Meta Killed AI Token Dashboard

Meta built an internal scoreboard that ranked employees by how many artificial intelligence tokens they used, then pulled it down days later after the ranking escaped the building. The board handed out labels like “Token Legend,” “Session Immortal,” “Model Connoisseur,” and “Cache Wizard.” (aol.com) A token is the small unit an artificial intelligence model reads and writes, closer to a chunk of text than a full word. Counting tokens is like counting how much fuel a car burns, not how many passengers it delivered. (finance.yahoo.com) The dashboard was nicknamed “Claudeonomics,” and it tracked usage across more than 85,000 Meta employees over a rolling 30-day window. It also surfaced the top 250 users inside the company, which turned a back-end cost meter into a public office game. (theinformation.com, dnyuz.com) That game got big fast because Meta is already pushing hard on artificial intelligence inside the company. Recent internal training programs have taught employees to use artificial intelligence agents for coding, reports, and daily work, while some teams have set targets for how much code should be written with artificial intelligence help. (businessinsider.com, cio.economictimes.indiatimes.com) Once a company tells people that artificial intelligence use is important, a leaderboard tells them something narrower: use more of it than the person next to you. Reports around the dashboard said some employees were leaving artificial intelligence agents running just to drive their token totals higher. (blog.tmcnet.com, businessinsider.com) The numbers were huge enough to make the distortion obvious. One report said Meta employees used more than 60 trillion tokens in 30 days, and another said the top user alone had reached 281 billion tokens over that period. (theinformation.com, blog.tmcnet.com) Meta’s official reason for removing the board was that usage data had been shared externally. But the board’s short life also showed the problem with treating a raw activity count as a performance score inside a company with nearly 79,000 employees at the end of 2025. (timesofindia.indiatimes.com, sec.gov) This is the old office-metric problem in a new costume. If you rank salespeople by calls, they make more calls; if you rank engineers by lines of code, they write more lines; if you rank workers by tokens, they burn more tokens. (businessinsider.com, the-decoder.com) Meta did not shut down its broader push to get employees using artificial intelligence tools. It shut down one especially visible meter that confused “using the tool a lot” with “doing the job well.” (peoplematters.in, businessinsider.com)