Amazon Memo Urges Engineers to Use AI More Judiciously

Amazon has circulated internal guidance urging software engineers to be more selective in using artificial intelligence, discouraging so-called tokenmaxxing, where workers use AI tools extensively without producing commensurate value. The memo, circulated in September and posted to Amazon’s website on Tuesday, revises how the company defines the role of software development engineers and managers as AI reshapes software engineering. Amazon said it is still early in understanding AI’s impact but emphasized that AI will be central to future software engineering practices. The guidance comes as Amazon’s business units accelerate their use of AI under CEO Andy Jassy, who has said AI could change the speed and scope of daily work and may reduce some corporate roles through automation while creating new ones. Across the industry, companies are grappling with rising AI costs and the difficulty of measuring whether AI tools improve productivity, efficiency, or business outcomes. Some organizations have set limits on token usage, the units processed and generated by AI models, or shifted to lower-cost open-source tools to control spending. Amazon said it has not imposed caps on employee token usage. In May, the company shut down an internal leaderboard after discovering that some employees were using large amounts of AI tokens to climb rankings rather than deliver meaningful work. In the memo, Amazon described metrics such as token consumption, AI adoption rates, code changes, and code reviews as flawed indicators of AI value because they can encourage excessive spending and unnecessary complexity. The company did not tell employees to disregard those metrics entirely, but said token usage and code generation should be assessed with care and alongside longer-term business outcomes. Marc Brooker, a vice president and distinguished engineer at Amazon Web Services, said the company is focusing on productivity measures such as feature velocity, including launch frequency and successful production changes. He described Amazon’s approach as a return on investment, saying the company should invest tools, time, tokens, and human effort into problems that matter to customers and the business. Amazon also warned against using AI when it does not add value, saying that recognizing where AI does not yet help is a sign of engineering judgment rather than a failure.