Zuckerberg Tells Meta Employees AI Agent Progress Is Slower Than Expected

Meta Chief Executive Mark Zuckerberg told employees that progress on the company’s artificial-intelligence “agents” has been slower than expected, and said the slower pace does not mean he is giving up on the technology, multiple reports said.

The comments, described as part of an internal town hall, characterize Meta’s agent development as lagging behind expectations over a recent period of effort. The reports portray Zuckerberg as disappointed that the work is not moving faster, while still emphasizing that Meta intends to continue investing in agent capabilities.

Yahoo Finance’s report on the episode framed Zuckerberg’s message as a recognition of shortfalls in AI agent momentum, even as Meta continues to pursue the broader idea of systems that can carry out tasks on a user’s behalf rather than simply answer questions. In that view, “AI agents” are intended to act across steps, using tools and context to complete objectives, which is harder than producing text that matches a prompt.

Reuters, via a republished market news page, also said Zuckerberg told employees that AI agent progress was slower than expected. The Reuters-based reporting similarly centers on the discrepancy between results and the speed Meta wanted, without presenting new product metrics or external benchmarks.

Meta, which has been restructuring its engineering and product roadmaps around AI, is now facing an environment where expectations for agent-like systems are rising across the industry. The difficulty is not only building models that can reason, but also making them reliable in practice, integrated with tools, and safe enough for repeated real-world use.

In broader terms, Zuckerberg’s admission highlights a tension common in frontier AI product development. Timelines can slip when teams attempt to move from research demonstrations to durable systems that perform consistently across tasks, interfaces, and user goals. In Meta’s case, that includes connecting agent behavior to the company’s platforms and workflows, which typically requires substantial engineering beyond model training alone.

What Meta did not disclose in the reported remarks, at least in the coverage available publicly, is how far behind targets the company is, whether it has adjusted delivery schedules, or what specific capabilities are missing. The reports do not cite new releases, performance numbers, or revised dates tied to agent deployments.

For investors and observers, the next announcement to watch is whether Meta’s public AI updates and product releases begin to reflect faster improvements in agent behavior, such as better task completion or clearer user-facing integrations, or whether leadership continues to set expectations for a longer runway.