🔍 Read the full analysis: Amazon’s Advice On Tokenmaxxing Connects To Its Bigger AI Bets on ThorstenMeyerAI.com
TL;DR
Amazon published engineering guidance on October 27 telling software teams not to treat AI usage, token counts or code volume as goals. The memo links AI work to customer and business outcomes, while the company’s Nova models and Anthropic investment show different paths in its broader AI strategy.
Amazon’s senior engineering leaders published guidance on October 27 telling software engineers not to use AI token counts or adoption rates as performance targets, saying those measures can encourage waste without showing whether customers or the business benefit. The memo’s advice comes as Amazon’s own AI strategy faces different tests: the company has scaled back several Nova models, while its investment in Anthropic has risen substantially in value on paper.
The guidance, published on Amazon’s news site, introduces a revised role guideline for software engineers. It was written by the company’s most senior engineers and presented by Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business. Rather than make AI use a fixed part of the job description, the document says definitions written now could quickly become outdated. It centers the role on impact for customers, the business and the team.
The memo describes token usage, AI-tool adoption, code changes and reviews as “deeply flawed metrics” when used as targets. Those figures can still serve as indicators: the document says they currently correlate well with meaningful outputs. But when teams are rewarded for the inputs themselves, the measures may increase costs and complexity. It gives two examples: “tokenmaxxing is at odds with frugality” and measuring lines of code is at odds with “invent and simplify.”
Its instructions are to examine both light and heavy AI users, ask why their work differs, and have leaders use the tools on their own production work. Engineers whose tasks do not benefit from current AI tools are asked to explain what they found and share it. The memo calls that outcome “a finding, not a failing”—not a reason to use AI for its own sake.
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 AI Bets Face an Outcome Test
The guidance matters beyond how Amazon evaluates engineers because it sets out a standard for judging the company’s own AI investments: usage is not the same as value. A tool can generate more tokens or code while adding expense, maintenance work or complexity. The memo’s emphasis on impact puts the question on whether AI improves a product, a customer experience or a business process.
That distinction is relevant as Amazon makes substantial investments in both its own models and outside providers. The source reports that Amazon’s Nova line has been pared back, while its Anthropic holding has gained sharply in reported value. Those developments are not directly evaluated in the engineering memo, and they do not by themselves establish which strategy will succeed. They do, however, show why measuring outcomes rather than activity is an important question for Amazon’s AI spending and product plans.
For employees, the memo also offers a practical signal: managers should assess what work accomplishes, not how many AI tools a person uses. For other companies adopting generative AI, Amazon’s position is a warning against turning convenient activity data into a proxy for productivity. The memo does not say that AI tools are unhelpful; it says that the value must be judged in the work.
Popular right now
LG C2 Series 65-Inch OLED evo 4K Smart TV (OLED65C2PUA, 2022)
Samsung 48-Inch S90H OLED 4K Smart TV (2026)
Samsung 42-Inch S90H OLED 4K Smart TV Bundle
As an affiliate, we earn on qualifying purchases.
From Internal Leaderboards to Official Guidance
The memo follows reports of internal efforts to track AI use. In May, the Financial Times reported that some Amazon employees were assigning unnecessary jobs to an internal agent tool called MeshClaw to increase their token counts. The report also described a weekly AI-use target and internal leaderboards. These accounts were attributed to people familiar with the practice; they are reported claims, not details confirmed in the memo.
One part of the earlier episode was confirmed publicly. At the end of May, Amazon told Business Insider that an informal employee-created leaderboard called KiroRank had been “deprecated.” The company said the tracker 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 guidance puts a similar principle into the company’s published engineering role expectations.
The broader company strategy has had mixed developments. Business Insider reported in July that most flagship Nova models had been moved to a “keep the lights on” status. Amazon’s lifecycle information, as described in the source, lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. Separately, the source says Amazon invested about $18 billion in Anthropic and reported the holding at about $190 billion in a June 30 filing. Those are figures attributed to the source’s cited reporting and filing; they are not a measure of realized cash returns.
„“Tokenmaxxing is at odds with frugality.”“
— Amazon engineering memo
What the Memo Does Not Settle
The memo does not provide a new single metric for measuring AI’s contribution, or specify how managers should weigh AI use against other factors in performance reviews. It says input measures can correlate with outcomes today, but does not quantify that relationship or describe how it varies among teams. How Amazon will apply the guidance in individual evaluations remains unclear.
The reported Nova changes and Anthropic valuation also do not establish which products will produce durable returns. A paper valuation is not the same as cash realized, and the source does not supply a full comparison of the costs, revenue or customer results of Amazon’s different AI bets. The company’s third-quarter results may add financial detail, but it is not yet clear whether Amazon will report figures that isolate AI’s effect.
Reports of employees inflating token counts came from media accounts and anonymous sources. Amazon’s confirmation that KiroRank was deprecated does not independently verify every detail of those reports. The October memo confirms the company’s position on metrics, but it does not disclose how widespread tokenmaxxing was or whether any employees were formally evaluated on token totals.
Amazon’s Earnings Are the Next Checkpoint
Amazon is scheduled to report third-quarter results on October 29, one day after the memo’s publication. Those results could provide updates on the company’s broader business and AI investments, though the source does not indicate that Amazon will break out returns from Nova, Anthropic or internal AI tools separately.
For engineers and managers, the immediate next step is applying the new role guidance: examine how AI helps or fails on real work, document the reasons, and look at both lower- and higher-adoption teams. Whether that approach changes performance reviews or how Amazon tracks AI use over time has not been specified. The company’s stated test, for now, is impact rather than volume.
Key Questions
What did Amazon tell its engineers about AI use?
Amazon’s October 27 memo says engineers should not be judged by token counts, AI adoption rates, lines of code or similar input measures used as targets. It says the focus should be on impact for customers, the business and the team.
Does Amazon’s memo tell employees to stop using AI?
No. The memo says some adoption measures currently correlate with important outputs and that teams not adopting AI may be falling behind. Its distinction is between using tools where they help and using them only to raise usage numbers.
What is tokenmaxxing?
Tokenmaxxing refers to increasing the amount of text an AI model processes or generates to boost a usage count or ranking. The source says the October memo criticized that practice as inconsistent with Amazon’s principle of frugality.
What did Amazon say about its KiroRank leaderboard?
Amazon told Business Insider in May that the informal, employee-created KiroRank leaderboard had been deprecated and was not intended to promote AI use for its own sake. Reports of other token-inflating behavior were not fully confirmed by that statement.
What is the next major date for Amazon’s AI strategy?
Amazon is scheduled to report third-quarter results on October 29, 2026. The available source does not say whether the company will provide separate financial results for its AI models or Anthropic investment.
Source: ThorstenMeyerAI.com
