- Mark Zuckerberg told an internal town hall that Meta‘s AI agents have progressed slower than planned.
- He said the corporate restructuring around AI did not go as “clean” as it could have, and that executives misjudged the timing.
- Meta’s April model, Muse Spark, posted solid benchmark scores but did not match OpenAI or Anthropic.
- Meta’s AI chief presented a more optimistic view at the same meeting, per Reuters.
What Happened
Meta chief executive Mark Zuckerberg acknowledged at an internal town hall on Thursday that the systems known as AI agents have not advanced as fast as expected, according to an audio recording obtained by Reuters and reported by The Decoder on July 3, 2026. He said the company’s restructuring did not go as “clean” as it could have and that executives misjudged the timing.
The remarks are a rare public-facing concession from a CEO who has spent the past year presenting AI as Meta’s central priority.
Why It Matters
The admission carries weight because Zuckerberg reorganized Meta around AI over the past year and spent heavily to do it. He put Alexandr Wang in charge of the AI division, rebranded it as Meta Superintelligence Labs, and offered top researchers nine-figure packages to move from rivals. A concession that the payoff has not arrived reframes those bets as still-unproven rather than early wins, and it lands while competitors continue to ship.
It also exposes a gap between Zuckerberg’s assessment and that of his own AI leadership, which complicates how outsiders should read Meta’s progress reports.
Technical Details
Reuters quoted Zuckerberg saying the “trajectory of the agentic development over at least the last four months hasn’t really accelerated in the way that we expected,” and that the bets on the new structure “haven’t come to fruition yet.” He described the restructuring as not going as “clean” as it could have, attributing part of the problem to misjudged timing. In April, Meta released Muse Spark, the first model in a new lineup; it posted solid benchmark scores but did not match the leading systems from OpenAI or Anthropic. At the same town hall, Meta’s AI chief painted a more optimistic picture, according to Reuters, indicating a split within leadership over how to characterize the last four months of work.
Who’s Affected
The candor affects Meta’s AI staff, including the high-priced hires brought into Meta Superintelligence Labs on the promise of a fast-moving effort, and investors tracking the return on a year of aggressive spending. Alexandr Wang, who leads the division, is directly named in the restructuring Zuckerberg described as imperfect, which puts his mandate under scrutiny. Rivals OpenAI and Anthropic remain the benchmark Meta is measured against, and Muse Spark’s inability to match them is the concrete shortfall behind the abstract concern.
What’s Next
Zuckerberg framed the setbacks as timing and structure problems rather than a change of direction, so the near-term signal is whether Meta’s next model release closes the gap with OpenAI and Anthropic that Muse Spark did not. The tension to watch is between his caution and his AI chief’s optimism: if the next model underperforms, the gap between those two messages becomes the story. For now, Meta’s superintelligence bet remains funded and staffed, but on Zuckerberg’s own account it has yet to deliver.
