Zuckerberg Admits Meta's AI Agent Push Is Moving Slower Than Planned

Meta CEO Mark Zuckerberg made a rare admission to his own staff this week, saying the company's push into AI agents has not moved as quickly as he had hoped. Speaking at an internal town hall, he acknowledged that the progress made over the last several months fell short of what executives had projected when they began planning big organisational changes earlier this year.

Back in May, Meta cut about 10% of its global workforce and shifted roughly 7,000 employees onto teams focused on artificial intelligence. The idea was to free up resources for heavy AI investment and prepare the company to benefit from AI tools that could make work more efficient. Zuckerberg had also told employees at the time that no more companywide layoffs were expected this year, though many staff remained doubtful.

Now, months later, Zuckerberg says the bet hasn't fully paid off yet. He explained that when the restructuring was being planned in January and February, top executives worried Meta wasn't moving fast enough to keep up with rapid changes in AI technology. They were especially optimistic about tools like Claude Code, a coding assistant from AI startup Anthropic, believing such technology would advance quickly. That optimism hasn't been matched by reality, he said, describing the company's new organisational bets as not yet having borne fruit.

Zuckerberg also admitted that the reorganisation itself, including the layoffs, wasn't executed as cleanly as it should have been, and that leadership had misjudged the timing of the changes. This comes as Meta and other big technology companies pour enormous sums into AI infrastructure, with Meta alone expected to spend up to $145 billion this year as part of an industry-wide spending spree that could top $700 billion.

Despite the setback, Zuckerberg struck an optimistic note about the near future, telling employees he expects Meta to start seeing meaningful returns from its AI investments within the next three to six months.

The same town hall also touched on a separate controversy: Meta's mouse-tracking software, which monitors employees' mouse movements and digital activity to help train AI systems. The programme was paused last month after a data security incident raised concerns about exposed employee information. Meta's Chief Technology Officer Andrew Bosworth told staff that a review found no employee data had actually been used in AI training. He added that if the programme resumes, employees will be given the choice to opt in, a reversal from earlier this year when workers were told they could not opt out at all.