Zuckerberg tells Meta staff AI agents are progressing slower than expected

On July 2, 2026, at an internal meeting with Meta staff, Mark Zuckerberg acknowledged that the company's AI agent development is moving slower than he had expected. It is a rare public signal that even the largest AI companies have to revise their expectations for agentic systems.

What was said at the closed meeting

According to sources familiar with the contents of the meeting, the Meta chief told employees that the company's AI agent efforts are not moving at the speed he anticipated. He was speaking specifically about agentic systems: programs that can carry out multistep tasks on their own, operate tools and make decisions without constant human involvement.

The specific technical reasons for the lag have not been disclosed publicly. The very fact that such an admission got outside the corporate walls is notable: Meta traditionally keeps its internal strategic agenda closed. In recent years the company has taken an increasingly aggressive position in AI, from the open release of the Llama series to the large-scale integration of Meta AI into WhatsApp, Instagram and Facebook. Against that background, publicly acknowledging a slowdown in development is an atypical move.

Why AI agents turned out harder than the industry expected

In 2024–2025, major tech companies, including Meta, OpenAI, Google and Anthropic, were nearly unanimous in calling 2025–2026 the "era of AI agents." The logic seemed convincing: language models had learned to reason and plan, and the next step was to teach them to act reliably. It looked like a matter of a few months of optimization on top of models that already worked.

In practice, the move from an "understanding" model to an "acting" agent proved considerably harder. The main problems:

• Errors accumulate in long chains of actions, and a failure at an early step breaks all the logic that follows
• Real user scenarios turned out to be incomparably more varied than lab tests and benchmarks
• Integration with external systems, such as banks, payment services and corporate tools, produces failures that are hard to predict
• Users are not ready to delegate important tasks to agents without reliable control over every step

Meta has invested heavily in this area: it is developing the open Llama model series, building Meta AI into messengers and social networks with an audience of more than three billion users, and creating agentic tools for corporate customers. That Zuckerberg is voicing disappointment now suggests the gap between ambition and reality is wider than originally assumed.

What this means

Zuckerberg's words are neither an admission of defeat nor a reversal of strategy. Meta remains one of the key players in open AI: it has its own computing capacity, world-class teams and an unprecedented user base for testing agentic systems in real-world conditions.

For companies and developers planning to build products on agentic technologies, it is a concrete reminder: the direction is right, the pace is real, and it is slower than the headlines of past years promised. Counting today on fully autonomous "digital employees" that work reliably without supervision is still premature. Most likely, Zuckerberg is using this admission as a signal to reset internal expectations and to step up work on the next round of agentic research.

*Meta has been designated an extremist organization and is banned in Russia.