🔍 Read the full analysis: What Amazon’s AI Investments Say About Measuring Results on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Amazon told software engineers to judge AI by its impact, not by token counts or tool usage, after some internal metrics encouraged unnecessary activity. The memo arrives as Amazon’s own Nova model portfolio has been scaled back and its Anthropic investment has risen sharply on paper; those developments do not establish a direct comparison of AI performance or returns.
Amazon published new guidance for its software engineers on October 27, warning that token counts and AI-tool adoption are poor performance targets when they are not tied to useful results. The memo, written by senior engineering leaders, follows reports of employees using an internal AI tool to inflate usage and comes as Amazon’s own Nova models have been scaled back while its investment in Anthropic has risen substantially in reported accounting value.
The memo, published on Amazon’s news site, introduces a revised role guideline for software engineers. It says engineers should be assessed on their impact on customers, the business and their teams, rather than on how much AI they use. The document calls token usage, AI adoption, code changes and reviews “deeply flawed metrics” when treated as targets. It warns that such targets can increase costs and make systems more complex.
The guidance does not say usage data is worthless or that engineers should stop using AI. It says those measures currently “correlate well” with outcomes that matter, but are stand-ins for results, which take longer to assess. Amazon advises managers to examine both low and high adopters: understand why some teams are not benefiting from AI and check whether heavy users are delivering better work. The memo says a finding that current AI tools do not help a particular task is not a performance failure.
The document was written by senior engineers and introduced by Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business. It also asks leaders to use AI on their own production work and explain where the tools helped and where they did not. The authors say they avoided fixed AI requirements in the role definition because the technology changes quickly.
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 Scorecards Can Mislead
The memo is a practical warning for companies adopting generative AI: measuring activity is not the same as measuring value. Token counts or weekly adoption rates can show whether tools are being used, but they do not establish that the tools have improved customer experience, software quality, delivery time or costs. If workers are rewarded for usage alone, they may generate activity that does not serve a business need.
Amazon’s own portfolio gives the issue added relevance, though it is not a controlled comparison. Business Insider reported in July that most flagship Nova models had been moved to “keep the lights on.” Amazon’s lifecycle information also lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. Meanwhile, the source material reports that Amazon’s investment in Anthropic was carried at about $190 billion on June 30, after roughly $18 billion in investment. That is an accounting value, not cash proceeds or proof that Anthropic’s products will produce a particular return for Amazon.
The difference matters to employees and investors alike. For engineers, the memo signals that judgment about whether a tool is useful is part of the job. For businesses, it suggests that AI programs should be judged against outcomes and costs, not leaderboards. Amazon’s third-quarter results, due October 29, may offer more company-wide financial context, but they will not by themselves isolate the impact of any one AI investment.
Products Worth Considering
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Samsung Galaxy Tab S10 FE 128GB Wi-Fi Android Tablet, Blue
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Usage Leaderboards to Guidance
The memo follows months of debate about internal AI tracking. In May, the Financial Times reported that Amazon employees had given an internal agent tool called MeshClaw unnecessary tasks to increase their token counts. The accounts were anonymous and remain reported allegations, not independently established facts. The same reporting described a target for more than 80% of developers to use AI weekly; that figure should be understood as part of the report, not as a confirmed company-wide policy.
Amazon did confirm one related development. At the end of May, a company spokesperson told Business Insider that KiroRank, an informal employee-built leaderboard, had been deprecated. The company 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 memo goes further by setting out a company-wide engineering principle: focus on outcomes, while using adoption data to identify where tools do or do not help.
Amazon’s AI strategy includes both models it develops and investments in other companies. Its Nova models are offered through AWS, while Anthropic develops models independently. The reported changes to Nova and the rise in the stated value of Amazon’s Anthropic holdings are separate developments; they do not show that one caused the other or that the memo was written in response to the investment’s performance.
“Tokenmaxxing is at odds with frugality.”
— Amazon’s engineering memo
What the Metrics Cannot Yet Show
The memo does not provide quantified evidence showing how much AI has improved Amazon’s engineering output, or how its teams will measure customer and business impact in practice. It says usage indicators correlate with important outputs today, but does not publish the underlying data or define a single replacement scorecard. The guidance instead leaves room for managers to apply judgment and examine individual cases.
The reported Anthropic valuation is also not a realized return. The source material describes Amazon’s holdings as carried at about $190 billion in a June 30 filing, but it does not establish what portion of that value comes from the investment, how it may change, or how much Amazon has earned in cash. Nor do the reported Nova retirements reveal the full financial results of Amazon’s model-development effort. Amazon’s October 29 earnings could add detail on spending and performance, but the company may not break out results for particular AI products or investments.
Amazon’s Earnings and AI Measures
Amazon is scheduled to report third-quarter results on October 29, one day after publishing the memo. Those results will provide a new view of the company’s overall business and may include comments on AI-related investment, but specific disclosures on Nova, Anthropic or engineering productivity are not confirmed in the supplied material.
The more immediate test of the memo will be how managers apply it. Amazon says leaders should review both low and high levels of AI adoption, use the tools themselves and record where they help. Whether that approach leads to published measures of customer impact, software quality, costs or delivery times remains to be seen. For now, the company’s stated standard is clear: AI use is an input; useful results are the measure.
Key Questions
What did Amazon announce on October 27?
Amazon published engineering guidance saying software engineers should be assessed on their impact on customers, the business and their teams, not on token counts or AI-tool use alone.
Is Amazon telling engineers to stop using AI?
No. The memo says adoption and usage measures can correlate with useful outcomes, but warns against turning them into targets. It asks teams to judge whether AI actually helps their work.
What was tokenmaxxing?
It refers to using an AI system more, including for unnecessary tasks, to raise token counts. The Financial Times reported that Amazon employees had done this with an internal tool; those accounts were anonymous.
What does Amazon’s Anthropic investment value mean?
The source material says Amazon’s holdings were carried at about $190 billion on June 30, after roughly $18 billion in investment. That figure is a reported accounting value, not cash proceeds or a guaranteed future return.
When will Amazon report its next quarterly results?
Amazon is scheduled to report third-quarter results on October 29, 2026. It is not yet clear whether the company will provide detailed figures for individual AI products or investments.
Source: ThorstenMeyerAI.com
