Meta CEO Mark Zuckerberg has acknowledged that the company’s ambitious push into AI agents is advancing more slowly than anticipated, offering one of the clearest indications yet that building autonomous AI systems remains significantly more challenging than many technology companies initially expected.

Speaking during an internal town hall, Zuckerberg said Meta’s large-scale organizational restructuring—including workforce reductions and the reassignment of thousands of employees to AI-focused roles—had not delivered the expected pace of progress. While executives believed earlier this year that AI agent capabilities would accelerate rapidly, those expectations have not yet been fully realized.

The comments come after Meta invested heavily in reshaping the company around artificial intelligence, laying off roughly 10% of its workforce while redirecting substantial engineering resources toward AI development. The strategy was designed to support massive infrastructure investments and prepare the company for a future in which AI agents can autonomously perform increasingly complex tasks for users.

Despite the slower-than-expected progress, Zuckerberg emphasized that Meta is not changing its long-term direction. Instead, he suggested the company expects more meaningful returns from its AI investments within the next three to six months as research, infrastructure, and product development continue to mature.

The remarks provide an important reality check for the broader AI industry. While large language models have advanced rapidly, transforming them into reliable autonomous agents capable of independently completing complex workflows remains a far more difficult engineering challenge. Many leading AI companies continue investing aggressively despite uncertainty around deployment timelines.

The town hall also addressed concerns surrounding Meta’s internal AI training practices. Following a review of a controversial employee activity tracking program, company leadership stated that employee data had not been incorporated into AI model training. Meta also indicated that any future use of the software would require employees to opt in, reversing its earlier mandatory approach.

Zuckerberg’s comments reinforce a broader industry trend: AI infrastructure spending continues at record levels, but translating that investment into dependable, production-ready AI agents is proving more gradual than early optimism suggested.