Amazon Updates AI Guidance to Discourage Tokenmaxxing
Amazon updated its role guidance for software engineers and managers on October 27, saying that token consumption, AI adoption rates, and code-change counts should not be standalone performance targets. The company said engineers should still learn and use AI, combining it with human judgment and domain expertise, while judging performance by long-term contributions to customers, the business, and their teams.
Amazon’s guidance provides a notable workplace policy for AI-assisted software engineering: managers are asked to test whether higher AI usage improves delivery rather than rewarding usage volume itself.
亞馬遜出手整頓 Tokenmaxxing 亂象,AI 使用量不再是工程師唯一績效指標 | TechNews 科技新報


亞馬遜(Amazon)於 10 月 27 日更新了軟體工程師與主管的角色指引,針對內部日益盛行的「tokenmaxxing」現象踩下煞車,明確提醒員工不應將 AI 使用量視為唯一的衡量標準。
根據公司最新的說法,Token 消耗量、AI 工具採用率,以及程式碼變更或審查次數,雖然皆可做為參考,但若將其視為目標而非觀察指標,可能會導致錯誤的誘因,甚至推高成本並增加系統複雜度。亞馬遜形容這些指標是「存在嚴重缺陷」的衡量方式,並強調在評估工程師表現時,應回歸到對客戶、業務與團隊的長期影響。