🔍 Read the full analysis: What Does AI Success Look Like? Amazon’s Bets Offer A Perspective on ThorstenMeyerAI.com
Get hardware and tech essentials 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’s engineering leadership said software engineers should be judged on customer and business impact, not token counts or AI-tool usage. The memo’s message comes as reports point to a reduced role for several Nova models and Amazon’s Anthropic investment has risen substantially in value on paper; Amazon’s third-quarter results are due October 29.
Amazon published engineering guidance on October 27 telling software teams not to treat AI usage, token counts or code volume as measures of success. The memo says engineers should be assessed on impact for customers, the business and their teams—a distinction that lands as Amazon’s own Nova models and its investment in Anthropic show different results, and the company prepares to report quarterly earnings.
The memo, written by Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business, introduces a revised role guideline for software engineers. Amazon said the guideline was written by its most senior engineers. It does not prescribe a fixed set of AI duties, arguing that specific definitions could quickly become outdated as tools change.
Instead, the document distinguishes between outcomes and inputs. Customer and business outcomes can take time to measure, so teams have used measures such as AI-tool adoption, token usage, code changes and reviews as proxies. The memo calls those measures “deeply flawed” when used as targets, warning that they can increase costs and complexity without demonstrating better results. It also says the input measures currently “correlate well” with outputs that matter. That qualification means Amazon is not dismissing usage data; it is rejecting its use as a standalone score.
The guidance asks leaders to examine both low and high levels of AI adoption: why some teams use the tools less, and whether heavy users are actually delivering more. It also calls for managers to use AI on their own production work and describe when it helped and when it did not. The memo says an engineer’s conclusion that current AI tools do not help a particular task can be useful evidence, not 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
Amazon’s Outcome Test for AI
The memo matters because companies are still working out how to tell whether workplace AI produces value. Usage is easy to count; improved customer outcomes, faster delivery or lower costs are harder to attribute and may take longer to appear. Amazon’s position is that the easier measure should not replace the harder one, especially if staff can improve their scores simply by generating more AI activity.
That distinction affects both managers and workers. A token leaderboard can reward activity that consumes computing resources without improving a product. Conversely, a team that uses AI selectively might deliver useful work without appearing highly engaged under a usage target. Amazon’s proposed approach asks leaders to look at adoption data as a diagnostic, then investigate what the numbers mean rather than treating them as proof of performance.
The corporate stakes are visible in Amazon’s own portfolio, though the comparison is not a controlled test. The source material reports that several Nova products were retired or moved to maintenance, while Amazon’s Anthropic stake rose sharply in value on paper. Those developments show that building a model and investing in an outside model maker carry different risks and outcomes; they do not by themselves establish which strategy will produce lasting customer or business gains.
Products Worth Considering
Logitech MX Keys S Wireless Keyboard, Graphite
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
Apple iPad Air 11-inch (M2), 512GB, Wi-Fi, Space Gray
As an affiliate, we earn on qualifying purchases.
From Token Leaderboards to Model Bets
The memo follows reports of internal efforts to increase AI usage. In May, the Financial Times reported that some Amazon employees assigned unnecessary tasks to an internal agent tool, MeshClaw, to raise token counts. Those accounts were anonymous and are reported allegations, not independently established facts here. The same reporting described a target for more than 80% of developers to use AI weekly and internal tracking of token use.
Amazon did confirm to Business Insider that an employee-created leaderboard called KiroRank had been “deprecated.” The company described it as an informal tracker 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 memo turns that broader caution into formal engineering guidance.
On model development, Business Insider reported in July that most flagship Nova models had been moved to “keep the lights on.” Amazon’s lifecycle information, as cited in the source material, lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. Separately, Amazon invested about $18 billion in Anthropic; a June 30 filing reportedly carried the holdings at about $190 billion. That is a reported paper valuation, not cash proceeds or a measure of operating performance.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering memo
What the Scorecard Cannot Show
The memo does not provide a universal method for attributing improvements to AI, or specify how teams should weigh customer outcomes against delivery speed, cost and quality. It acknowledges that long-term results are slow to measure, leaving managers to exercise judgment while using adoption figures as signals rather than targets.
It is also unclear how much the reported Nova changes reflect a broader retreat from Amazon’s model-building effort, or how the company will develop and market its remaining models. The reported $190 billion value for Amazon’s Anthropic holdings is a valuation recorded in a filing, not a realized return. The available material does not establish how either model strategy will perform over time or what Amazon’s earnings will show.
Earnings and AI Measures Ahead
Amazon is scheduled to report its third-quarter results on October 29, one day after the memo’s publication. Those results may provide updated financial information, but they will not by themselves settle whether Amazon’s AI tools or model investments are improving customer outcomes. The source account says it will not anticipate the figures.
Within the company, the next test is whether teams apply the guidance in performance discussions and project reviews. Leaders are being asked to inspect low and high adoption, use AI themselves on real work and explain where it helps. Any clearer evidence of impact will depend on how Amazon measures results over time and what the company discloses about its AI products and investments.
Key Questions
What did Amazon’s memo change?
It introduced revised engineering guidance that places customer, business and team impact ahead of token counts or AI-tool usage as measures of performance.
Is Amazon telling employees to stop using AI?
No. The memo says AI adoption can correlate with valuable outputs and asks leaders to examine how teams use the tools. Its objection is to treating usage itself as a performance target.
What does “tokenmaxxing” mean here?
It refers to increasing AI token usage—text processed or generated by a model—without a useful work outcome, such as by giving a tool unnecessary tasks.
What is known about Amazon’s Nova models and Anthropic investment?
The source material cites reporting that most flagship Nova models were moved to maintenance and says several Nova products are listed as retired. It also reports that Amazon’s Anthropic stake was valued at about $190 billion in a June 30 filing after an investment of about $18 billion. That figure is a paper valuation, not realized cash.
When will Amazon report its next quarterly results?
Amazon is scheduled to report third-quarter results on October 29, 2026. The figures and any additional detail about AI performance remain unknown until the company reports.
Source: ThorstenMeyerAI.com
