Amazon tells engineers to stop gaming its AI leaderboard. There was an AI leaderboard.

Amazon published revised guidelines for software engineers on Tuesday, announcing the age of the "Renaissance Developer" who uses AI thoughtfully rather than indiscriminately. The document is sensible enough. The context that required it is somewhat less flattering.

The context: Amazon engineers were running AI agents on deliberately meaningless tasks to inflate their token counts and climb an unofficial internal leaderboard called Kirorank. Amazon shut Kirorank down in May, after the practice started generating real computing costs for no real output. A separate unnamed company, per Axios, accidentally spent $500 million on Claude in a single month after neglecting to put any limits on employee usage.

So now Amazon has a word for the behaviour it accidentally incentivised: tokenmaxxing. Using more AI tokens as a proxy for working harder, regardless of whether any of it was useful. The memo notes, with commendable patience, that this is "at odds with frugality."

The VP who wrote the memo, Terk Terkowitz, does land one genuinely good line: "If your work does not benefit from current AI tooling, that is a finding, not a failing." That's worth saying. It just shouldn't have taken a $500 million accident at a competitor and an internal leaderboard gamification scandal to make it worth saying.