Amazon keeps AI-use metrics, but warns engineers not to chase them

Amazon shut down an internal leaderboard in May after discovering employees were consuming more tokens to climb the rankings, a practice known as tokenmaxxing. The company is now warning software engineers against maximizing AI usage for its own sake while retaining usage measures as part of how it assesses work, CNBC reported.

The revised guidance circulated internally in September before being posted publicly on October 27. It asks employees to exercise judgment rather than treat token consumption or code generation as productivity targets. Tokens are the units of data AI models process and generate.

Managers are now encouraged to examine users at the 10th and 90th percentiles of AI adoption: understand why the lightest users are using the tools less, and check whether the heaviest users are increasing output accordingly, according to Fortune’s account of the guidance. Terk Terkowitz, vice president and distinguished engineer at Worldwide Amazon Stores, warned that turning usage measures into targets could raise costs, while rewarding lines of code could add unnecessary complexity.

“If your work does not benefit from current AI tooling, that is a finding, not a failing,” Terkowitz wrote in the memo, quoted by Fortune. Engineers should document the reasons, share them with peers and managers, and work toward solutions.

The expectations extend to managers’ own work. Leaders are expected to use AI tools in production, understand where they help or fall short, and explain those judgments to their teams, Business Insider reported.

Amazon has also looked at launches and successful changes in production to assess engineering productivity, Marc Brooker, a vice president and distinguished engineer at Amazon Web Services, told CNBC. He said the company wants to avoid spending tokens on problems that do not matter, while preserving room for experimentation.