Overview
- The company posted a memo Tuesday that tells software engineers to be selective with AI and discourages 'tokenmaxxing,' the practice of driving up token use to signal productivity.
- The memo formalizes earlier internal moves, including the deprecation of an internal leaderboard that ranked staff by token consumption and executive warnings this year about gaming metrics.
- Amazon defines tokens as the billing units for model prompts and responses and says raw token counts, AI-adoption rates, and simple code metrics are 'deeply flawed' measures that can raise costs and complexity.
- The guidance says choosing not to use current AI tooling for a task is an acceptable, documentable finding and that teams should be judged by long-term customer and business outcomes rather than token volume.
- Amazon stopped short of imposing firm usage caps and instead asked managers to apply judgment, monitor outliers, and watch for wasted spend—a stance that could influence industry norms for measuring AI productivity.