Meta AI Agents Lag Behind Hype After Layoffs, Zuckerberg Admits
Internal optimism about tools like Anthropic’s Claude Code prompted deep cuts and reshuffled thousands of staff, but Meta concedes the payoff has not materialized yet.
At an internal town hall, Mark Zuckerberg acknowledged that the pace of AI agent development over the past four months has failed to meet Meta's expectations. The confession, first reported by Reuters, marks a rare retreat from the rapid-automation narrative that drove the company's aggressive restructuring earlier this year.
According to Zuckerberg, leadership was "super-optimistic" about developer tools like Anthropic's Claude Code when planning reorganizations last winter, fearing the company would fall behind the accelerating curve. In response, Meta laid off roughly 10% of its workforce and transferred more than 7,000 employees into dedicated AI units. Now, Zuckerberg admits the new structure has yet to bear fruit. Crucially, he clarified that the staff cuts were dictated by mounting capital expenditures and budget discipline, not real-world productivity gains delivered by autonomous agents.
Internal messaging split almost immediately. On the same day, Alexandr Wang, head of Meta Superintelligence Labs, claimed publicly on X that Zuckerberg was merely describing the industry as a whole. Internally, Wang reportedly assured employees that Meta's in-training model, codenamed Watermelon, has already matched OpenAI's flagship GPT-5.5 across internal benchmark suites.
Which benchmarks were used remains unspecified, and Meta declined to elaborate. Even if the claim holds, GPT-5.5 debuted in April, while OpenAI has already previewed GPT-5.6. The pressure on Wang is tangible: a year after Meta completed its $14.3 billion deal involving Scale AI and offered massive compensation packages to lure researchers, its only shipped result remains Muse Spark, an April release that never aimed for the frontier tier.
Meta is keeping its annual capex guidance unchanged at $125 billion to $145 billion, with Zuckerberg insisting visible returns should appear within three to six months. Whether those returns manifest as autonomous software engineering or merely higher infrastructure bills remains unresolved.
Source: habr
Comments
Help shape the next version: Add context or suggest a correction. AI review can add points toward a rewrite. Reviews and updates may take time; a full meter does not guarantee a new version.