Zuckerberg says Meta's AI agents are moving slower than expected

Why it matters

Meta is testing whether a founder-led AI reorganization can force agent products into existence. Zuckerberg's town hall suggests capital and headcount moves are ahead of product proof.

Mark Zuckerberg told Meta employees on Thursday that the company's AI agent work has moved more slowly than leadership expected, a rare internal admission from the founder and CEO as Meta tries to turn one of the largest AI infrastructure budgets in the market into products employees and users can actually use.

AI agent development over the last four months has not "accelerated in the way we expected," Zuckerberg said during an internal town hall, according to a recording heard by Reuters. He also said a company reorganization that included major job cuts was not as "clean" as it could have been and that Meta's bets on the new structure "haven't come to fruition yet," Reuters reported.

That is a sharper message than the one Zuckerberg has been selling to investors for much of 2026. In January, he told investors Meta users would start seeing new AI models and products "over the coming months," with agentic shopping tools singled out as a focus area, according to TechCrunch's account of the call. The thesis was that Meta's advantage would come from personal context: social graph, content history, interests and relationships. Six months later, Zuckerberg is telling employees that the agent work has not sped up as expected.

The gap matters because Zuckerberg has tied Meta's internal operating model, capital budget and product roadmap to AI agents. In April, Meta raised its 2026 capital expenditure forecast to $125 billion to $145 billion, including principal payments on finance leases, up from a prior range of $115 billion to $135 billion. Reuters reported Thursday that Meta is projected to spend as much as $145 billion on AI infrastructure this year, a significant share of Big Tech's more than $700 billion AI outlay.

Zuckerberg is still asking employees to wait for the payoff. He said at the town hall that Meta should start seeing more significant benefits from its AI investments within the next three to six months, Reuters reported. That timeline pushes the accountability window into late 2026, after a year in which Meta has already reorganized teams, shifted workers into AI workflows and raised spending guidance.

The town hall also adds a second data point to a pattern Reuters reported in June. In an internal memo seen by Reuters, Zuckerberg told employees Meta had made mistakes in its AI workforce shift and would "almost certainly make more." Reuters reported then that Meta had laid off 10% of its global workforce in May and transferred 7,000 employees to new initiatives related to AI workflows. Zuckerberg also said Meta would try to find new roles for employees who had been reassigned to train AI models.

For founders building agent startups, Meta's stumble is useful context. The largest consumer distribution platform in the West can buy compute, hire researchers and reorganize thousands of employees, but the hard part remains product reliability: agents that can take action across messy user contexts without producing expensive errors. Zuckerberg's mother-test framing from earlier this year captured the problem plainly: there are many agents being built, but far fewer he would trust for ordinary users.

Meta's current bet is broader than a chatbot upgrade. The company is trying to make agents work across commerce, advertising, internal engineering and consumer assistants while also building enough data-center capacity to support that shift. In January, Zuckerberg created Meta Compute, a top-level initiative meant to build tens of gigawatts of AI infrastructure this decade. Axios reported this week that Meta has also explored selling cloud computing services and access to AI models, a move that would let the company monetize excess capacity if its own product rollout takes longer than planned.

That optionality cuts both ways. If Meta can sell spare compute at attractive prices, Zuckerberg has a financial release valve for an infrastructure buildout that is running ahead of proven agent revenue. If the agents remain behind schedule, the company risks looking like a hyperscaler by accident: long on chips and data centers, short on products that justify the buildout inside Facebook, Instagram, WhatsApp and Meta's ad business.

The town hall also touched a separate internal AI training controversy. Meta Chief Technology Officer Andrew Bosworth said a review of a recent data security incident involving the company's mouse-tracking software indicated that no employee data was included in AI training, according to Reuters. The program, which Reuters said tracks employee mouse movements and digital activity for AI training, was paused last month while Meta investigated an exposure of sensitive data. Bosworth said that if Meta turns the program back on after the review, it will be opt-in. Reuters reported that when Meta first installed the software on U.S. employees' computers in April, Bosworth told them there was no way to opt out.

That detail explains why the agent slowdown is also an internal labor story. Zuckerberg has not simply funded an AI lab around the edges of Meta. He has pushed the company to reorganize work around AI, asked employees to train models and accepted disruption from layoffs and team transfers. Thursday's message was an acknowledgment that organizational force has not yet translated into the acceleration he expected.

Meta still has the main asset that lets Zuckerberg keep pressing: the core advertising business throws off enough cash to fund a large AI program longer than most rivals can tolerate. The open question is how long investors and employees will accept a three-to-six-month horizon if the agent products continue to arrive slower than the infrastructure bills.