🔍 Read the full analysis: Why Token Counts Don’t Tell The Whole Story About AI At Amazon on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Amazon’s October 27 engineering memo says token counts and AI adoption can help teams spot patterns but should not become performance targets. The guidance comes as reporting describes most flagship Nova models being wound down, while Amazon’s Anthropic investment has risen sharply in value on paper.
Amazon published new guidance for its software engineers on October 27, warning that token counts and other AI activity measures should not be used as performance targets. The memo says teams should focus on results for customers, the business and colleagues—a distinction with added significance as the company’s own Nova model program and its investment in Anthropic show different outcomes.
The guidance appears in a revised role guideline for Amazon software engineers, written by the company’s most senior engineers and published on its own news site. It was presented by Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business. The document says long-term outcomes matter, but can take time to measure, leading some teams to rely on stand-ins such as AI-tool adoption, token use, code changes and reviews.
The memo calls those measures “deeply flawed” when disconnected from results or used as targets. In its view, rewarding token consumption can conflict with Amazon’s frugality principle, while measuring lines of code can work against its aim to simplify. The memo does not say usage data is useless: it says these inputs currently “correlate well” with important outputs, while warning that turning a correlation into a goal can raise costs and complexity.
Instead, leaders are asked to examine both low and high adoption. The memo recommends looking at the 10th percentile to understand why some teams use AI less, and the 90th percentile to check whether heavy users are producing more. It also asks leaders to try the tools on their own production work and record tasks where they help and where they do not. The document says a lack of benefit from current AI tools can be a valid finding, rather than 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
From AI Activity to Business Outcomes
The guidance matters because employee scorecards can shape how technology is used. If teams are rewarded for tokens consumed or tools opened, they may produce more activity without improving products or services. Amazon’s memo argues that such targets can also encourage unnecessary spending and more complicated systems. Its alternative is not to ignore adoption, but to treat it as a signal to investigate rather than a result in itself.
The company’s wider AI portfolio offers a practical reason to make that distinction. Business Insider reported in July that most flagship Nova models had been moved to a “keep the lights on” status. Amazon Web Services’ lifecycle page lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. By contrast, Amazon’s filing valued its Anthropic holdings at about $190 billion as of June 30, compared with about $18 billion invested, according to the source material. That valuation is on paper, not a realized return, and does not by itself establish how well either AI strategy is performing for customers.
For workers, the memo suggests that using AI more is not automatically evidence of better work. For managers, it sets out a way to ask where tools improve outcomes and where they do not. For customers and investors, it reinforces that adoption figures alone cannot establish whether a company’s AI spending is delivering useful products or measurable business value.
Popular right now
Roborock Qrevo CurvX Robot Vacuum and Mop
KitchenAid Artisan Series 5-Quart Tilt-Head Stand Mixer, Dried Rose (KSM150PS)
De'Longhi Magnifica Evo Automatic Espresso Machine with Milk Frother – Silver
As an affiliate, we earn on qualifying purchases.
Earlier Usage Targets at Amazon
The memo follows reports that some Amazon employees used internal tools for unnecessary tasks to raise their token totals. The Financial Times reported in May that staff were giving an internal agent called MeshClaw extra work to inflate usage. That account, including reports of an adoption target and internal leaderboards, is based on reported employee accounts and is not confirmed in the memo.
There is a separate on-record development. At the end of May, an Amazon spokesperson told 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 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 guideline moves that message from an informal warning into official engineering role guidance. The memo says it avoids fixed AI requirements because definitions may become outdated quickly, and instead emphasizes durable measures: impact on customers, the business and the team.
“Tokenmaxxing is at odds with frugality.”
— Amazon engineering memo
What the Metrics Cannot Show
The memo does not provide company-wide data showing how much AI has improved software output, customer experience or costs. Its statement that adoption measures currently “correlate well” with relevant outputs is Amazon’s characterization; the guidance does not publish the underlying data or explain how the relationship was measured.
The reported Nova status and the valuation of Amazon’s Anthropic stake also do not offer a complete comparison of the two strategies. The source material does not specify the operating results attributable to Nova, the terms behind the reported investment valuation, or whether the paper valuation will translate into realized gains. Amazon’s scheduled third-quarter results may provide financial context, but the figures had not yet been reported at the time of this article.
Amazon’s Results Due October 29
Amazon is scheduled to report third-quarter results on October 29, 2026, the day after publication of the memo. Those results may give investors further financial context, but the source material does not establish that Amazon will break out returns from Nova, Anthropic or employee AI usage.
Within engineering, the memo calls for managers to examine adoption at both ends of the range, test AI tools on production work and document where they help or fall short. The next meaningful evidence will be whether teams use those findings to change their work—and whether Amazon reports measurable improvements in customer or business outcomes, rather than activity totals alone.
Key Questions
What did Amazon’s October 27 memo change?
It introduced revised role guidance for software engineers, saying teams should judge AI work by customer, business and team impact, not token counts or tool usage alone.
Does Amazon say engineers should stop using AI?
No. The memo says adoption metrics can correlate with important outputs, but warns against using them as targets. It asks leaders to examine both low and high adoption and determine whether tools improve results.
What does tokenmaxxing mean in this story?
It refers to using AI tools for extra or pointless tasks to increase token counts. Earlier reports described this behavior among some Amazon employees; those accounts are not confirmed by the October memo.
What does the memo say about Amazon’s AI models?
The memo is about engineering performance measures, not a formal review of Amazon’s models. Separately, Business Insider reported that most flagship Nova models were moved to a “keep the lights on” status, and AWS lists several Nova products as retired.
When will Amazon report its next results?
Amazon is scheduled to report its third-quarter results on October 29, 2026. The source material does not say whether the company will report AI-specific returns or usage figures.
Source: ThorstenMeyerAI.com
