Meta minimizes role of token maxing in employee evaluations
Executive Take
Counting AI usage as a performance metric backfires because employees optimize for the metric, not the work. Judge output and impact instead, and drop the leaderboard entirely.
Executive Summary
Meta told employees it will not use AI adoption dashboards or token counts in annual performance reviews. An internal memo from executives Maher Saba and Santosh Janardhan said managers should instead judge output quality, speed, problem complexity, and scope. Meta previously ran a token-usage leaderboard that led engineers to do unnecessary tasks to boost scores. Amazon deleted a similar leaderboard after seeing the same gaming behavior.
Why It Matters
HR leaders building AI adoption incentives should note that usage-based metrics invite gaming, not real productivity gains. This is a live case study on how to design (and undo) flawed performance criteria.
Bizquad Perspective
The real lesson isn't that token counting failed, it's that Meta and Amazon both had to publicly reverse course after rolling it out company-wide first.