan Was Halted
Several pressures converged before the second layoff wave was stopped. The AI agent technology fell short of expectations, investors criticized Meta's massive AI budget, and employees pushed back inside the company.
The timing was striking. Zuckerberg halted the November layoff wave on the evening of May 19, only hours before the first round of layoffs was due to begin. That reversal suggests the company was still reassessing the plan at the point when the restructuring was already close to becoming real for employees.
Reuters reported that Zuckerberg later acknowledged in July that the agent technology had not accelerated as quickly as he had expected. That admission matters because Project OT depended on AI agents becoming useful fast enough to support a smaller workforce structure.
Employee Backlash Became A Major Obstacle
The internal reaction was not limited to quiet concern. Employees openly resisted the plan after they came to believe that tracking software logging mouse clicks and keystrokes was being used to train AI systems that could replace them.
That belief triggered a wave of angry posts on Meta's internal network, Reuters reported. For employees, the concern was not only that AI might change their jobs. It was that their own activity could be used to build the tools that would make those roles unnecessary.
Internal sentiment fell from 74 to 55 percent. That drop became another sign that the plan was damaging confidence inside the company at the same time the agent technology was failing to meet expectations.
The Productivity Gap At The Core Of The Reversal
The logic behind Project OT was straightforward: if AI agents could take on enough work, Meta could operate with fewer people in many areas. But that logic only works if the agents produce reliable productivity gains.
According to the source, that expectation did not hold. The agents did not deliver the gains Meta expected, and Zuckerberg admitted in July that the technology had not sped up as fast as anticipated. Without that acceleration, the planned reduction of many teams by up to 60 percent became harder to defend.
This is the central lesson from the episode as described by Reuters: replacing work at scale is not just a staffing decision. It depends on whether the underlying technology can actually carry the workload. If the AI systems are not ready, the remaining human teams may face a larger burden without the promised support.
What The Episode Shows About AI Restructuring
Meta's reversal shows how quickly an AI workforce plan can run into practical limits. The company had a codename, internal documents, a staffing concept and a scheduled second layoff wave. But the plan still depended on technology performance, investor confidence and employee acceptance.
All three became problems. Investors criticized the scale of AI spending. Employees reacted strongly to fears that workplace tracking data could be used to train replacements. The AI agents, meanwhile, did not deliver the expected productivity gains.
The result was a pause at a critical moment. The second wave planned for November was halted before the first round began. Project OT therefore became less a clean example of AI replacing work and more a case study in the constraints around doing so: the tools must work, the workforce must trust the process, and leadership must be able to justify the costs.