Zuckerberg's Plan to Replace Meta Employees with AI Collapses

Mark Zuckerberg wanted to rebuild Meta into an "AI-native" company, where autonomous AI agents would take over most day-to-day tasks, overseen only by small, handpicked teams of humans. Six months after the launch of a secret program codenamed Project OT, the plan lay in ruins, and the company quietly walked back its most radical part.

According to reporting from outlets including The Globe and Mail and Engadget, Project OT called for more than an ordinary restructuring. It aimed at a structural change to Meta's working model: smaller teams, fewer layers of management, and AI agents handling tasks that previously belonged to engineers, researchers and product managers. By June, at least 11 units within the company had rolled out pilot, experimental teams operating under the new model.

Numbers that worried leadership

Internal data disclosed by Chief Technology Officer Andrew Bosworth showed a gap between AI activity and real-world results. Agents were generating dramatically more code than before, but little of that work translated into features that actually reached users. At the same time, serious outages and security incidents kept rising, and employees spent more and more time cleaning up after errors from the systems that were supposed to lighten their load.

Zuckerberg himself admitted in July, during an unplanned appearance at an internal company meeting, that he had misjudged how fast agentic technology would advance.

The trajectory of agentic technology development over at least the last four months hasn't really accelerated the way we expected it to - Mark Zuckerberg, CEO of Meta

Revolt on Workplace and employee surveillance

Alongside the technical problems, the plan sparked open conflict with staff. In April, as part of the Model Capability Initiative, Meta deployed software that tracked employee activity on US computers, logging mouse movements, keystrokes and screenshots across more than 200 applications, and even capturing emails and messages exchanged with European colleagues.

The reaction was fierce. Employees flooded the internal Workplace platform with critical posts, dark humor and elephant emojis referencing the wave of layoffs, calling the company a "factory for mining employee data." Privacy organizations, including NOYB, pointed to possible violations of the EU's GDPR, since data collected for internal communication purposes was reportedly being used to train AI systems without explicit consent. Meta ultimately suspended the monitoring program.

A last-minute retreat

The turning point came on the night of May 19 into May 20. Hours before announcing the first wave of layoffs, affecting about 10 percent of the workforce, Zuckerberg unexpectedly canceled plans for a second wave of cuts, scheduled for November and comparable in scale to the 2023 layoffs, which had cut 25 percent of staff. In an internal message to employees, he wrote that he did not plan further workforce reductions this year.

Meta officially maintains that Project OT was only scenario planning, not a final plan to eliminate jobs. But the actual sequence of events, from the layoffs, to the suspension of the surveillance program, to the cancellation of the second wave of cuts, suggests that artificial intelligence was not ready to take over as large a share of duties as had been assumed at the start of the year.

What it means for companies beyond Meta

The Project OT story lands at a moment when many tech companies are publicly attributing layoffs to advances in automation, while quiet rehires of laid-off workers show that reality is often messier than press statements suggest. Meta, a company with some of the largest computing resources and research budgets in the industry, shows that even at a market leader, AI agents have not yet replaced humans at the scale expected just six months ago.

Despite the failure of its workforce-reduction plan, Meta is not backing away from its AI investments. The company is standing by its plan to spend roughly $130 billion on AI infrastructure in 2026, and the broader initiative to restructure the organization around agentic technology has not been formally abandoned, despite the failure of its first attempt.