🔍 Read the full analysis: Measuring AI Beyond Tokens: The Signal In Amazon’s Investment Bets 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 other AI-usage measures can be useful signals but should not become targets for judging engineers. The guidance lands as the company’s Nova models have been scaled back, according to reporting and AWS lifecycle notices, while Amazon’s Anthropic stake was valued at about $190 billion in a June filing. The memo does not establish that the investment has generated operational returns.
Amazon told software engineers on October 27 to judge AI work by its effects on customers, the business and teams—not by token counts or how often employees use AI tools. The memo’s timing draws attention to two different company bets: Amazon’s Nova models have been scaled back, according to reporting and AWS notices, while its investment in Anthropic was carried at about $190 billion in a June 30 filing, a paper valuation that does not by itself show business returns.
The guidance was published on Amazon’s website and comes from Terk Terkowitz, a vice president and distinguished engineer in the company’s stores business. It accompanies a revised role guideline for software engineers, written by senior engineers. Rather than define a fixed set of AI duties, the document says technology is changing too quickly for a lasting definition: “Any definition we write now will likely look antiquated by this time next year.”
The memo calls token usage, AI-tool adoption, code changes and reviews “deeply flawed metrics” when used as targets. It argues that such measures can be disconnected from useful outcomes and, when pursued for their own sake, increase costs and complexity. The memo does not say those measures have no value: it says they currently “correlate well” with results that matter, while cautioning against treating a correlation as a performance score.
Its instructions are to examine both low and high levels of AI adoption, ask why teams are using or avoiding the tools, and check whether heavy use produces better work. Engineers are also encouraged to document cases where current AI tools do not help. The memo says leaders should use the tools on their own production work and explain where they help and where they do 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
Why Amazon Is Changing the Scorecard
The memo addresses a practical workplace problem: usage is easy to count, while quality, customer benefit and long-term productivity are harder to measure. If employees are rewarded for producing tokens, lines of code or AI-assisted edits, they may increase activity without improving the work. Amazon’s guidance makes clear that the company wants managers to distinguish between using a tool and getting a useful result from it.
That distinction matters beyond Amazon because companies are still trying to determine whether generative AI improves engineering output enough to justify its costs. The memo offers a management approach, not proof that AI is delivering a particular productivity gain. Its advice is to use adoption measures as diagnostic signals—to find teams that may need help or investigation—rather than as goals that employees must hit.
Amazon’s own model and investment choices provide a separate, limited illustration of the difference between activity and outcomes. A large investment valuation is not equivalent to revenue or operating profit, and a model being maintained is not evidence that it has achieved broad commercial success. The available facts point to contrasting bets, but do not establish which will produce the stronger long-term business results.
Products Worth Considering
Logitech MX Brio Ultra HD 4K Webcam, Graphite
EcoFlow Smart Home Panel 2 Whole-Home Backup System Hub
DJI Osmo Pocket 3 Creator Combo Vlogging Camera with Mic 2, 1 CMOS, 4K/120fps, 3-Axis Stabilization, 128GB Card
As an affiliate, we earn on qualifying purchases.
Nova, Anthropic and Earlier Leaderboards
Amazon’s engineering guidance follows earlier reported efforts to track employee AI use. In May, the Financial Times reported that some staff assigned unnecessary tasks to an internal agent tool, MeshClaw, to raise token counts. The accounts were attributed to employees and remain reported claims, rather than details confirmed in the October memo. The same reporting described a weekly AI-use target and internal usage leaderboards.
Amazon did confirm to Business Insider in May that a leaderboard called KiroRank had been “deprecated.” The company said it was an informal tracker made by a group of employees and was not 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 new memo puts a similar principle into official engineering role guidance.
Amazon’s model portfolio is also changing. Business Insider reported in July that most flagship Nova models had been moved to a “keep the lights on” status. AWS lifecycle information lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. Separately, Amazon’s June 30 filing valued its Anthropic holdings at about $190 billion; the source material says Amazon had invested about $18 billion. That valuation is a reported accounting figure, not cash proceeds or a measure of Anthropic’s operating performance. Amazon was due to report third-quarter results on October 29, 2026.
“Tokenmaxxing is at odds with frugality.”
— Amazon engineering memo, published October 27
What the Measures Cannot Show
The memo does not provide a quantified estimate of how much AI has improved engineering productivity, nor does it specify a single replacement metric for managers to use. It says usage measures correlate with outcomes at present, but the source material does not give the underlying data, measurement period or size of that relationship. The degree to which AI adoption has affected delivery speed, quality or customer results therefore remains unclear.
The financial and product comparisons also have limits. The $190 billion Anthropic holding value is an accounting valuation in Amazon’s filing, not a realized gain, and the supplied material does not detail how it was calculated. The reporting about Nova’s reduced status and the AWS retirement notices concern particular models; they do not establish that Amazon has abandoned all in-house AI development. Amazon’s third-quarter results, scheduled for October 29, may provide more financial context, but the information here does not include those results.
Amazon’s Results and New Guidance
The next dated milestone is Amazon’s scheduled third-quarter earnings report on October 29, 2026. That release may show how the company describes its AI-related spending and business performance, although quarterly results would not by themselves settle questions about long-term productivity or the value of individual model bets.
Within engineering, the memo points managers toward reviewing low- and high-adoption teams, examining where tools help or fail, and sharing those findings. The company has not announced a new numerical scorecard in the source material. The practical test of the policy will be whether leaders use AI metrics to investigate work and improve outcomes rather than to reward usage volume.
Key Questions
What did Amazon’s engineering memo say about AI tokens?
It said token use and other activity measures can be useful signals, but are deeply flawed as targets because they may reward activity without showing customer or business impact.
Is Amazon telling engineers to stop using AI?
No. The memo does not call for stopping AI use. It says teams should examine where the tools help and avoid measuring success by adoption or token volume alone.
What is known about Amazon’s Anthropic investment?
Amazon’s June 30 filing valued its Anthropic holdings at about $190 billion, according to the source material. That is a paper valuation, not proof of realized profit or operating returns.
Has Amazon abandoned its Nova models?
The source material says Business Insider reported that most flagship Nova models were moved to a “keep the lights on” status, and AWS lists several models as retired. Those details do not establish that Amazon has ended all in-house AI development.
When is Amazon’s next earnings report?
Amazon was scheduled to report its third-quarter results on October 29, 2026, two days after the engineering memo was published.
Source: ThorstenMeyerAI.com
