🔍 Read the full analysis: The Measure Of AI At Amazon: What Its Bets Say About Results 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’s engineering leadership said Oct. 27 that software engineers should be judged by customer and business outcomes, not token counts or AI-tool usage. The memo’s message lands as the company’s Nova models have been scaled back, according to reporting and AWS lifecycle listings, while Amazon’s Anthropic holdings were valued at about $190 billion in a June 30 filing.
Amazon’s senior engineering leadership told software engineers on Oct. 27 to measure AI by its effects on customers, the business and teams—not by token counts or how often employees use AI tools. The guidance comes as the company’s Nova model line has been scaled back, while its investment in Anthropic was carried at about $190 billion in a June 30 filing, according to the source material. Amazon is due to report third-quarter results on Oct. 29.
The memo, published on Amazon’s news site, introduces a revised role guideline for software engineers written by the company’s most senior engineers. It says measures such as token usage, AI-tool adoption, code changes and reviews can help teams understand activity, but are flawed when used as targets. The memo warns that tying goals to those inputs can raise costs and add complexity without proving that customers or the business are better served.
The guidance does not tell engineers to stop using AI. It says input measures currently correlate with outcomes that matter, and describes teams that have adopted AI more heavily as delivering more, while teams that have not adopted it are falling behind. The proposed response is to examine both low and high adoption: understand why some teams are not finding useful applications, and check whether heavy users are producing better results.
The memo also says engineers should document cases where current AI tools do not help, treating that as a useful finding rather than a performance failure. Leaders are asked to use the tools on their own production work and explain where they helped and where they did not. The guidance avoids defining AI use as a permanent job requirement, saying such a definition could quickly become outdated.
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 memo makes a practical distinction that matters beyond Amazon: usage is not the same as value. A token count can show that a system was used, but it cannot by itself show that software improved, a customer problem was solved or a team worked more effectively. When activity becomes a target, employees may have an incentive to generate more activity rather than better outcomes.
For Amazon, the issue is tied to substantial investment and competition in generative AI. The company sells its own models through AWS and has also invested heavily in Anthropic, whose models it does not control. The source material says Amazon invested about $18 billion in Anthropic and reported the holdings at about $190 billion on June 30. Those are figures from the supplied account, not a measure of operating revenue or proof that the investment has produced equivalent cash returns.
The contrast between a scaled-back in-house model portfolio and a large paper valuation on an external investment sharpens the question for investors and customers: which AI spending is producing durable results? Amazon’s Oct. 29 earnings report may offer financial context, but the memo itself provides no quantified estimate of AI’s contribution to revenue, costs or productivity.
Products Worth Considering
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How Amazon Reframed AI Measurement
The memo follows earlier reports of internal pressure to increase AI use. The supplied source says the Financial Times reported in May that some Amazon employees assigned unnecessary jobs to the MeshClaw agent to raise token counts. It also reports that an internal target called for more than 80% of developers to use AI weekly. These details are based on reporting and anonymous employee accounts, rather than figures confirmed in the memo.
Amazon did confirm to Business Insider in late May that an employee-created leaderboard called KiroRank had been discontinued. The company said it was an informal tracker and was not intended to promote AI use for its own sake. Senior vice president Dave Treadwell also told staff not to use AI simply to use it. The Oct. 27 memo takes that principle further by placing outcome-based judgment in official engineering guidance.
On the product side, Business Insider reported in July that most flagship Nova models had been moved to a “keep the lights on” status. AWS lifecycle information cited in the supplied material lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. That evidence indicates changes to specific products; it does not establish that Amazon has abandoned all model development or its broader AWS AI services.
“Tokenmaxxing is at odds with frugality.”
— Amazon engineering memo
What the Results Do Not Show Yet
The memo provides no quantified productivity or financial results from Amazon’s AI adoption. It says adoption metrics correlate with outcomes, but does not give the underlying data, define the outcomes in detail or show whether gains outweigh the costs of tools and infrastructure. The degree to which AI contributed to the reported performance of high-adoption teams is not specified.
The Anthropic figure is a reported value for Amazon’s holdings as of June 30, not evidence that the investment has been realized in cash. The supplied material also says gains on the stake exceeded Amazon’s operating income in the second quarter, but does not include the filing details needed to independently assess that comparison. The status of every Nova model and the company’s future model-development plans are also not established by the cited product retirements.
Amazon’s third-quarter results had not yet been reported in the source material. It is therefore not possible from these details to say how the company’s AI spending affected quarterly revenue, operating costs or margins.
Amazon’s Earnings Are Next
Amazon is scheduled to report its third-quarter results on October 29, 2026, one day after the memo’s publication. The results may provide a fresh view of the company’s financial performance, but readers should distinguish reported company-wide figures from any specific measure of AI’s contribution.
Further evidence will depend on whether Amazon publishes clearer outcome measures for engineering work, and whether its teams can show that AI tools improve customer or business results without encouraging unnecessary usage. The memo directs leaders to examine both low and high adoption, but the source material does not give a schedule for that review or say whether Amazon will disclose its findings.
Key Questions
What did Amazon’s engineering memo change?
It set out revised role guidance that emphasizes impact on customers, the business and teams. It cautions against using token counts, AI adoption, code changes or reviews as targets in place of outcomes.
Is Amazon telling engineers not to use AI?
No. The memo says AI adoption can correlate with useful results and asks leaders to examine where tools help. Its warning is against using AI merely to raise usage numbers or treating usage itself as proof of value.
What happened to Amazon’s Nova models?
Business Insider reported that most flagship Nova models were moved to a “keep the lights on” status. AWS lifecycle information cited in the source lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. The supplied information does not show that Amazon has ended all Nova development.
How much was Amazon’s Anthropic investment worth?
The source material says Amazon invested about $18 billion and reported its Anthropic holdings at about $190 billion as of June 30. That is a reported valuation of the holdings, not the same as cash proceeds or operating revenue.
When will Amazon report its next results?
Amazon’s third-quarter results are scheduled for October 29, 2026. The source material does not yet provide those results or a specific breakdown of AI’s effect on performance.
Source: ThorstenMeyerAI.com
