Meta: Zuckerberg Disappointed by the Pace of AI Agents

Meta: Zuckerberg Disappointed by the Pace of AI Agents

Mark Zuckerberg acknowledged weaknesses in the company's restructuring during an internal meeting. The AI agents around which Meta has reorganized are progressing more slowly than expected, Zuckerberg stated. Meanwhile, the head of AI presented a more optimistic vision.

The CEO of Meta admitted during an internal meeting on Thursday that the systems known as AI agents have not advanced as quickly as anticipated, according to an audio recording obtained by Reuters. The company's restructuring has not gone as "smoothly" as it could have, he said, and executives misjudged the timing.

"The trajectory of agent development over the past four months has not really accelerated as we expected," Reuters quoted Zuckerberg as saying. The bets on the new structure "have not yet paid off."

This admission carries significant weight. Zuckerberg has spent the past year fully investing in catching up in AI. He appointed Alexandr Wang to lead the AI division, renamed it Meta Superintelligence Labs, and offered nine-figure sums to attract top talent from competitors. In April, the company launched Muse Spark, the first model in a new lineup. It received good benchmark scores but did not match those of OpenAI or Anthropic.

A Restructuring Focused on Agents

Meta laid off about ten percent of its global workforce in May and transferred approximately 7,000 employees to AI teams. The goal was to fund an expensive AI infrastructure and optimize the efficiency of AI-powered workflows. During planning in January and February, executives feared they were not moving quickly enough, Zuckerberg stated. At that time, executives were "super optimistic" about tools like Claude Code from Anthropic.

Meta plans to spend up to $145 billion on AI infrastructure this year, a significant portion of the more than $700 billion that major tech companies are collectively investing. Zuckerberg expects more tangible results in the next three to six months. A Meta spokesperson declined to comment.

According to Bloomberg, Meta is also building a cloud business to sell excess AI computing capacity to external clients.

Meta's AI Head Takes a Different Tone

During the same meeting, the head of AI, Alexandr Wang, took a different tone, according to Business Insider. The upcoming Meta model, named "Watermelon," has caught up to OpenAI's flagship model, GPT-5.5, he stated, citing unspecified benchmarks. "Watermelon, our next model after Avocado, is currently in training," Wang said, according to Business Insider. "Watermelon uses an order of magnitude more computing power than Avocado," he added, referring to the internal codename for Muse Spark, which was launched in April.

On X, Wang entered damage control mode. Zuckerberg was talking about the progress of the entire industry, not specifically Meta's AI efforts, he said. An update to Muse Spark with major improvements in coding capabilities and agents is coming soon. A coding model comparable to Claude Opus from Anthropic is expected to follow "very soon," and users will appreciate what the team has been "working on."

Employee Tracking for AI Training Remains Unresolved

Separately from the discussion about models, CTO Andrew Bosworth addressed Meta's controversial mouse-tracking software during the same meeting. This tool records employees' mouse movements and digital activity to generate training data for AI. Meta had suspended the program after potentially sensitive data was exposed.

An internal review revealed that no employee data had been integrated into AI training, Bosworth stated. When Meta first installed the software on U.S. employees' machines in April, it informed them that there was no way to opt-out.

If the program restarts after the review is completed, it will operate on an opt-in basis. "For those who are comfortable, that's great, they can contribute to this kind of large human survey," Bosworth said. "For those who are not, that's not a problem."