Amazon told its engineers that token usage, AI adoption rates, and counts of code changes or reviews are “deeply flawed” measures of how teams use AI, in a memo it published on Tuesday after sharing it internally in September. The memo doesn’t tell staff to ignore those numbers, but it says they can drive up costs and complexity, so teams should weigh them with judgment and be measured on long-term results instead. Amazon also wants engineers to skip AI where it doesn’t help, and the memo says that when work doesn’t benefit from the tools, it’s “a finding, not a failing.” Amazon says it hasn’t capped token use, but in May it took down an internal leaderboard because employees were tokenmaxxing, running up AI usage regardless of output, to climb it. Wow, who could’ve guessed that would’ve happened if you put up a leaderboard?
