What happened
Amazon has formalized internal guidance telling software engineers to be selective about when they use AI, and published it on the company's website on October 27 (Source: https://particle.news/story/amazon-formalizes-guidance-discouraging-tokenmaxxing-among-engineers). The memo, which circulated internally in September and was reported by CNBC, explicitly warns against “tokenmaxxing”, the practice of driving up token use to signal productivity, and says token use, AI adoption rates, code changes and code reviews are deeply flawed measures when used to assess teams or organizations (Source: https://dailyaibrief.com/news/amazon-warns-engineers-against-tokenmaxxing-VBSj1CHT; Source: https://www.cnbc.com/2026/10/27/amazon-ai-coding-tokenmaxxing.html). Amazon defines tokens as the billing units for model prompts and responses, and says raw token counts and simple code metrics do not capture engineering value (Source: https://particle.news/story/amazon-formalizes-guidance-discouraging-tokenmaxxing-among-engineers).
What the memo says
The guidance is unusually direct for a company that has pushed hard on AI adoption. It tells engineers that “tokenmaxxing is at odds with frugality”, one of Amazon's leadership principles, and that measuring lines of code is at odds with “invent and simplify”, while acknowledging that the same input metrics correlate well with the outputs that matter (Source: https://fourweekmba.com/ai-amazons-new-engineer-guide-puts-outcomes-above-ai-token-use/). On when to skip AI entirely, the memo states: “If your work does not benefit from current AI tooling, that is a finding, not a failing,” and adds that “an engineer who can explain where AI does not yet help their work is demonstrating the judgement we are asking for” (Source: https://jingletree.com/amazon-s-anti-tokenmaxxing-memo-says-engineers-should-be-pickier-about-using-ai-296308.html).
The KiroRank backstory
The memo formalizes lessons Amazon learned the hard way. In late May 2026, Amazon shut down KiroRank, an internal leaderboard that ranked software engineers by how many tokens they consumed on Kiro, the company's AI coding platform, after employees ran pointless tasks through the tool to inflate their scores (Source: https://www.aipolicydesk.com/blog/ai-adoption-metrics-perverse-incentives-2026). Senior Vice President Dave Treadwell told staff at the time: “Please don't use AI just for the sake of using AI” (Source: https://aiuntethered.com/news/amazon-ai-leaderboard-elimination-rising-costs/). Earlier company policy had reportedly required a large share of employees to use AI weekly, which pushed staff to game the metric, and media estimates put the cost of the token-burning behavior at hundreds of millions of dollars per month (Source: https://m.36kr.com/p/3835782264894595). Employee accounts described how the leaderboard, linked to internal awards, incentivized gaming the system, with some staff running scripts that fired endless non-productive prompts at AI models to inflate usage statistics (Source: https://agentupdate.ai/news/amazon-shuts-down-ai-leaderboard-after-token-cheating/). Meta also removed its own “Claudeonomics” leaderboard within days of it leaking in April, showing the pattern is industry-wide (Source: https://corti.com/from-tokenmaxxing-to-token-discipline-the-2026-reckoning-in-ai-assisted-engineering/).
Amazon's framing: the Renaissance Developer
Amazon's engineering leadership framed the guidance positively: what matters most is the outcome, not which tools are used, and the era now belongs to the “Renaissance Developer” who blends AI tools with human judgment and domain expertise to build entirely new things (Source: https://www.aboutamazon.com/news/company-news/amazon-software-engineer-role-guideline-ai). Analysis of the memo notes that Amazon treats spending as a signal, not a goal, and credits teams that have embraced AI most, which Amazon calls “frontier teams” (Source: https://fourweekmba.com/ai-amazons-new-engineer-guide-puts-outcomes-above-ai-token-use/). The revision underscores that the developer role and its output are not defined by AI use (Source: https://www.aboutamazon.com/news/company-news/amazon-software-engineer-role-guideline-ai).
Why it matters for enterprise AI buyers
For B2B buyers, Amazon's reversal is a data point about how even the most aggressive AI adopters are recalibrating: they are moving from raw AI usage as a KPI to outcome-based evaluation. Employee commentary on Hacker News has long questioned whether official claims that token stats would not factor into performance evaluations hold up in practice, since managers still scrutinize the data (Source: https://chyshkala.com/blog/amazon-s-100m-token-burning-scheme-exposes-big-tech-s-ai-metrics-madness). The broader industry takeaway, as analysts summarize the 2026 shift, is a move from tokenmaxxing to token discipline: measuring AI investments by business results rather than by adoption volume, and treating cost as a signal to optimize rather than a score to maximize (Source: https://corti.com/from-tokenmaxxing-to-token-discipline-the-2026-reckoning-in-ai-assisted-engineering/).
What the memo signals about AI governance
The memo is also a governance document, and that is the part enterprises should copy. Amazon has moved from asking employees to maximize AI usage to asking them to maximize judgment, and it has written down what judgment means in this context: knowing when AI does not help is a finding, not a failing (Source: https://jingletree.com/amazon-s-anti-tokenmaxxing-memo-says-engineers-should-be-pickier-about-using-ai-296308.html). For companies building their own AI adoption policies, the sequence of events at Amazon, leaderboard, gaming, cost blowout, shutdown, formal guidance, is a case study in why usage metrics without outcome metrics generate perverse incentives (Source: https://www.aipolicydesk.com/blog/ai-adoption-metrics-perverse-incentives-2026). The practical design lesson is to tie AI measurement to business outcomes, publish the rules explicitly, and remove any metric that employees can game by spending tokens (Source: https://corti.com/from-tokenmaxxing-to-token-discipline-the-2026-reckoning-in-ai-assisted-engineering/). Amazon's revised role guideline, emphasizing that the developer role is not defined by AI use, is a model for how large organizations can institutionalize that balance (Source: https://www.aboutamazon.com/news/company-news/amazon-software-engineer-role-guideline-ai).
For buyers of AI tools and cloud services, the memo carries a pricing implication as well: if enterprises stop optimizing for token volume and start optimizing for outcomes, demand patterns for model usage will shift toward higher-value, more selective workloads, which is a factor to watch in AI procurement budgets for 2027 (Source: https://fourweekmba.com/ai-amazons-new-engineer-guide-puts-outcomes-above-ai-token-use/).
