Amazon memo urges engineers to use AI while warning against flawed metrics

Amazon says its “Renaissance Developer” combines AI tools with human judgment and domain expertise to build new things.[1] The company says a developer’s role and output are not defined by whether that person uses AI.[1] At the same time, Amazon describes AI as changing software engineering faster and more fundamentally than previous technology trends, and says AI will be central to future practice.[1] The guidance urges engineers who have not begun using AI to learn it and apply it to their daily work as soon as possible.[1]

Amazon says long-term outcomes are the best basis for evaluation, and calls the metrics under discussion deeply flawed when they are disconnected from what matters.[1] The memo warns that making input metrics into targets can raise costs and complexity, citing tokenmaxxing and lines-of-code measurement as examples.[1] It also says input metrics correlate well with important outputs and expects that to remain true for some time.[1] The tension is practical: Amazon does not reject input measures outright, but warns that treating them as goals can distort the work they are meant to represent.[1]

Amazon recommends examining teams at the 10th percentile of AI adoption to learn why they are not using AI effectively, and teams at the 90th percentile to check whether adoption is matched by delivery gains.[1] The company says sufficient understanding means using AI tools on production work, describing tasks where they helped and did not, and explaining that judgment to the team.[1] Amazon also says that documenting why AI does not help, sharing the reasoning with peers and a manager, and working toward solutions demonstrates the judgment it expects.[1]

Amazon says teams adopting AI most heavily have seen increased productivity and delivery, while teams or individuals not adopting AI or increasing output metrics are falling behind.[1] A CNBC report says Amazon has not imposed caps on employees’ token usage.[2] Amazon shut down its internal AI-use leaderboard in May after employees used tokenmaxxing to climb its ranks; The Register reported that workers created junk tasks to rise on KiroRank, an employee-built leaderboard.[2][3] The Register’s Quinn characterized the memo as an effort to make engineering’s future feel less bleak, while noting its mixed signals.[3] Quinn said he could not confirm whether the reported 250 job cuts announced that week were related to the memo.[3]