Mark Zuckerberg Admits Meta AI Agent Development Is Slower Than Expected

During a series of internal town halls in early July, Mark Zuckerberg admitted that the development of AI agents has not accelerated as expected over the last four months. He conceded that executives miscalculated the timing of a major reorganization in May, which involved laying off approximately 8,000 employees and reassigning 7,000 others to AI teams. Zuckerberg noted the transition was not as "clean" as intended and that initial strategies were driven by over-optimism regarding external tools like Anthropic's Claude Code.

These organizational struggles coincide with a massive financial push, as Meta plans to spend up to $145 billion on AI infrastructure this year, funding the effort partly through $25 billion in debt. To mitigate risks and manage excess capacity, the company is reportedly establishing a cloud business to sell AI compute and negotiating a $6.5 billion deal with Samsung Electronics to produce 2nm AI chips. Despite these efforts, shares fell up to 7% following Zuckerberg's disclosures, as investors worry infrastructure spending is outpacing commercial demand.

Internal morale has further declined due to the rollout of the Applied AI division, which CTO Andrew Bosworth described as "atrocious." Employees also petitioned against the Model Capability Initiative, a keystroke-logging program used for AI training that was suspended on June 22 after a data leak. Bosworth stated that the program would be strictly opt-in if it ever resumed. Despite the turmoil, Meta reported strong first-quarter 2026 revenues of $56.31 billion.