Meta softens AI metrics in employee reviews, pivots from token obsession

Meta is recalibrating how it measures engineering productivity—shifting from raw "token" output (a proxy for AI model usage that invited gaming) toward broader "AI-driven impact" language. The company burned through a cycle of metric-driven culture that rewarded volume over outcomes and is now correcting course before the metric itself becomes cargo cult theater. The shift surfaces a real tension in AI-first organizations: how to incentivize meaningful AI adoption without creating perverse incentives that inflate usage metrics rather than actual business value.