Amazon told developers to use AI every week and ranked them on a leaderboard. They started using it on tasks they didn't need, asking the AI questions they already knew the answer to. The Financial Times broke the story Tuesday. There's a name for it now: tokenmaxxing.
Amazon built an in-house chatbot called MeshClaw. It can answer emails, write code, and run small tasks for you. The company required 80% of developers to use it weekly, with a leaderboard counting each person's usage. Usage is measured in tokens, which is roughly the AI version of a phone bill: the more questions you ask, the higher your number climbs.
Developers gamed it quickly. They asked the AI for information their company's own documents already had, and started practice projects nobody had requested. One Microsoft engineer told The Pragmatic Engineer they purposely ask the AI for information they could find in their company's handbook, because looking it up in the handbook directly doesn't show up on the leaderboard.
Amazon's official line was that the leaderboard wouldn't count toward performance reviews. Several employees told the Financial Times they don't believe that. One said the default setup, where the AI can act on your behalf without asking first, terrifies them.
Meta had the same problem on a much worse scale. A staff engineer there built a leaderboard called Claudeonomics ranking all 85,000 Meta employees by AI usage. Top users got fake titles like "Token Legend." Meta workers used 60 trillion tokens in one 30-day stretch. At what AI companies charge customers, the bill comes to roughly $9 billion. Neither Mark Zuckerberg nor the CTO who publicly praised the practice made the top 250 of his own company. Meta killed the dashboard 48 hours after the news leaked.
Microsoft runs its own version. Nvidia CEO Jensen Huang says he'd be "deeply alarmed" if a $500,000-a-year engineer wasn't burning $250,000 worth of AI tokens annually.
UnitedHealth wants staff in its Optum services division doing at least one ChatGPT or Copilot query a day. Bloomberg reported it this week. It's the gentlest version of the leaderboard idea anyone has tried. The same wall is waiting: when usage is the metric, employees produce more usage. Productivity is a different conversation.
A 2026 industry survey by AI governance firm ModelOp found two-thirds of big companies rely on estimates to judge whether their AI spending is paying off. According to CNBC, nearly every Fortune 500 company now tracks some version of employee AI use. The Amazon story is what that turns into eight months later.
UnitedHealth is spending $1.5 billion on AI this year. Whatever metric they pick, developers will figure out how to game it before bonus season.
UnitedHealth Group,
$UNH, is tracking how often some employees use artificial intelligence tools as part of a push to embed the technology throughout its operations, per Bloomberg.