🔍 Read the full analysis: What Amazon’s Bets Say About Turning AI Use Into Outcomes on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Amazon published engineering guidance on October 27, 2026, warning that token counts and other AI-use measures can become harmful targets when they replace outcome-based judgment. The memo says those measures currently correlate with results, but asks leaders to examine whether both low and heavy AI adopters are delivering value. The guidance comes as Amazon’s Nova models and its Anthropic investment offer contrasting examples of AI bets.
Amazon on October 27 told its software engineers not to treat AI usage as a measure of success, publishing guidance that says customer, business and team outcomes matter more than token counts, tool adoption or code volume. The memo acknowledges that AI-use measures currently correlate with results, but warns that turning them into targets can increase costs and complexity instead of improving work.
The guidance was published on Amazon’s news site and comes from Terk Terkowitz, a vice president and distinguished engineer in the company’s stores business. It introduces a revised role guideline for software engineers, written by the company’s most senior engineers. The document does not set a fixed definition of AI use in the job, saying that rules written now could quickly become outdated. Instead, it centers the role on impact on customers, the business and the team.
The memo calls token usage, AI tool adoption, code changes and reviews “deeply flawed metrics” when used as targets. It says these measures are inputs, not the outcomes that matter, and warns that pursuing them as goals can drive costs up and make systems more complex. At the same time, the authors say the measures “correlate well” with important outputs today. The memo therefore does not tell engineers to stop using AI; it argues against scoring them simply for using it more.
For managers, the guidance recommends examining both the 10th percentile of AI adoption and the 90th percentile. The purpose is to understand why some teams use the tools little or not well, and whether the heaviest users are actually delivering more. It also asks leaders to use AI on their own production work and describe a task where it helped and one where it did not. Engineers whose work does not benefit from current tools are asked to document why and share that finding.
Six months at Amazon
From token leaderboards to an outcomes memo
Amazon spent 2026 learning the same lesson twice: once with how its engineers use AI, once with its own AI models.
- May 2026Token leaderboards backfireFinancial Times (via The Decoder): staff give an internal agent tool pointless tasks to inflate token counts. Reported, anonymous sources.
- Late May 2026KiroRank “deprecated”Amazon spokesperson to Business Insider. SVP Dave Treadwell: “Please don’t use AI just for the sake of using AI.”
- 24 Jul 2026San Francisco AGI Lab site closesConfirmed by Amazon to GeekWire. Frontier model research continues under Pieter Abbeel.
- 28 Jul 2026Most flagship Nova models wound downBusiness Insider, people familiar: Premier, Omni, Reel, Canvas to “keep the lights on”. Amazon: models “remain one of the most important things we’re working on”.
- 14 / 30 Sep 2026Nova Sonic v1, Nova Reel, Nova Canvas reach end of lifeAWS Bedrock model lifecycle page. Requests to them now fail.
- 26 Oct 2026Anthropic files its S-1Amazon notes: $6.2 billion principal, 12% ownership cap that Amazon can waive. Anthropic commits at least $110 billion to Amazon through April 2036.
- 27 Oct 2026The memo: outcomes, not tokenmaxxingUsage and throughput metrics are “deeply flawed” as targets. “Tokenmaxxing is at odds with frugality.”
- 29 Oct 2026Amazon’s third-quarter resultsAnnounced by Amazon. Not yet published when this article went live.
Sources: Amazon engineering memo (27 Oct 2026); Business Insider (29 May and 28 Jul 2026); GeekWire (24 Jul 2026); The Decoder (12 May 2026, summarising the Financial Times); AWS Bedrock model lifecycle (read 28 Oct 2026); Anthropic S-1 (26 Oct 2026).
Powered by Thorsten Meyer AI
Why AI Usage Targets Can Backfire
Amazon’s memo addresses a measurement problem facing companies adopting generative AI: activity is easier to count than value. Token totals and adoption rates can show whether tools are being used, but they do not by themselves establish that customers are better served, work is more reliable or costs have fallen. If a metric becomes a target, employees may optimize for the number rather than the work it was meant to represent.
The guidance matters to engineers and managers because it makes judgment part of the performance standard. A team that uses AI less may have a sound reason, while high usage is not proof of better results. For companies setting adoption goals, the memo offers a distinction: retain usage data as a diagnostic signal, but assess the effects on delivery, quality, cost and customers before treating adoption as success.
The policy also reflects a broader business question in Amazon’s AI strategy: whether investments produce outcomes, not just activity or attention. The company’s own models and its backing of Anthropic are different kinds of bets, but each invites scrutiny of what Amazon gets from the resources committed. The memo itself does not evaluate those investments, and the company’s upcoming earnings report will be a separate source of financial information.
Popular right now
Dyson Ball Animal 3 Corded Upright Vacuum
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Philips 3300 Series Fully Automatic Espresso Machine with LatteGo Milk System
As an affiliate, we earn on qualifying purchases.
From Leaderboards to Engineering Guidance
The memo follows reports that some Amazon employees used internal AI tools for unnecessary tasks to increase token counts on leaderboards. In May, the Financial Times reported that staff used an internal agent tool called MeshClaw this way and that AI usage was tracked. Those accounts were attributed to employees and should be treated as reported allegations, not as findings established by the new memo.
Amazon did confirm one related development: at the end of May, a spokesperson told Business Insider that the KiroRank leaderboard had been deprecated. The company described it as an informal tracker created by a group of employees and said it was never intended to encourage AI use for its own sake. Senior vice president Dave Treadwell also told staff, “Please don’t use AI just for the sake of using AI.” The October guidance turns that message into formal engineering role principles.
The article’s investment comparison draws on separate reporting and company records. Business Insider reported in July that most flagship Nova models had been moved to “keep the lights on”; AWS lifecycle information also lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. The source material says Amazon invested about $18 billion in Anthropic and that a June 30 filing valued its holdings at about $190 billion. Those figures describe a reported investment and a paper valuation, not realized cash returns.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering memo
What the Guidance Does Not Measure
The memo does not provide public data showing how much AI has improved Amazon’s engineering output, reduced costs or changed customer outcomes. Its statement that usage metrics currently correlate well with results is not accompanied in the source material by figures, a measurement period or details of how that relationship was assessed. The guidance also does not set a replacement numerical score or define one universal way to measure impact across different engineering teams.
The reported tokenmaxxing accounts came from news reports and anonymous employees; Amazon’s confirmed statement about KiroRank does not independently verify every reported behavior or establish how widespread it was. The company’s Nova and Anthropic figures also come from distinct sources and describe different things: model status, investment and a reported paper valuation. They do not establish a direct comparison of financial returns. Amazon’s October 29 earnings report may provide additional company-wide information, but its results will not by themselves show whether this specific engineering guidance is working.
Amazon’s Earnings and Adoption Data
Amazon is scheduled to report third-quarter results on October 29, one day after the memo’s publication. The company’s earnings may add context about its wider business and AI investments; the available source material does not supply the results in advance. The memo itself sets no public deadline for a follow-up assessment.
The next test for the engineering guidance will be how managers use the recommended reviews of low- and high-adoption teams. If Amazon later reports whether greater AI use improved delivery, quality, customer impact or costs, readers will have a clearer basis for judging the policy. For now, the stated standard is to treat usage figures as signals to investigate, not as a score of individual or team performance.
Key Questions
What did Amazon tell its software engineers?
Amazon said engineers should be judged by customer, business and team impact, not by token counts or how often they use AI tools. The memo cautions that usage measures can be misleading when made performance targets.
Does the memo say engineers should stop using AI?
No. The guidance says AI-use measures currently correlate with important outputs. It argues that managers should assess whether AI use produces results rather than reward usage for its own sake.
What does “tokenmaxxing” mean here?
It refers to using an AI tool in ways that increase token counts without a useful work outcome. Amazon’s memo says that behavior conflicts with its principle of frugality.
What should happen when AI does not help with a task?
The memo says that lack of benefit is a finding, not a failing. It asks engineers to document why the tools did not help and share that information with peers and managers.
When is Amazon’s next earnings report?
Amazon is scheduled to report its third-quarter results on October 29, 2026. The source material does not include those results or say that the report will address the engineering guidance directly.
Source: ThorstenMeyerAI.com
