🔍 Read the full analysis: What Amazon’s AI Bets Tell Us About Measuring Value on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Amazon’s October 27 engineering memo warns that token counts and other AI-use metrics can become costly, misleading targets, even though they currently correlate with useful results. The message lands alongside contrasting developments in Amazon’s AI portfolio: reports say several Nova models were wound down, while Amazon’s Anthropic stake was valued far above its investment on paper.
Amazon published new guidance on October 27 telling software engineers to measure AI’s value by its effects on customers, the business and their teams—not by how many tokens they use. The memo matters beyond internal performance reviews: it arrives as Amazon’s own Nova models face reports of a pullback, while its investment in Anthropic has risen sharply in reported paper value.
The memo, written by Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business, sets out revised role guidance for the company’s software engineers. Amazon says the guidance was written by its most senior engineers and deliberately avoids defining the job around particular AI tools. The authors say that rules focused on current tools could quickly become outdated; they instead center the role on impact and judgment.
Amazon describes token usage, AI-tool adoption, code changes and reviews as “deeply flawed metrics” when used as targets. The memo says such inputs can be disconnected from meaningful outcomes, raise costs and add complexity. It gives two examples: tokenmaxxing conflicts with the company’s frugality principle, and targeting lines of code conflicts with “invent and simplify.”
The guidance does not say those measures are useless. It says they currently “correlate well” with outcomes that matter, and that teams adopting AI most heavily are delivering more. The distinction is between using metrics to spot patterns and using them to score workers. Amazon advises leaders to examine both low and high adoption, understand why adoption differs, and test whether heavy users are producing better results.
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 Rejects AI Usage Scores
The memo addresses a measurement problem facing companies adopting generative AI: activity is easier to count than value. Token totals, tool adoption and code changes can show that a system is being used, but do not by themselves establish that work improved or customers benefited. If workers are rewarded for those numbers, they may use AI where it adds little value or generate unnecessary output.
For Amazon, the stakes include both productivity and cost. The memo warns that treating inputs as targets can increase spending and system complexity. Its approach is to retain usage data as a signal while asking managers to investigate outcomes. That gives workers room to report that a tool did not help a particular task, rather than treating non-use as automatic underperformance.
The company’s portfolio adds a concrete, though incomplete, comparison. Nova’s reported retreat suggests that building models is not itself proof of success; the Anthropic stake’s reported valuation suggests a different kind of return, but one measured on paper. Neither comparison alone establishes the customer impact or long-term economics of Amazon’s AI strategy.
Products Worth Considering
Dell XPS 13 9350 AI Professional Laptop 13.4 120Hz FHD+ Display
Dell XPS 13 9350 Laptop, 13.4 FHD+ Display, Intel Ultra 7 256V, 16GB DDR5, 1TB SSD
Logitech MX Keys S Wireless Keyboard, Graphite
As an affiliate, we earn on qualifying purchases.
From Token Leaderboards to Model Bets
The memo follows reports of internal efforts to track AI use. The Financial Times reported in May that some Amazon employees assigned unnecessary tasks to the internal agent tool MeshClaw to increase token counts. The report also described a weekly AI-adoption target and internal leaderboards. Those accounts were anonymous and are reported claims, not details independently confirmed in the supplied material.
Amazon did confirm to Business Insider at the end of May that a leaderboard called KiroRank had been deprecated. The company described it as an informal tracker created by employees and said it was not intended to encourage AI use for its own sake. Senior vice president Dave Treadwell also told staff not to use AI just for the sake of using AI. The October memo now puts that principle into official engineering guidance.
Amazon’s model-building and investment choices offer further context. Business Insider reported in July that most flagship Nova models had been moved to “keep the lights on.” Amazon’s lifecycle page lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. Separately, the source material says Amazon invested about $18 billion in Anthropic and that a June 30 filing valued its holdings at about $190 billion. That is a reported paper valuation, not cash proceeds from a sale.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering memo
What the Results Cannot Yet Show
The memo does not provide a company-wide estimate of how much AI has improved output, reduced costs or changed customer outcomes. Its statement that adoption metrics currently correlate with useful outputs is not accompanied in the supplied material by figures, a measurement period or a description of how the correlation was established. Correlation is not a measured return, and the memo does not set out a replacement numerical scorecard.
The status and performance of Nova models also remain unevenly described. Business Insider’s report concerns most flagship models, while Amazon’s lifecycle page identifies specific retired services; neither establishes that every Nova model or related product has been discontinued. The Anthropic figure is a valuation reported in a filing, not evidence that Amazon has realized the gain or that it will persist.
Amazon’s third-quarter results are due October 29, after the memo’s publication. They may add financial context, but the supplied information does not establish what the company will disclose about AI spending, Nova or returns from its Anthropic investment. The results should not be assumed to answer those questions unless Amazon reports the relevant details.
Amazon’s October 29 Results
The next scheduled milestone is Amazon’s third-quarter results on October 29. Investors and employees will be looking for any company disclosures that help explain AI-related costs, model strategy and the business contribution of its partnerships. No specific AI figures or management comments have been confirmed in the supplied material, so the results may not resolve the measurement questions raised by the memo.
Inside engineering, the guidance points to a practical next step rather than a new leaderboard: managers are asked to use AI tools on their own production work, compare tasks where they help with tasks where they do not, and share the reasons. Amazon’s stated test is whether the work produces better outcomes for customers, the business or the team—not simply whether usage rises.
Key Questions
What did Amazon tell engineers about AI use?
Amazon said engineers should focus on customer, business and team impact, rather than token counts or other usage measures as performance targets.
Does Amazon say AI adoption metrics have no value?
No. The memo says these metrics currently correlate with outcomes, but warns against turning them into targets. It calls for using them to investigate patterns, not as a direct measure of individual performance.
What happened to Amazon’s KiroRank leaderboard?
Amazon told Business Insider that KiroRank, an informal employee-created tracker, had been deprecated. The company said it was not intended to promote AI use for its own sake.
Were Amazon’s Nova models discontinued?
The supplied material says Business Insider reported that most flagship Nova models were moved to “keep the lights on,” and Amazon’s lifecycle page lists several Nova services as retired. That does not confirm that all Nova models were discontinued.
When will Amazon report its next results?
Amazon’s third-quarter results are scheduled for October 29, 2026. Whether the company provides further details about AI costs or returns is not yet clear.
Source: ThorstenMeyerAI.com
