Meta’s effort to reorganize around AI agents is not delivering results as quickly as its leadership expected. At an internal town hall on Thursday, Mark Zuckerberg acknowledged that the company’s restructuring was not as smooth as it could have been and that the timing of its agentic AI push was misjudged.
The comments matter because Meta has spent the past year trying to close the gap in artificial intelligence. The company has shifted teams, increased its focus on infrastructure, and placed AI agents near the center of its strategy.
What Zuckerberg told employees
According to an audio recording obtained by Reuters, Zuckerberg said the development of AI agents has moved more slowly than planned. He described the corporate restructuring as not going as “clean” as it could have and said executives misread the timing.
Reuters quoted Zuckerberg as saying, “trajectory of the agentic development over at least the last four months hasn't really accelerated in the way that we expected.” He also said the bets on the new structure “haven't come to fruition yet.”
Those remarks point to a practical problem for Meta: reorganizing around AI agents is not the same as quickly producing working systems that transform internal workflows or products. The company expected faster acceleration, but Zuckerberg’s comments suggest the payoff is still developing.
A major AI reorganization is still settling
Meta’s AI push has involved major internal changes. Zuckerberg put Alexandr Wang in charge of the AI division, rebranded it as Meta Superintelligence Labs, and offered top talent nine-figure sums to lure them away from rivals.
In April, Meta released Muse Spark, the first model in a new lineup. The model posted solid benchmark scores, but it did not match OpenAI or Anthropic.
The restructuring also involved deep workforce changes. Meta laid off roughly ten percent of its global workforce in May and moved about 7,000 employees into AI teams. The goal was to fund expensive AI infrastructure and find efficiency gains through AI-powered workflows.
When planning began in January and February, Zuckerberg said senior leaders were concerned that Meta was not moving quickly enough. At the time, executives were “super optimistic” about tools like Anthropic’s Claude Code.
Meta also plans to spend up to $145 billion on AI infrastructure this year. That figure is part of the more than $700 billion Big Tech is collectively putting into AI. Zuckerberg said he expects more tangible results within the next three to six months, while a Meta spokesperson declined to comment.
Wang offered a more optimistic view
At the same town hall, Alexandr Wang described Meta’s AI progress in more positive terms, according to Business Insider. He said Meta’s upcoming model, code-named “Watermelon,” has caught up with OpenAI’s top model GPT-5.5, citing benchmarks that were not specified.
Business Insider quoted Wang as saying, “Watermelon, our next model after Avocado, is currently in training.” He added, “Watermelon uses an order of magnitude more compute than Avocado.” Avocado is the internal code name for Muse Spark, which shipped in April.
Wang also responded on X, saying Zuckerberg had been discussing the progress of the entire industry rather than Meta’s AI efforts specifically. He said a Muse Spark update with major improvements to coding and agentic capabilities is coming soon.
He also said a coding model on par with Anthropic’s Claude Opus would follow “pretty soon,” and said users would like what the team has been “cooking.”
The agent question is bigger than model benchmarks
The tension between Zuckerberg’s remarks and Wang’s more positive update shows why AI agents are difficult to assess. A model can improve on benchmarks while the broader agentic systems that companies want to build still take longer to become useful at scale.
Meta’s strategy depends on more than one upcoming model. It includes infrastructure spending, internal workflow changes, employee reassignments, and new tools for coding and agentic capabilities.
The source material also points to another part of the plan: according to Bloomberg, Meta is building a cloud business to sell excess AI compute capacity to outside customers. That would connect Meta’s infrastructure spending to a possible external business line, not only internal model development.
Employee tracking remains unresolved
The town hall also covered a separate issue involving employee data. CTO Andrew Bosworth addressed Meta’s mouse-tracking software, which records mouse movements and digital activity from employees to generate AI training data.
Meta had paused the program after potentially sensitive data was exposed. Bosworth said an internal review found that no employee data made it into AI training.
When Meta first installed the software on U.S. employees’ machines in April, Bosworth had told them there was no way to opt out. If the program restarts after the review is complete, he said it will run on an opt-in basis.
“For people who are comfortable, that's great, they can contribute to this kind of great human survey,” Bosworth said. “To people who are not, it is not an issue.”
For Meta, the immediate picture is mixed. The company is spending heavily, reorganizing aggressively, and preparing new models. But Zuckerberg’s own comments show that its AI agent push is still behind where leadership expected it to be.