Mark Zuckerberg said AI agent progress has been slower than expected in an internal town hall

What Happened

During an internal town hall, Mark Zuckerberg reportedly told Meta employees that progress on AI agents has been slower than the company anticipated, and that achieving "superintelligence" — Meta's stated long-term AI ambition — will take longer than some of its most bullish public messaging has suggested. The remarks came alongside news that Meta is shifting an internal AI training program to opt-in status after a data leak, a move that suggests the company is also recalibrating how it handles employee data used to train its models.

Why It Matters

Zuckerberg's candor is notable given how aggressively Meta, along with rivals like OpenAI, Google, and Anthropic, has marketed autonomous agents as the next transformative wave of AI — systems capable of independently completing multi-step tasks, from coding to scheduling to enterprise workflows. Public admissions from a major AI lab leader that agent progress is lagging expectations offer a useful reality check for an industry saturated with confident roadmaps and product launches.

For enterprise buyers evaluating autonomous AI agents, this matters directly. Many organizations have been piloting agentic tools for customer service, IT operations, and internal automation based on vendor promises of near-term reliability. If even Meta, with vast compute and research resources, is finding agent development harder than expected, it reinforces a growing pattern: today's agents still struggle with the reasoning consistency, long-horizon planning, and error recovery needed for high-stakes autonomous deployment. Enterprises should treat aggressive agent-adoption timelines with caution and prioritize narrow, well-scoped use cases over ambitious full autonomy.

The Data Leak Angle

The decision to make Meta's internal AI training program opt-in following a data leak also raises broader questions about how AI companies source training data from their own workforces — and what happens when that process breaks down. Making participation voluntary is a meaningful governance shift, and it may foreshadow similar scrutiny of data practices across the industry as companies race to gather high-quality training signals for agentic systems, which often require rich behavioral and interaction data to learn from.

Broader Context

Zuckerberg has previously framed superintelligence as a defining goal for Meta, backed by enormous infrastructure investment and high-profile research hires. Tempering expectations internally, even while maintaining an ambitious external narrative, suggests Meta is managing a gap between its long-term vision and near-term technical reality. For the wider AI agents market, this is a signal worth watching: hype cycles around autonomous agents may be outpacing actual capability, and enterprises building strategy around near-term agent maturity should factor in longer timelines and more rigorous evaluation before committing to broad deployment.