🔍 Read the full analysis: Amazon’s AI Bets Show Why Token Use Isn’t The Whole Scorecard 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 Oct. 27 saying developers should be judged by customer, business and team outcomes—not token use or other AI activity counts. The memo still says those measures can correlate with results, while Amazon’s Nova and Anthropic investments illustrate why AI activity alone is an incomplete scorecard.
Amazon published new guidance for its software engineers on October 27, warning that token counts and other AI-use measures should not become performance targets. The memo says engineers should be evaluated on impact for customers, the business and their teams—a distinction that matters as companies try to judge whether AI tools produce useful results rather than simply more activity.
The guidance was written by Amazon’s most senior engineers and applies to all software engineers, according to the company’s publication. Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business, introduced the revised role guideline. It does not define an engineer’s job by a particular AI tool, reflecting the memo’s view that such definitions could quickly become outdated.
The memo describes token usage, AI-tool adoption, code changes and reviews as inputs rather than outcomes. It calls them “deeply flawed metrics” when used as targets, saying that doing so can raise costs and add complexity. At the same time, it says those measures currently “correlate well” with outcomes that matter. The memo’s position is not that engineers should stop using AI, but that managers should not mistake the amount of use for the value produced.
Amazon’s guidance asks leaders to examine both low and high levels of AI adoption: the 10th percentile to understand why some teams use the tools less, and the 90th percentile to check whether heavy users are actually delivering more. It also asks leaders to use the tools on their own production work and identify tasks where they help and where they do not. Engineers whose work does not benefit from current AI tools should record and share that finding, the memo says.
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 Activity Is Not the Outcome
The memo addresses a measurement problem for employers adopting generative AI: usage is easy to count, while improvements in customer experience, quality, speed or cost can take longer to establish. Treating a count as a goal can reward behavior that increases activity without improving the work. Amazon’s examples make the risk concrete: tokenmaxxing conflicts with the company’s stated principle of frugality, while using lines of code as a target conflicts with “invent and simplify.”
For workers, the guidance signals that heavy AI use is not automatically a measure of strong performance. It also gives employees a way to report when a tool does not help, rather than treating low adoption as a failure by itself. For companies, the approach suggests using usage information to investigate how tools are working, while judging success by what changes for customers and the business.
Amazon’s own AI portfolio is relevant to that distinction, though it does not prove the memo was written in response to the performance of any particular product. The company’s in-house Nova models and its investment in Anthropic represent different bets. Their reported progress and financial value cannot be reduced to the volume of AI activity inside Amazon.
Products Worth Considering
Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Black
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
DJI Mic 3 Wireless Microphone System (2 Transmitters, 1 Receiver, Charging Case)
As an affiliate, we earn on qualifying purchases.
From Internal Rankings to Formal Guidance
The memo followed reports of internal efforts to track AI use. In May, the Financial Times reported that Amazon employees assigned unnecessary tasks to an internal agent tool, MeshClaw, to increase token counts. The FT also reported a company target for more than 80% of developers to use AI weekly and internal token-use leaderboards. Those details were based on reported accounts, including anonymous employee accounts, and should not be treated as confirmed company-wide policy.
Amazon did confirm to Business Insider at the end of May that an informal employee-built leaderboard called KiroRank had been “deprecated.” The company said the tracker was not 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 a formal statement about how engineering roles should be assessed.
The source article also describes contrasting outcomes in Amazon’s model strategy. Business Insider reported in July that most flagship Nova models had been moved to “keep the lights on”; AWS lifecycle information lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. Separately, the source reports that 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 a realized return.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering memo
What the Metrics Do Not Show
The memo does not provide data quantifying how much AI use has improved productivity, software quality or customer outcomes across Amazon. It says input measures correlate with important outputs at present, but gives no figures, time period or method for that relationship. It is also unclear how individual teams will translate the guidance into performance reviews or which outcome measures they will use.
The reported Nova status and Anthropic valuation provide different kinds of evidence: product lifecycle decisions and a paper value attached to an investment. Neither, on its own, establishes the overall financial return from Amazon’s AI strategy. Amazon’s third-quarter results, scheduled for October 29, may provide further company-level information, but the source material does not supply those results.
Amazon’s Results and Team Measures
Amazon is scheduled to report third-quarter results on October 29, 2026, one day after the memo’s publication. Those results may shed light on the company’s broader business and investment performance, but they are not a direct test of whether the new engineering guidance improves software work.
The next practical test will be how managers apply the guidance: whether they use adoption and token data as diagnostic signals, examine both low and high usage, and compare AI-assisted work with outcomes that matter to customers and teams. The memo does not set a timetable for reviewing those measures or describe a new company-wide scoring system.
Key Questions
What did Amazon’s October memo tell engineers?
It said engineers should be judged on customer, business and team impact, not on token counts or other AI-use measures treated as targets.
Is Amazon telling developers to stop using AI?
No. The memo says usage measures can correlate with results and asks leaders to examine adoption. Its warning is against treating how much AI is used as the result itself.
What does “tokenmaxxing” mean in this story?
It refers to using an AI tool in ways that increase token consumption or leaderboard rankings without a useful work outcome. The memo says that approach conflicts with Amazon’s principle of frugality.
What is unclear about Amazon’s AI performance?
The memo gives no figures showing how AI use affects productivity, quality or customer outcomes across the company. The reported status of Nova and the paper valuation of Amazon’s Anthropic stake also do not, by themselves, establish the overall return on its AI strategy.
Source: ThorstenMeyerAI.com
