Zuckerberg says Meta’s AI agents have advanced slower than expected

Mark Zuckerberg told Meta employees on July 2 that the company’s AI agents have not advanced as quickly as he expected, offering a blunt internal snapshot of a business trying to sell speed while still testing its own limits. The admission landed after Meta’s sweeping May restructuring, which cut about 10% of the global workforce, moved roughly 7,000 employees into AI-focused roles and closed about 6,000 open jobs.

The changes were meant to push workers into new AI initiatives and strip out layers of management, but Zuckerberg told employees the overhaul had not been as clean as it could have been. He also said the company had miscalculated the timing of the move. In January and February, he said, executives were already worried the company might not move fast enough, and they had watched rivals and tools such as Anthropic’s Claude Code closely as they weighed how urgently to act.

That urgency has been matched by spending. Meta has projected capital expenditures of as much as $145 billion this year for AI infrastructure, part of more than $700 billion in industrywide AI outlays across Big Tech. Zuckerberg said he expects more meaningful benefits from those investments within three to six months, a timeline that keeps the company’s investors and employees waiting for proof that the strategy can translate into products and business gains.

Meta’s reorganization has also become a workplace issue, not just a technical one. The company’s AI-focused teams include Applied AI Engineering, Agent Transformation Accelerator XFN and Central Analytics, groups built to develop AI agents that can autonomously carry out tasks now done by employees. But the shake-up has already triggered pushback inside the company, along with concerns about morale as the workforce absorbs layoffs, role changes and a faster push into automation.

The internal pressure has been compounded by a separate privacy problem. Meta paused its Model Capability Initiative after an internal data leak exposed employee conversations, performance information and meeting transcriptions, and the company classified the incident as a SEV 2. That episode added another layer of distrust around a program already tied to AI training and workplace surveillance.

Zuckerberg had framed the layoffs in May by saying “success isn’t a given.” Two months later, Meta is still betting that AI agents will eventually justify the reorganization, but the company’s own chief executive has now acknowledged that the payoff is arriving more slowly than expected.