Amazon has issued new guidance for its software engineers on how to use artificial intelligence (AI) in their daily work. The memo, shared internally in September and published on 27 October, 2026, makes clear that the company wants AI to be used for solving customer problems and improving products — not simply for boosting usage statistics.
The company is discouraging a practice known as tokenmaxxing, where employees consume large amounts of AI resources without improving the quality of their work. Artificial intelligence systems process information in units called tokens, and while token prices have dropped, heavy usage can still drive up costs under usage based pricing models. Amazon warns that measuring productivity by token consumption, code generation, or the number of code reviews can give a false picture of effectiveness.
Instead, engineers are being asked to focus on outcomes such as successful product launches, better customer experiences, and measurable business impact. The memo also acknowledges that AI is not suitable for every task. Using judgment to decide when AI adds value — and when it does not — is considered a sign of maturity, not resistance.
Amazon’s move highlights a growing industry trend to tie adoption to quality, efficiency, and customer value, not raw usage metrics. This shift reinforces the need to train employees on responsible AI use, set performance measures around outcomes, and discourage wasteful practices that inflate costs without benefit.
By reframing AI as a tool for impact rather than volume, Amazon is signalling that the future of work will depend on smart application of technology, balanced with human judgment.
