🔍 Read the full analysis: A Closer Look At Amazon’s AI Bets And The Meaning Of Measurable Results 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 saying software engineers should be judged by customer, business and team impact—not token counts or AI-tool usage. The memo still treats adoption data as a potentially useful signal, while the results of Amazon’s Nova models and its Anthropic investment offer contrasting company-level examples. Amazon’s third-quarter results were scheduled for October 29.
Amazon published new guidance for its software engineers on October 27, warning that token counts and other measures of AI activity should not be used as performance targets. The memo says engineers should be assessed on impact for customers, the business and their teams, a distinction that matters as Amazon invests in its own AI models and in outside model maker Anthropic.
The guidance was written by the company’s most senior engineers and applies to all Amazon software engineers. In a memo published on Amazon’s news site, Terk Terkowitz, a vice president and distinguished engineer in the stores business, said measures such as AI-tool adoption, tokens used, code changes and reviews can be misleading when turned into targets. The memo says those targets can raise costs and add complexity without establishing that the work produced a useful result.
The memo does not say that usage data has no value. It says those input measures currently “correlate well” with outcomes that matter, while warning against treating correlation as a reason to score employees by activity. It recommends examining both low and high adoption: leaders should investigate why some teams are not getting value from AI and whether heavy users are delivering more.
The guidance also asks leaders to use AI tools in their own production work and document cases where the tools helped or did not. It says an engineer who finds current AI tools do not benefit a particular task is “demonstrating the judgement we are asking for.” That frames non-use in a given case as a finding to explain, not automatically a performance failure.
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 Amazon Is Changing Its AI Scorecard
The memo addresses a practical problem facing employers adopting generative AI: activity is easier to count than value. Token totals, tool use and lines of code can rise even when a task takes longer, costs more or produces no improvement for customers. Amazon’s guidance makes clear that such figures may help managers spot patterns, but should not stand in for evidence of results.
For employees, the distinction affects how AI adoption may be evaluated. The memo asks teams to use the tools where they help, while allowing engineers to explain when they do not. For companies, it offers a more restrained way to assess adoption: compare work and outcomes, then use usage metrics as clues rather than quotas. The memo does not provide a new numerical measure of productivity or say how Amazon will audit whether teams follow the guidance.
The policy also lands as Amazon’s investments show different reported outcomes. Business Insider reported in July that most flagship Nova models had been moved to a “keep the lights on” status. Separately, Amazon’s June 30 filing valued its Anthropic holdings at about $190 billion, against roughly $18 billion invested, according to the source material. That is a paper valuation, not a realized return, and does not by itself establish the commercial performance of either company’s AI products.
Products Worth Considering
Logitech MX Keys S Wireless Keyboard, Graphite
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Black
As an affiliate, we earn on qualifying purchases.
From Token Leaderboards to Outcomes
The memo follows earlier reports of internal usage tracking. The Financial Times reported in May that some Amazon employees gave an internal agent tool, MeshClaw, unnecessary tasks to increase token counts. Those accounts were anonymous and remain reported claims, not details confirmed by Amazon in the material provided here. The FT also reported a weekly AI-use target for developers.
At the end of May, Amazon confirmed to Business Insider that an informal employee leaderboard called KiroRank had been “deprecated.” The company said the tracker was created by a group of employees and was never intended to encourage AI use for its own sake. The new guidance turns the broader concern into an official statement about how engineering work should be evaluated.
Amazon’s Nova models are part of its effort to develop AI systems for AWS customers. AWS lifecycle information lists Nova Canvas, Nova Reel and the first Nova Sonic as retired, while Business Insider reported that most flagship Nova models had been moved to maintenance status. Amazon’s Anthropic investment is a separate bet on an external model developer. The company’s third-quarter results, due October 29, were the next scheduled financial update, but the memo itself did not offer new earnings figures.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering memo
How Amazon Will Measure Results
The memo does not specify a single replacement metric, set thresholds for judging improved productivity or explain how managers will apply the guidance across teams. It also does not say whether existing internal adoption targets have been removed. The statement that usage metrics correlate with outcomes is the memo’s assessment; it does not establish that AI use caused those results.
The financial picture for the company’s AI efforts is also incomplete. The reported $190 billion Anthropic holding value is a balance-sheet figure, not cash received from a sale, and the source material does not detail how it was calculated. Business Insider’s account of Nova’s status and AWS’s retirement listings describe different aspects of the model portfolio; Amazon has not provided a full accounting of the returns or costs of its AI bets in the information cited here.
Earnings and Policy Follow-Through
Amazon’s next scheduled milestone was its third-quarter earnings report on October 29, one day after publication of the memo. Those results may offer more financial context, but the source material does not establish that Amazon planned to disclose AI-specific revenue, costs or returns.
The next test of the guidance will be how engineering leaders use it in practice: whether they review the reasons behind low adoption, check outcomes among heavy users and stop treating activity figures as goals. Any assessment of the company’s AI strategy will also depend on later evidence about customer demand, model performance and the financial contribution of Nova and Anthropic—not usage totals alone.
Key Questions
What did Amazon’s new engineering memo change?
It says software engineers should be evaluated on customer, business and team impact, rather than token counts or AI-tool activity used as performance targets.
Does Amazon want engineers to stop using AI?
No. The memo says adoption data can correlate with useful outcomes and encourages leaders to examine how teams use AI. Its warning is against using activity measures as goals or using AI solely to increase usage.
What is tokenmaxxing?
It refers to increasing the number of AI tokens used, including through unnecessary tasks, to improve a usage score or leaderboard position. Reports about Amazon employees doing this were attributed to anonymous accounts in the Financial Times report.
What did Amazon say about its KiroRank leaderboard?
Amazon confirmed to Business Insider that KiroRank had been deprecated. The company described it as an informal tracker made by employees and said it was not intended to promote AI usage for its own sake.
When were Amazon’s third-quarter results due?
The source material said Amazon was scheduled to report its third-quarter results on October 29, 2026, the day after the memo’s October 27 publication.
Source: ThorstenMeyerAI.com
