Meta stops grading employees on AI usage, but Hatch tokens keep climbing

Meta told employees this week that performance reviews will no longer depend on how much they use AI tools, three employees who received the internal message tell WIRED. Workers testing Meta's new agentic tool Hatch say their token consumption keeps rising.

• Almost a year ago, Meta said it would grade workers on their "AI-driven impact," evaluating how much they used chatbots and agents in daily work, Business Insider reported at the time
• That usage correlated with labels like "AI Native," "AI First," and "AI Enabled," according to a lawsuit filed in July by about two dozen employees who argue Meta violated US antidiscrimination law when laying them off in May; Meta denies the allegations
• New guidance this week drops those criteria for looser language, noting that outcomes "can be supported by AI or other means"; The Information reported Wednesday that Meta told engineers company-wide it will not use AI adoption dashboards or token counts to evaluate impact
• The tokenmaxxing culture peaked when one employee built an internal leaderboard tracking AI usage and handing out titles like "Token Legend," before taking it down in April once the dashboard leaked

Meta spokesperson Tracy Clayton said the update emphasizes what the company says has been the standard all along: employees get evaluated on their contributions, and labels such as "AI Native" were never used for performance reviews. Employees describe the change to WIRED as subtle but welcome, since it removes pressure to use AI in situations where it doesn't help. A couple of months before this reversal, Meta had moved the opposite way, rationing employees' AI usage to control the compute bill.

The reviews are loosening while Meta pushes a new incentive to use AI: Hatch, the agentic tool the company let employees test on corporate devices for weeks ahead of a public launch. Hatch can browse the web and operate other applications on its own. Some employees avoid connecting it to personal email or calendar, citing privacy worries and the risk that it breaks something important. An earlier Meta project that tracked employee keystrokes to train AI models, since paused, left many of them distrustful of the company's AI tools to begin with. Workers say Hatch still burns through more tokens than standard chatbots or coding assistants, even without a leaderboard pushing them toward it.

The stakes go beyond compute bills. Meta walked back a broader plan last week: Zuckerberg canceled a second wave of the company's AI-native reorg after a rise in AI-linked software bugs and slower-than-expected agent progress, Reuters reported. He has said Meta doesn't expect further mass layoffs this year, but some employees still expect that a version of Hatch that works could bring the cuts back.

This article is for informational purposes only and does not constitute investment advice.