Meta chief executive Mark Zuckerberg acknowledged at an internal town hall on July 2 that the company’s sweeping AI restructuring has fallen short of expectations, according to a recording heard by Reuters.
He said the development of AI agents had not accelerated over at least the past four months in the way leadership anticipated, and that the company’s bets on its new structure “haven’t come to fruition yet.” Zuckerberg also conceded that the reorganization, which included major job cuts, was “not as clean” as it could have been and that executives had misjudged the timing of the changes. It is a rare public admission of miscalculation from one of the industry’s most aggressive AI investors.
The restructuring in question was severe. In May, Meta laid off about 10% of its global workforce and reassigned roughly 7,000 employees to AI-focused teams, moves that drew internal pushback and hurt morale. Zuckerberg said the plans were shaped in January and February, when conversations with top staff centered on fears that Meta was not moving fast enough to adapt, and that leadership was “super optimistic” at the time about coding tools like Anthropic’s Claude Code.
The clear implication is that expectations for how quickly AI agents could take over real work outran what the technology has so far delivered. Since then, Zuckerberg and other executives have tried to soften parts of the reorganization without reversing course, and he told employees in May, to some skepticism, that he did not expect further companywide layoffs this year.
Notably, the admission partly decouples two narratives the industry often merges. Meta’s headcount cuts were driven by the cost pressures of funding AI infrastructure, not by agents already replacing workers, a distinction that undercuts the common framing of AI-era layoffs as proof that the software is doing human jobs.
Zuckerberg struck an optimistic note about the future, saying he expects Meta to see more significant benefits from its AI investments within three to six months. The scale of the wager is enormous: Meta is projected to spend as much as $145 billion on AI infrastructure this year, a large slice of the more than $700 billion Big Tech is expected to pour into the technology.
Why It Matters
The comments are a candid data point in the intensifying debate over whether AI spending is running ahead of returns. Coming from a frontier operator with every incentive to project confidence, Zuckerberg’s acknowledgment of a gap between “agents exist” and “agents are doing the work we planned” carries weight.
It reframes the challenge as one of organizational and integration friction rather than raw model capability, since Meta’s Llama benchmarks are not the issue. That friction is precisely the problem that Amazon’s $1 billion forward-deployed-engineering push and Microsoft’s $2.5 billion Frontier unit are being paid to solve, an irony given that both are Meta’s infrastructure partners. For enterprises copying the same playbook, the warning is that committing to heavy AI spending before agents deliver measurable output can open a costly gap in the books.
The Privacy Sub-Plot
The town hall also touched a separate controversy. Chief technology officer Andrew Bosworth said a review of a data security incident involving Meta’s employee-monitoring software found no employee data was used in AI training. The company paused the program last month while investigating the exposure of sensitive data.
The software, installed on US employees’ computers in April, tracks mouse movements and digital activity to gather data for AI training, and Bosworth had originally told staff there was no way to opt out. He now says that if the program resumes after the review, it will be opt-in, framing it as a voluntary “human survey” for those comfortable participating. The reversal underscores the friction, internal and external, that Meta’s rapid AI push keeps generating.
