🔍 Read the full analysis: Amazon’s AI Investments Bring The Outcomes Question Into Focus on ThorstenMeyerAI.com
Get smart everyday buys 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 saying AI usage measures such as token counts should not become performance targets; teams should focus on customer, business and team outcomes. The memo comes as reporting describes most flagship Nova models as maintained rather than actively developed, while Amazon’s Anthropic stake was valued at about $190 billion in a June 30 filing. Amazon’s third-quarter results are due October 29.
Amazon published revised guidance for software engineers on October 27, warning that token counts and other AI-use measures should not become performance targets. The memo says engineers should be judged by their impact on customers, the business and their teams—a policy shift arriving as Amazon’s own Nova models and its investment in Anthropic present sharply different reported results.
The guidance was written by Amazon’s most senior engineers and published on the company’s news site. It says that measures including AI tool adoption, token usage, code changes and code reviews can serve as indicators, but become “deeply flawed metrics” when used as targets rather than measures. The memo warns that targets based on these inputs can increase costs and system complexity without showing whether work improved.
The memo does not say AI use has no value. It says the input measures currently “correlate well” with important outputs, and describes teams that have adopted AI more extensively as delivering more, while teams that do not adopt it are “falling behind.” Its distinction is between observing usage to understand work and using usage to score employees.
It directs leaders to examine both low and high adoption: understand why teams around the 10th percentile use AI less, and check whether those around the 90th percentile are actually delivering more. Engineers whose work does not benefit from current tools should document why, the memo says. Leaders are also asked to use AI on their own production work and report where it helped and where it did not.
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 Shifts From Usage to Impact
The guidance matters because internal measures can shape how employees work, particularly when a company sets targets or builds leaderboards around them. If workers are rewarded for generating more tokens or code rather than solving customer problems, they may spend resources on activity that produces no useful result. Amazon’s warning frames this as both a cost and engineering-quality issue, linking token-heavy behavior to its stated principles of frugality and simplifying systems.
For companies adopting generative AI, the memo offers a practical distinction: usage data can help locate teams that need support, but it is not proof of productivity. Whether AI speeds up a task, improves quality or benefits customers remains the more consequential test. The memo’s recognition that current adoption measures correlate with outcomes also cautions against reading it as a retreat from AI; it calls for judgment about when the tools help, not a blanket usage quota.
The wider corporate stakes are visible in Amazon’s own AI portfolio. The source material reports that Amazon’s in-house Nova effort has seen flagship models moved largely to maintenance, while its Anthropic investment has risen sharply in value on paper. Those figures are not directly comparable measures of product performance: one concerns model development and the other a financial holding. Together, though, they sharpen the question of how Amazon decides which AI investments produce durable value.
Popular right now
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Samsung 42-Inch S90H OLED 4K Smart TV Bundle
Vitamix 5200 Professional-Grade Blender, 64 oz. Container, Black
As an affiliate, we earn on qualifying purchases.
Nova, Anthropic and Earlier Usage Targets
Before the October guidance, Amazon employees’ use of AI had already drawn attention. The Financial Times reported in May that some staff gave an internal agent tool, MeshClaw, unnecessary tasks to raise token counts on internal leaderboards. Those accounts were anonymous and are reported allegations, not independently established facts. The source also says Amazon had a weekly AI adoption target; that detail likewise comes from the reporting, rather than the October memo.
Amazon did confirm one related change. At the end of May, a spokesperson told Business Insider that the KiroRank leaderboard had been deprecated, describing it as an informal tracker created by employees and saying it was never intended to encourage AI use for its own sake. Senior vice president Dave Treadwell also told staff not to use AI just for the sake of using it.
The investment comparison is based on separate developments. Business Insider reported in July that most flagship Nova models had been moved to a “keep the lights on” posture; Amazon’s lifecycle information lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. The source says Amazon invested about $18 billion in Anthropic, with its June 30 filing valuing the holding at about $190 billion. That reported valuation is a paper value, not cash proceeds or a measure of operating profit. Amazon is scheduled to report third-quarter results on October 29.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering memo, published October 27
How AI Results Will Be Measured
The memo does not specify a single replacement score or a company-wide formula for measuring AI’s effect on productivity. It says long-term outcomes are difficult and slow to measure, and leaves room for leaders to apply judgment. It is also unclear how the guidance will affect individual reviews, team targets or future AI adoption reporting.
The source material does not establish how much of the reported Nova wind-down reflects a permanent strategy change, nor does a paper valuation of Anthropic determine the eventual return on Amazon’s investment. Amazon’s October 29 results may add financial context, but the figures and any company commentary are not available in the supplied material. Claims about past token inflation remain based on anonymous reporting; Amazon’s confirmation concerned the deprecation and informal status of KiroRank.
Amazon’s Earnings and Policy in Practice
Amazon is scheduled to release its third-quarter results on October 29, one day after publishing the memo. Those results could provide updated financial figures and management commentary, but the available source material contains no earnings numbers or forecast, so none can be stated here.
For now, the next test is how managers put the guidance into practice: whether they use adoption metrics to identify support needs, check whether heavy use produces better work, and accept documented cases where AI adds little. Further details on Nova’s product plans and the value of the Anthropic stake would also help clarify how the company evaluates its AI investments beyond usage and headline valuation.
Key Questions
What did Amazon’s new engineering guidance say?
Amazon said measures such as token use, AI adoption and code changes can be useful indicators but should not become performance targets. It says engineers should be evaluated on impact on customers, the business and their teams.
Is Amazon telling employees to stop using AI?
No. The memo says AI adoption currently correlates with important outputs and that teams that do not adopt AI may fall behind. Its warning is against using AI just to raise usage figures, rather than assessing whether it helps the work.
What happened to Amazon’s Nova models?
Business Insider reported in July that most flagship Nova models had been moved to a “keep the lights on” posture. Amazon’s lifecycle page lists Nova Canvas, Nova Reel and the first Nova Sonic as retired, according to the supplied source material. The full status and future plans for the Nova portfolio are not established here.
How much is Amazon’s Anthropic investment worth?
The supplied source says Amazon invested about $18 billion and its June 30 filing valued its Anthropic holdings at about $190 billion. That is a reported paper valuation, not cash received or a guaranteed eventual return.
When will Amazon report its next results?
Amazon’s third-quarter results are scheduled for October 29, 2026. The supplied material does not include the results or any accompanying company commentary.
Source: ThorstenMeyerAI.com
