🔍 Read the full analysis: What Counts As AI Progress? Amazon’s Bets Point To Outcomes on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Amazon published engineering guidance on October 27 saying software teams should judge AI by its effect on customers, the business and colleagues—not by token counts or tool adoption. The guidance comes as Amazon’s own Nova model lineup has been scaled back and its Anthropic investment has gained substantial reported value; the full financial picture remains to be seen in the company’s October 29 earnings.
Amazon published new guidance for its software engineers on October 27, warning that targets based on AI token use, tool adoption or code volume can reward activity rather than results. The guidance says engineers should be assessed on impact for customers, the business and their teams, as the company’s own AI-model efforts and its investment in Anthropic offer contrasting examples of where its AI strategy stands.
The document, published on Amazon’s news site, introduces revised role guidance for software engineers and was written by the company’s most senior engineers. It was presented by Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business. The guidance says long-term outcomes are difficult to measure quickly, which led some teams to use inputs such as token usage, adoption of AI tools, code changes and reviews as proxies.
Amazon calls those inputs “deeply flawed metrics” when used as targets. The document warns that optimizing for them can increase costs and system complexity. It does not say teams should stop using AI: the memo says adoption metrics currently “correlate well” with important outputs, while cautioning that correlation does not make usage itself the goal.
The guidance asks leaders to examine both low and high adopters. The 10th percentile can help explain why some teams do not use the tools effectively; the 90th percentile can show whether heavy use produces better work. Engineers are also asked to record cases where AI does not help, while leaders should test the tools on their own production work and explain where they did and did not prove useful.
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 Usage Targets Can Mislead
The guidance addresses a measurement problem facing companies adopting generative AI: activity is easier to count than value. Token volumes, tool logins and lines of code can be recorded quickly, but they do not establish that a product improved, work became faster or customers benefited. Treating those indicators as quotas can encourage employees to use AI even when it adds little, with possible costs in computing and maintenance.
Amazon’s position is more qualified than a rejection of adoption tracking. The memo says usage indicators can help identify patterns, and that teams adopting AI are currently delivering more. But leaders are told to use the metrics diagnostically, then examine the work itself. The distinction is between measuring adoption and rewarding it: a high count may prompt a closer look, not settle whether a team performed well.
For workers, the guidance gives official standing to reporting that an AI tool is not useful for a task, provided they explain why. For companies, it puts customer and business outcomes ahead of easily gamed internal scoreboards. Whether this approach changes performance reviews or investment decisions is not specified in the published guidance.
Products Worth Considering
Jackery HomePower 3000 Portable Power Station with Two 200W Solar Panels
Apple AirPods Pro 3 Wireless Earbuds
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
As an affiliate, we earn on qualifying purchases.
Amazon’s Two AI Bets
The memo follows earlier reports of usage tracking inside Amazon. The Financial Times reported in May that some employees assigned unnecessary tasks to an internal agent called MeshClaw to raise token counts. That account relied on employee reports and should be treated as attributed reporting, not as a finding confirmed in Amazon’s October guidance. Amazon did confirm to Business Insider at the end of May that an informal employee-created leaderboard called KiroRank had been “deprecated,” saying it was not intended to encourage AI use for its own sake.
The company’s model portfolio offers a second part of the story. Business Insider reported in July that most flagship Nova models had been moved to “keep the lights on” status. Amazon’s lifecycle information lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. These developments indicate changes to parts of the lineup, but do not establish that Amazon has abandoned its broader model program.
Amazon has also invested in Anthropic, a model developer it does not control. The supplied source says Amazon invested about $18 billion and that a June 30 filing valued its holdings at about $190 billion. Those figures describe a reported, paper valuation, not cash proceeds or realized profit. Amazon was due to report third-quarter results on October 29, one day after publication of the guidance.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering guidance, published October 27
What the Guidance Leaves Open
Amazon has not said how the new guidance will affect individual performance reviews, promotions or team targets, or whether any existing usage dashboards will change. The document does not provide a replacement numerical score for AI’s contribution; instead, it calls for judgment and examination of actual work.
The reported status of Nova models and the valuation of Amazon’s Anthropic stake also require care. A paper valuation is not a realized return, and the supplied figures do not explain how Amazon calculates the full economics of the investment. The October 29 earnings report may offer company-wide financial results, but the source does not establish that Amazon will break out AI returns or model performance in detail.
Earnings and Engineering Practice
Amazon’s third-quarter results, scheduled for October 29, 2026, are the next near-term company update identified in the source. They may provide more information about Amazon’s overall financial performance, though whether they will clarify the returns from its AI investments is unknown.
Inside engineering, the practical test will be whether managers use the new guidance to examine work outcomes rather than set token or adoption quotas. The memo directs leaders to try AI tools on their own production tasks and share examples of where they help or fail. It does not give a timetable for reviewing the policy or report results from its rollout.
Key Questions
What did Amazon announce?
Amazon published revised guidance for software engineers saying AI use should be judged by outcomes for customers, the business and teams, rather than by token counts or tool adoption alone.
Is Amazon telling engineers to stop using AI?
No. The memo says adoption indicators can correlate with important outputs and asks leaders to examine them. Its warning is against turning usage into a target that rewards activity without showing value.
What does the guidance say about AI tools that do not help?
Engineers are encouraged to document why a tool does not benefit their work. The memo describes that as a finding rather than a failure, making task-specific judgment part of the evaluation.
What does Amazon’s Anthropic investment valuation mean?
The supplied source says Amazon’s holdings were valued at about $190 billion on June 30, against about $18 billion invested. That is a reported paper valuation; it should not be read as cash received or a realized profit.
When is Amazon’s next reported milestone?
Amazon is scheduled to release its third-quarter results on October 29, 2026. It is not clear whether the company will provide detailed figures on returns from AI models or its Anthropic investment.
Source: ThorstenMeyerAI.com
