Meta's AI Agent Progress Lags Internal Expectations, Zuckerberg Says

Meta acknowledges slower-than-expected AI development

Meta's push toward artificial general intelligence is taking longer than the company anticipated, CEO Mark Zuckerberg told employees during an internal town hall Thursday. While the company continues investing tens of billions in AI infrastructure and talent, progress on AI agent technology has not matched internal timelines, according to two sources who attended the meeting.

Zuckerberg maintained that Meta remains on a "journey to superintelligence" and expects to see benefits within three to six months, but emphasized the competitive AI landscape requires significant effort. The comments, first reported by Reuters, reflect growing tension between Meta's aggressive AI ambitions and the practical challenges of rapid development.

Meta declined to comment on the internal discussion.

Keystroke tracking program becomes optional

In a significant policy reversal, Chief Technology Officer Andrew Bosworth announced that Meta's AI training program — which captured employees' keystrokes and mouse movements to train models — will be opt-in only if it resumes. The mandatory version had sparked considerable employee backlash over privacy concerns.

Meta paused the program last month after an internal leak exposed employee conversations and keystrokes to colleagues. Bosworth acknowledged the rollout damaged morale and trust within the company, though he noted it generated more useful data than expected.

The retreat mirrors another recent climbdown: Meta gave engineers the option to leave its Applied AI task force last month after reassigning thousands to the unit, a move some employees called "the undraft."

Why it matters

Meta's internal struggles reveal the human cost of the AI arms race. Even companies with massive resources face fundamental tradeoffs between development speed and workforce trust. The keystroke-tracking reversal and task force opt-outs suggest that top-down mandates — however technically valuable — can backfire when they erode the morale of the very engineers building the technology. With Bosworth warning that morale is "probably one of the worst it's ever been" in Meta's 20-year history, and the company having laid off roughly 8,000 employees in May, Meta's AI strategy is colliding with organizational realities that money alone cannot solve.

Morale challenges mount

The policy changes come as Meta grapples with workforce sentiment. Weeks before the town hall, Bosworth warned staff that morale had reached historic lows following the May layoffs that eliminated 10% of Meta's workforce.

The combination of aggressive AI timelines, controversial data collection practices, and significant headcount reductions has created a challenging environment for the employees expected to deliver on Zuckerberg's superintelligence vision.

Details of the town hall were first reported by Reuters and Business Insider.