Zuckerberg says Meta’s AI agents are progressing slower than expected
Zuckerberg said Meta’s AI agents have lagged his expectations, even as the company is pouring up to $145 billion into AI infrastructure this year.
Mark Zuckerberg told Meta employees that the company’s AI agents had not advanced as quickly as he expected, a striking admission from a chief executive who has made AI central to Meta’s next phase. He said the last four months of agent development had not “accelerated in the way” he expected and that some of the company’s restructuring bets had “haven’t come to fruition yet.”
The comments landed as Meta is still digesting a major internal overhaul. The company cut about 8,000 jobs, roughly 10% of its corporate workforce, and reassigned about 7,000 employees into AI-focused roles, including teams tied to agent development and productivity measurement. Zuckerberg also acknowledged that the reorganization was not as clean as it could have been, underscoring how much execution risk remains inside the company’s AI push.
The gap matters because Meta has been positioning AI agents as the next interface layer for its products, advertising systems and consumer services. That ambition sits alongside a far larger spending commitment: some reports project Meta will spend as much as $145 billion on AI infrastructure in 2026. In an industry race where OpenAI, Google and Microsoft are all spending heavily on compute, talent and data centers, Zuckerberg’s remarks suggest the technology is still catching up to the investment and the sales pitch.
He said he expects Meta to start seeing more meaningful benefits from its AI investments within the next three to six months, a tighter timetable that will be watched closely by investors and rivals. For Meta, the challenge is not just building more capable systems, but making them reliable enough to plan, act and complete tasks without failing at the moments users notice most.
The comments also came against a backdrop of internal scrutiny. Meta paused its mouse-tracking program in June 2026 while it investigated data-security concerns, and Andrew Bosworth said a review found no employee data was included in AI training. That episode added another layer of pressure to a company already trying to prove that its AI investments can move from internal experiments to dependable consumer products.
This article was produced by Prism’s automated news system from verified source data, official records, and press releases, then run through automated quality and moderation checks before publishing. The system is built and supervised by the people who set the standards it runs under. Read our full AI policy.
Did this article answer your question?
