Meta has halted a sweeping corporate overhaul that would have replaced thousands of jobs with AI agents. That is according to the news agency Reuters, citing internal documents and conversations with employees. Just hours before the first wave of layoffs in May, CEO Mark Zuckerberg pulled the second, larger round. The case is a rare, well-documented rejection of the idea that autonomous software can take over entire departments in the near term.
Key takeaways
- Under the name Project OT, Meta planned, according to Reuters, to have AI agents handle much of its routine work and keep only small teams for oversight.
- Two waves of layoffs were planned, in May and November 2026. Individual teams were to shrink by up to 60 percent, not the company as a whole.
- Zuckerberg canceled the November wave shortly before the first round. Meta instead cut about 10 percent of positions and moved some staff into AI projects.
- Internal metrics showed the AI agents did not deliver the expected productivity gains: far more generated code, but barely more finished features for users.
- At the same time, employee morale collapsed, partly because of monitoring software that recorded keystrokes and mouse movements.
The plan behind Project OT
The idea took shape, according to Reuters, at a leadership retreat in January 2026 at Zuckerberg’s compound in Hawaii. The model was young companies built around AI tools from the start. Meta was to become an AI-native company in which software handles most of the daily work that thousands of people do today. What would remain were small, highly paid teams steering a fleet of AI agents. The company employs roughly 75,000 people.
Meta confirms that Project OT exists and describes it as a year-long effort to cut costs, reorganize teams, and shift staff into priority areas. It acknowledges that in some scenarios individual departments were to become up to 60 percent smaller. At the same time, it stresses that the plan was never to cut 60 percent of the entire workforce.
Why the overhaul failed
Two developments tipped the balance. First, the autonomous agents did not deliver what leadership expected. Internal figures showed that the volume of code changes to Meta’s platforms and infrastructure rose 220 percent year over year, while the number of new or improved features for users grew only 36 percent. Infrastructure teams reported unreliable AI code, and uncontrolled agents triggered larger outages. Serious technical and security incidents rose 40 percent, and the time to fix them rose 70 percent. That AI agents judge their own performance and time needs poorly was shown shortly afterward by an independent study, which we covered separately.
Second, morale in the workforce turned. Many employees read the overhaul as an announcement of their own replacement. That was amplified by software that logged keystrokes and mouse movements, officially also to gather training data for the AI. The internal satisfaction score fell from 74 to 55 percent, and employees pushed forward efforts to form unions. That observation matches earlier studies finding that AI at work has so far mainly shifted tasks rather than eliminating whole professions.
What Meta now says and does
Zuckerberg promised the workforce more stability and said he does not expect any further company-wide layoffs this year. At the same time, he conceded that the AI agents had not developed as quickly as leadership had assumed. The monitoring software was suspended, and some of the reassigned employees returned to their previous roles. Publicly, Meta is framing the reversal with the message that it is a human-centered company.
The planned second wave in November is canceled. The cut that already happened stays in place, but at around 10 percent of positions it is far smaller than originally intended. For the affected units, this is less a change of course than a shift in emphasis: AI tools are still to be expanded, just as a complement rather than a replacement.
Takeaway: what other companies can learn
Meta’s retreat is not proof that AI agents are useless inside a company. But it shows how wide the gap is between a demo and sustained operation. More code is not more product. When autonomous systems write errors into the infrastructure, the debugging eats up the time saved. And an overhaul that frames its own workforce as a cost item undermines exactly the cooperation it needs. For smaller firms weighing similar steps, the case is an expensive real-world experiment run at someone else’s expense.
Outlook
Meta will keep pushing AI aggressively into product development; the halted overhaul does not change that. The question is the pace. Zuckerberg’s admission that the agents are not there yet is notable for a company pouring billions into that very technology. Other large employers that publicly plan AI-driven job cuts will likely read the Reuters report closely. The next test is whether Meta keeps the reassigned employees for good or resumes the cuts in smaller steps once public attention fades.