Mark Zuckerberg told Meta employees at an internal town hall on Thursday that the company's work on AI agents had not "accelerated in the way we expected" over at least the last four months, according to a recording heard by Reuters. He said the company's bets on its new structure "haven't come to fruition yet," and told staff Meta had miscalculated the timing of the changes. In the same meeting, Meta's AI chief, Alexandr Wang, told employees the company's next model had "caught up" with OpenAI's flagship on benchmark tests.
Meta cut about 10% of its workforce, roughly 8,000 people, in May and reassigned about 7,000 employees to AI-focused teams, moves the company framed as a way to fund AI infrastructure and capture efficiency gains from AI-assisted work. Zuckerberg said executives had entered the year "super optimistic" about coding tools such as Anthropic's Claude Code, expecting them to speed development faster than they have. He told staff he expects Meta to see more substantial benefits from its AI spending within the next three to six months.
Key Takeaways
- Zuckerberg told a July 2 internal town hall that Meta's AI agent work hadn't 'accelerated in the way we expected' over the last four months.
- He said the restructuring bets 'haven't come to fruition yet' and expects a payoff within three to six months.
- AI chief Alexandr Wang claimed Meta's next model, Watermelon, has 'caught up' with OpenAI's GPT-5.5 on unspecified benchmarks.
- Meta plans $125-145 billion in 2026 AI capex; its shares closed down 4.9% at $582.90.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Zuckerberg traced the reorganization to conversations in January and February with what he called "our top people," who he said worried that "we weren't going to move fast enough to adapt." He acknowledged the changes were not as "clean" as they could have been and said he had made mistakes in the workforce shift and would "almost certainly make more," according to Reuters. Meta declined to comment.
Meta raised its 2026 capital expenditure forecast in April to between $125 billion and $145 billion, up from $115 billion to $135 billion, citing higher component prices and additional data-center costs. That spending puts Meta among the largest contributors to the more than $700 billion in AI infrastructure outlays Reuters attributes to big technology companies this year. Meta shares closed down 4.9% Thursday at $582.90. The company said in January it wants to build tens of gigawatts of compute under an initiative called Meta Compute, and Axios reported this week that Meta is weighing selling some of that capacity to outside customers.
Stay ahead of the curve
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
Wang said the company's next model, codenamed Watermelon, had "caught up" with OpenAI's GPT-5.5 on closely followed benchmarks, according to two people familiar with the meeting cited by Business Insider. Wang said Watermelon is still in training and uses "an order of magnitude more compute" than Avocado, Meta's internal name for Muse Spark, the model it released in April. Business Insider reported it was not clear which benchmarks Wang cited, and neither Meta nor OpenAI confirmed the claim. Muse Spark scored well on some benchmarks at launch but did not match OpenAI or Anthropic overall, Business Insider reported.
Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.
Chief Technology Officer Andrew Bosworth used the town hall to address the employee-monitoring program Meta paused last month. The software tracked workers' mouse movements and keystrokes to generate training data for agents that operate a computer. Meta halted it after a leak exposed internal conversations and performance information across the company, according to documents reviewed by Reuters. Bosworth said a review of the incident indicated no employee data had been included in AI training. If the program restarts, he said, it will be opt-in rather than mandatory. "For people who are comfortable, that's great, they can contribute to this kind of great human survey," Bosworth told staff. "To people who are not, it is not an issue." When Meta installed the software on U.S. employees' computers in April, Bosworth had told them there was no way to opt out.
Zuckerberg's three-to-six-month window would put a payoff toward the end of 2026, more than a year after Meta created its Superintelligence Labs unit under Wang and moved workers onto AI projects. Wang wrote on X on Thursday that an update to Muse Spark is coming, with gains in coding and agent tasks, and that a Meta model to rival Anthropic's Claude Opus would arrive "pretty soon." Meta has not published benchmark results for Watermelon or named a release date.
Frequently Asked Questions
What did Zuckerberg admit about Meta's AI agents?
At an internal town hall on July 2, Zuckerberg told employees that AI agent development had not 'accelerated in the way we expected' over at least the last four months, and that the company's restructuring bets 'haven't come to fruition yet.' He said he expects more substantial benefits from Meta's AI spending within three to six months.
What is Watermelon, and did it really catch GPT-5.5?
Watermelon is Meta's next model, still in training. AI chief Alexandr Wang told staff it had 'caught up' with OpenAI's GPT-5.5 on benchmarks, according to Business Insider. Wang did not specify which benchmarks, and neither Meta nor OpenAI confirmed the claim. Meta has not published results or set a release date.
How much is Meta spending on AI?
Meta raised its 2026 capital expenditure forecast in April to between $125 billion and $145 billion, up from $115 billion to $135 billion, citing higher component prices and data-center costs. That makes Meta one of the largest contributors to the more than $700 billion Reuters attributes to big technology firms' AI spending this year.
What happened with Meta's employee-monitoring program?
Meta paused a program that tracked workers' mouse movements and keystrokes to train agents after a leak exposed internal conversations and performance data. CTO Andrew Bosworth said a review indicated no employee data had been included in AI training, and that the program would be opt-in if it restarts. It was mandatory when installed in April.
How many jobs did Meta cut in the AI restructuring?
Meta cut about 10% of its workforce, roughly 8,000 people, in May and reassigned about 7,000 employees to AI-focused teams. Zuckerberg framed the moves as a way to fund AI infrastructure and capture efficiency gains, and acknowledged the reorganization was not as 'clean' as it could have been.
AI-generated summary, reviewed by an editor. More on our AI guidelines.



IMPLICATOR