WORLDTECHAmazon tells engineers to judge AI tools by delivery, not token use
According to TechSpot, Amazon has rewritten its engineering guidelines to warn staff against using AI token consumption as a measure of performance. The company wants AI coding tools assessed by the software they help ship.
What the revised guidelines advise
The revised engineering guidelines counsel against using AI adoption, token consumption, code changes and code reviews as performance goals. Managers are asked to weigh those figures alongside other indicators rather than reading them alone.
The company suggests looking at teams at both ends of the usage range. Low use may signal that engineers are overlooking ways to improve their work, while heavy use should raise the question of whether delivery is getting better enough to warrant it. As the report puts it, neither figure settles the matter by itself.
Tokenmaxxing and the earlier leaderboard
The guidance is aimed at tokenmaxxing, the deliberate inflation of AI consumption to appear productive on internal dashboards. Earlier this year, employees used MeshClaw, an internal AI agent that performs multi-step tasks, can deploy code and can interact with Slack, to push up their token counts.
Amazon closed an internal leaderboard in May after employees manipulated the rankings, according to the same report.
What Amazon says it has not done
Amazon states that it has not set caps on employees' token use. The guidance also allows engineers to conclude that a given task gains nothing from the tools on offer.
Open questions
- Will Amazon disclose how it weighs delivery indicators against token figures?
- Will closing the leaderboard lead to other changes in how AI use is tracked?