Zuckerberg admits mistakes in Meta's AI-driven workforce overhaul

Mark Zuckerberg is trying to frame Meta’s AI overhaul as a difficult transition rather than a strategic retreat, even as the company works through the fallout from sweeping job changes. In an internal memo, he acknowledged that the pace and complexity of the restructuring had created problems and said Meta would keep pushing ahead on artificial intelligence while trying to steady the workforce.

“Given the complexity of these changes, we’ve made mistakes and will almost certainly make more,” Zuckerberg wrote, while also saying he wanted to provide as much stability as possible. That admission matters because Meta has spent months redesigning how it operates around AI, with management flattened, teams reorganized and employees pushed toward new internal roles tied to model training and product development.

The scale of the change is already visible. In May, Meta laid off 10% of employees globally, moved 7,000 workers into new AI-related initiatives and closed 6,000 open roles. The company said those steps were meant to improve AI workflows and create smaller, flatter teams. With 77,986 employees at the end of March, the restructuring affected about 20% of the workforce.

Meta has also said it does not expect more company-wide layoffs in 2026, a signal aimed at calming employees after months of uncertainty. The company has said it will try to find new internal roles for workers reassigned to train AI models, but the memo made clear that the broader reorganization is not ending. In April, Meta raised its 2026 capital spending forecast to between $125 billion and $145 billion, underscoring how aggressively it is still investing in AI infrastructure.

The human cost of that push has become a central issue inside the company. Meta plans to increase team-building budgets, hold a July hackathon and expand offsites and corporate events to improve cohesion. Zuckerberg also said the company heard complaints that manager oversight had become too broad and planned to scale that back. The new Applied AI Engineering unit reportedly has as many as 50 individual contributors for every manager, a ratio that highlights how sharply the organization has been compressed.

Tension has risen alongside the restructuring. More than 1,600 Meta employees signed a petition protesting software that monitors clicks and keystrokes for AI training data, and Chris Cox addressed the “brutal” environment on a recent employee call. Taken together, the memo reads less like a pause in Meta’s AI race than an attempt to absorb the labor backlash without slowing the company’s push toward AI agents, flatter teams and a more automated internal structure.