AI
Meta Ditches AI Usage Metrics for Performance Reviews While Rolling Out Hatch Agent
Meta has scrapped requirements that employees demonstrate AI adoption to earn good performance ratings, even as it pushes a new autonomous agent tool that's driving up token consumption.
WIRED — AI
Meta has formally abandoned its practice of grading workers on how much they use artificial intelligence tools, reversing a controversial policy that sparked internal backlash and contributed to at least one discrimination lawsuit. The company announced this week that performance evaluations would shift focus away from AI adoption metrics and back toward measurable work outcomes—regardless of whether AI was involved.
The tokenmaxxing era comes to an end#
For much of the past year, Meta had been measuring employee performance through an "AI-driven impact" framework that essentially tracked how frequently workers deployed chatbots and AI agents. This metric translated into status labels like "AI Native," "AI First," and "AI Enabled," creating what industry observers called a "tokenmaxxing" culture—a race to consume ever more computational resources to appear AI-forward.
The gamification took on absurd proportions when an employee created an internal leaderboard ranking colleagues by token consumption, awarding titles such as "Token Legend" to those who burned through the most AI capacity. When the dashboard leaked publicly in April, the employee took it down, but the damage to company culture was done. Workers reported padding their usage with frivolous AI prompts simply to climb the rankings.
The old policy had particular consequences for employees on health and family leaves, who couldn't accumulate usage metrics and subsequently faced unfair performance penalties. About two dozen workers sued Meta in July, alleging the company violated antidiscrimination laws by laying them off in May partly due to their low AI adoption scores. Meta has denied the allegations.
A quieter pivot#
The new performance review guidance eliminates direct references to "AI usage" and "AI Native" status. Instead, Meta now emphasizes that work outcomes "can be supported by AI or other means," giving employees genuine flexibility in how they approach tasks. A company spokesperson stated that the evaluations had always focused on contributions rather than adoption, though the prior year's messaging told a different story.
Engineers across the organization were explicitly told this week that Meta "will not use AI adoption dashboards or token counts to evaluate impact"—a clear repudiation of the prior year's approach. For many employees, the shift comes as relief, removing the artificial pressure to force AI into workflows where it adds little value.
