🔍 Read the full analysis: Amazon Tells Engineers To Stop Tokenmaxxing. Its Own AI Bets Show What Measuring Outcomes Really Means 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 updated software-engineering guidance on October 27 warning that token counts and other AI-usage metrics can create costly, misleading incentives when treated as targets. The memo says teams should judge AI use by its effects on customers and business results; the status of Amazon’s Nova models and the value of its Anthropic investment offer a wider, but still incomplete, company-level scorecard.
Amazon published new guidance for software engineers on October 27, warning that token counts, AI-tool adoption and other activity measures should not become performance targets. The memo says engineers should be judged by their impact on customers, the business and their teams—a message that comes as the company’s own AI efforts show different outcomes across its in-house Nova models and its investment in Anthropic.
The memo, published on Amazon’s news site, was written by Terk Terkowitz, a vice president and distinguished engineer in the company’s stores business. It introduces a revised role guideline for software engineers, written by Amazon’s most senior engineers. The document avoids a fixed definition of AI use in the job, saying that one written now could look outdated within a year.
Amazon says long-term outcomes are difficult to measure quickly, so some teams have relied on inputs such as token usage, AI-tool adoption, code changes and reviews. The memo calls these “deeply flawed metrics” when used as targets: they can be disconnected from results and can increase costs and system complexity. It specifically says “tokenmaxxing is at odds with frugality” and that measuring lines of code conflicts with Amazon’s “invent and simplify” principle.
The guidance does not say that AI tools are unhelpful or that usage data should be ignored. The memo says those input measures currently “correlate well” with outcomes that matter, and that teams adopting AI are delivering more while teams that do not adopt it are “falling behind.” Its distinction is between using activity data to identify patterns and scoring people on activity alone.
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
From AI Activity to Business Results
The guidance addresses a practical problem for employers rolling out generative AI: usage is easy to count, but value is harder to establish. A high token total can show that a tool was used, but not that it improved a product, saved time, reduced costs or helped a customer. If employees are rewarded for the count itself, they may have reason to create activity that does not improve their work.
Amazon’s instructions point toward a more diagnostic approach. Managers should examine both the 10th percentile of AI adoption, to understand why some teams use the tools less, and the 90th percentile, to test whether heavy use is producing better results. The memo also says engineers should be able to explain when current AI tools do not help. It calls that a “finding, not a failing,” making room for professional judgment rather than a single usage target.
The company’s own AI portfolio adds relevance, but the comparison is not a direct test of its new guidance. The supplied report says most flagship Nova models were moved to maintenance, while Amazon’s Anthropic holding had risen substantially in value on paper. Those developments suggest that investment and usage are not the same as demonstrated customer impact; the memo itself does not claim that one caused the other.
Products Worth Considering
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Jackery Solar Generator 1000 v2 with 200W Solar Panel
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
As an affiliate, we earn on qualifying purchases.
How Amazon’s Usage Tracking Evolved
The memo follows earlier reports of internal tracking that rewarded AI activity. In May, the Financial Times reported that some Amazon staff assigned unnecessary tasks to an internal agent tool, MeshClaw, to increase token counts. The same report said Amazon had a target for more than 80% of developers to use AI weekly and tracked token use on internal leaderboards. Those accounts were anonymous and remain reported allegations, not details confirmed by the memo.
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 a group of employees and said it was never intended to promote 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 now sets out the principle in an official engineering role guideline.
The company-level comparison comes from separate reporting and records cited in the source material. Business Insider reported in July that most flagship Nova models had been shifted to “keep the lights on.” Amazon’s lifecycle page lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. The source also says Amazon invested about $18 billion in Anthropic and reported the holding at about $190 billion as of June 30. These figures describe an investment’s stated paper value, not cash proceeds or a measure of product performance.
“Tokenmaxxing is at odds with frugality.”
— Amazon engineering memo
What the Memo Does Not Resolve
The memo does not set out a single replacement metric or specify how managers should weigh customer impact, business outcomes and team contribution in performance reviews. It acknowledges that long-term results can be slow and difficult to measure, but the document does not give a universal method for attributing a result to AI rather than other changes in a team’s work.
It is also unclear how widely the reported tokenmaxxing occurred, how many employees were involved, or whether the reported weekly-adoption target applied across Amazon or only to particular teams. The memo’s statement that adoption metrics currently correlate with outcomes is not accompanied in the supplied material by underlying data, a time period or a quantified comparison. The Nova and Anthropic figures likewise do not establish the future commercial performance of either effort.
Amazon’s Results Due October 29
Amazon is scheduled to report third-quarter results on October 29, 2026, one day after the memo’s publication. The source article says it will not speculate about those figures; the earnings report may provide more information about the company’s financial position, but it is not clear whether Amazon will detail Nova’s prospects or the business impact of its AI tools.
For now, the next test is how teams apply the revised guidance. The memo asks leaders to use AI tools on their own production work and describe one task where the tools helped and another where they did not. It also directs managers to examine adoption at both ends of the distribution. Whether those practices replace usage leaderboards with clearer measures of customer and business outcomes remains to be seen.
Key Questions
What does Amazon mean by tokenmaxxing?
Tokenmaxxing means increasing AI token usage to improve a usage score or leaderboard position, rather than because the extra use helps complete the work. The memo says token counts should not be treated as a performance target.
Is Amazon telling engineers to stop using AI?
No. The guidance does not tell staff to stop using AI. It says AI adoption can correlate with useful results, while warning that usage alone is not proof of impact. Engineers are expected to use judgment about when the tools help.
What should managers measure instead?
The memo emphasizes long-term impact on customers, the business and the team. It does not prescribe one replacement score. It tells leaders to investigate both low and high AI adoption and ask whether heavier use is actually producing better work.
What happened to Amazon’s KiroRank leaderboard?
An Amazon spokesperson confirmed to Business Insider at the end of May that KiroRank had been deprecated. The company described it as an informal tracker made by a group of employees, not an official push to use AI for its own sake.
When will Amazon report its next quarterly results?
Amazon’s third-quarter results are scheduled for October 29, 2026. The supplied source does not confirm whether the company will provide detailed updates on Nova or its use of AI tools in engineering.
Source: ThorstenMeyerAI.com
