Amazon: Evaluate Software Engineers by Business Value, Not AI Usage

Amazon updated its software engineering guidelines to emphasize that performance should be evaluated based on the business value delivered rather than vanity metrics like AI usage volume or token consumption. The company acknowledged that while tracking metrics like token usage can offer operational insights, treating them as explicit performance targets risks driving counterproductive behaviors. This shift highlights a growing industry realization that traditional metrics fail to capture the true return on investment of AI coding tools. By discouraging empty productivity metrics, Amazon sets a precedent for tech giants trying to foster genuine efficiency rather than superficial compliance. Managers are instructed to investigate teams at both extremes: the bottom 10% to understand why they struggle with AI adoption, and the top 10% to verify if heavy AI usage genuinely accelerates delivery. Furthermore, engineers unable to benefit from AI are encouraged to explain why, and doing so demonstrates valuable human judgment.