Amazon memo tells engineers to stop chasing AI token usage

On October 27, 2026, Amazon issued an internal memo setting out how it wants engineers to use AI. According to a CNBC report covered by DEV Community, the memo told engineers to stop chasing token counts and to avoid using token usage, tool adoption rates, or code change counts as targets, describing these practices as 'seriously flawed.' The memo said AI effectiveness should instead be judged by long-term business outcomes.

Before this memo, some Amazon employees had reportedly used an in-house agent platform to run unnecessary tasks in order to inflate their numbers on internal leaderboards, which rewarded usage rather than results. Amazon shut one such leaderboard down in May. CNBC's report framed the memo as an instruction to be more selective about AI use, pushing back against adopting AI just to accumulate tokens, and noted that Amazon, under CEO Andy Jassy, has been integrating AI quickly across its sprawling business operations.

Earlier still, Amazon engineering leaders had updated the role guidelines for software development engineers in the AI era, stressing that what matters most is the outcome, not which tools are used. A separate target, cited by Tom's Hardware and referenced in the coverage, was to have 80%+ of developers using AI weekly. With the memo now public, the situation stands at Amazon formally discouraging token counts, tool adoption rates, and code change counts as measures of AI success, with a reported internal push for 80%+ of developers using AI weekly.