🔍 Read the full analysis: Tokenmaxxing Or Results? What Amazon’s AI Bets Reveal 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 published engineering guidance on October 27 telling software teams not to treat token use or AI adoption as performance targets. The memo comes as reporting describes a reduced role for several Nova models and Amazon’s investment in Anthropic has risen sharply in value on paper; Amazon’s third-quarter results are due October 29.
Amazon published new guidance for its software engineers on October 27, warning that token counts and other measures of AI use can become counterproductive when treated as performance targets. The memo says teams should judge work by its impact on customers, the business and colleagues—a message that lands as Amazon’s own Nova models and its investment in Anthropic show sharply different reported results.
The guidance was written by the company’s most senior engineers and published on Amazon’s news site. It updates role expectations for software engineers without prescribing a fixed set of AI rules. The memo says definitions of AI-related work could quickly become outdated, so it focuses instead on outcomes that endure. It identifies token usage, adoption of AI tools, code changes and reviews as input measures that can be useful to monitor but flawed as goals.
According to the memo, those measures currently correlate well with important outputs: teams adopting AI most heavily are delivering more, while teams that do not adopt it are falling behind. But using usage figures as targets can raise costs and add complexity without proving that the work improved. The memo asks leaders to examine both low and high adoption: why some teams use AI less, and whether heavy users are actually producing better results.
The document also calls on managers to use AI in their own production work and describe tasks where it helped and where it did not. It says engineers should record cases where current AI tools do not benefit their work, treating that as useful information rather than a performance failure. The guidance does not tell staff to stop using AI; it argues against scoring them simply for using more.
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 Scorecard
The memo addresses a management problem facing companies that are rapidly introducing AI: usage is easy to count, but value is harder to establish. A token leaderboard can reward activity even when the work adds no customer benefit. Amazon’s guidance makes the distinction explicit, linking such targets to higher costs and unnecessary system complexity.
The stakes extend beyond internal performance reviews. Amazon sells cloud computing and AI services through AWS, while also developing its own models and investing in outside providers. Its engineering guidance therefore offers a signal about how the company wants to evaluate AI deployment: not by the volume of model activity, but by whether teams deliver useful work. That is a stated management principle, not proof that Amazon has solved how to measure AI productivity.
For employees, the memo gives explicit legitimacy to reporting that a tool does not help a particular task. For customers and investors, the distinction matters because rising AI usage does not by itself establish better products, lower costs or stronger returns. Amazon’s upcoming quarterly report may provide financial context, but the source material does not establish that the memo will change reported results.
Products Worth Considering
Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, and Liquid Retina Display
Apple MacBook Pro 16-inch Laptop with M5 Pro, 18-Core CPU, 20-Core GPU, 48GB Memory, 2TB SSD, Space Black
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
As an affiliate, we earn on qualifying purchases.
From Token Leaderboards to Guidance
The memo follows reports of internal efforts to increase AI usage. In May, the Financial Times reported that some Amazon employees assigned unnecessary work to an internal agent tool, MeshClaw, to raise token counts. Those accounts were reported anonymously and are not independently confirmed here. The same reporting described a target for more than 80% of developers to use AI weekly.
At the end of May, an Amazon spokesperson confirmed to Business Insider 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 merely for the sake of using it. The October memo puts that message into formal engineering guidance.
The company’s model and investment history provides a wider lens. Business Insider reported in July that most flagship Nova models had been moved to a “keep the lights on” level of support. AWS lifecycle information lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. Separately, the source article says Amazon invested about $18 billion in Anthropic and that a June 30 filing valued its holdings at about $190 billion. Those figures are reported values, not cash proceeds or realized returns.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering memo
What the Metrics Cannot Show
The memo does not set out a single replacement formula for measuring AI’s contribution. It acknowledges that long-term outcomes are slow and difficult to measure, and says usage metrics currently correlate with outputs while warning against making them targets. How Amazon will assess individual or team performance in practice, and how it will distinguish AI’s effect from other factors, is not specified in the source material.
The Nova and Anthropic figures also require care. Reports that most flagship Nova models were moved to maintenance status do not establish that Amazon has ended all Nova development. The stated value of the Anthropic holding is an on-paper valuation; the material does not say Amazon sold the stake or realized that gain. Amazon’s third-quarter financial results had not yet been released at the time described, so their figures and any discussion of AI’s contribution remain unknown.
Quarterly Results Due October 29
Amazon is scheduled to report third-quarter results on October 29, one day after publication of the memo. Those results may add information about the company’s financial performance and AI-related investments, but no figures or management commentary are confirmed in the supplied material. The next test for the engineering guidance will be whether teams and managers apply its outcome-focused approach rather than replacing one usage leaderboard with another.
For now, the confirmed development is a change in the company’s stated expectations: engineers are not to be judged simply by how many tokens they generate or how often they use AI. Whether that approach produces measurable improvements in software, customer outcomes or costs remains to be seen.
Key Questions
What did Amazon’s October 27 memo change?
It issued revised guidance for software engineers, emphasizing customer, business and team impact rather than token use or AI adoption as performance targets.
Is Amazon telling engineers to stop using AI?
No. The memo says AI adoption is associated with stronger outputs at present, but warns against using adoption or token counts as goals. It asks teams to evaluate whether the tools improve the work.
What is tokenmaxxing?
In this account, it refers to using an AI tool for unnecessary tasks to increase token counts, sometimes to improve a position on an internal leaderboard. Earlier reports about specific employee behavior were based on anonymous accounts.
What is known about Amazon’s Nova models and Anthropic investment?
Business Insider reported that most flagship Nova models were moved to maintenance, while AWS lists several named Nova products as retired. The source material says Amazon invested about $18 billion in Anthropic and reported a June 30 paper valuation of about $190 billion for its holdings; it does not establish realized gains or the end of all Nova work.
When will Amazon report its next quarterly results?
Amazon is scheduled to report third-quarter results on October 29, 2026. The supplied material does not include those results.
Source: ThorstenMeyerAI.com
