🔍 Read the full analysis: How To Think About AI Outcomes Beyond Tokenmaxxing on ThorstenMeyerAI.com
Get the little things that make your day 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 warning that token counts and other AI-use metrics can create costs and complexity when treated as performance targets. The memo says those measures can still help teams spot patterns, but engineers should be judged on results and leaders should examine why AI helps some work and not other work.
Amazon published new guidance for its software engineers on October 27, warning that token counts and other measures of AI use should not become performance targets. The memo says engineers should be evaluated on customer, business and team impact, a policy shift that follows reports of staff using an internal AI tool for unnecessary tasks to raise their token totals.
The guidance was written by the company’s most senior engineers and introduced by Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business. It revises the role guidelines for software engineers across the company. Rather than add rules about specific AI tools, the document focuses on outcomes it says are less likely to become outdated: impact on customers, the business and colleagues.
The memo describes token usage, AI-tool adoption, code changes and reviews as input measures that some teams have used because long-term outcomes are harder to track. It calls such metrics “deeply flawed” when used as targets instead of ways to observe work. Setting them as goals, it says, can increase costs and make systems more complex. Its examples include “tokenmaxxing is at odds with frugality” and measuring lines of code conflicting with Amazon’s principle to “invent and simplify.”
The document does not say AI adoption is irrelevant. It says input measures currently “correlate well” with important outputs, and that teams adopting AI most heavily are delivering more while teams that do not adopt it are “falling behind.” Its distinction is between tracking use to understand performance and rewarding use for its own sake. The memo asks leaders to examine both the 10th and 90th percentiles of AI adoption, and to test whether heavy use is producing better results.
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 Measurable Results
The guidance matters to engineers and employers because AI-use targets can reward activity rather than value. A high token count may mean that a tool was used extensively, but it does not by itself show that customers received a better product, work was completed faster or costs fell. If teams are scored on usage, they may have reason to create more AI activity even when it adds little to the work.
Amazon’s proposed approach keeps adoption data in view but makes it diagnostic: leaders should ask why use is low or high and whether the work improved. That could make performance reviews less dependent on tool counts, while placing more weight on judgment and evidence of impact. The memo also gives employees a way to explain when current tools do not help: document the reason and share it with peers and managers.
The message has relevance beyond Amazon as companies look for ways to assess returns from AI investments. The memo offers a policy position, not evidence that the company has solved the measurement problem. It does not provide figures on savings, productivity or customer outcomes from the change.
Products Worth Considering
Dreame L50 Ultra Robot Vacuum and Mop
Bose S1 Pro+ Portable PA Speaker
Ring Floodlight Cam 2K Wired Security Camera, 3-Pack, Black
As an affiliate, we earn on qualifying purchases.
Reports That Prompted the Guidance
The memo follows reports of internal tracking and competition around AI use. In May, the Financial Times reported that Amazon employees were assigning unnecessary work to an internal agent tool, MeshClaw, to increase their token counts. That account, as summarized by The Decoder, also described a weekly AI-use target for developers and internal usage leaderboards. Those details were reported accounts, not independently confirmed in the memo.
Amazon did confirm to Business Insider at the end of May that an employee-built leaderboard called KiroRank had been deprecated. The company said the tracker was informal and “was never intended to promote the use of AI for usage’s sake.” Senior vice president Dave Treadwell also told staff: “Please don’t use AI just for the sake of using AI.” The October guidance takes that message beyond an informal tracker and places it in the company’s official engineering role guidelines.
The source article also compares Amazon’s internal policy with its broader AI portfolio. It cites a Business Insider report that most flagship Nova models were moved to “keep the lights on” and Amazon Web Services lifecycle information listing Nova Canvas, Nova Reel and the first Nova Sonic as retired. It contrasts those developments with Amazon’s investment in Anthropic. These examples frame a broader question about outcomes, but they do not establish that the memo was caused by the performance of either investment.
““Tokenmaxxing is at odds with frugality.””
— Amazon’s engineering memo
What the Memo Does Not Measure
The memo does not publish quantified results showing whether AI use has improved productivity, reduced costs or increased customer satisfaction at Amazon. It says adoption measures correlate with relevant outputs today, but the document does not define those outputs in detail or provide the data behind that assessment.
It is also not clear how individual teams will translate the guidance into performance reviews, what evidence will count as customer or business impact, or whether Amazon will change existing adoption targets. The source article says the company’s third-quarter results are due October 29, but those results will not by themselves establish the effect of this newly published engineering guidance.
How Teams Apply the New Guidance
The next test is how managers and engineering teams apply the guidelines in day-to-day work. The memo asks leaders to use AI tools on their own production tasks, identify a task where the tools helped and one where they did not, and explain the difference. It also calls on teams to examine both low and high adoption rather than assume either is a success or a problem.
Amazon’s third-quarter results are scheduled for October 29, one day after the memo’s publication. Those results may provide financial context for the company’s AI investments, but the available source material does not say they will report outcomes from this policy. Further details on how the guidelines will affect reviews, targets and team-level measurement remain to be seen.
Key Questions
What does “tokenmaxxing” mean?
Tokenmaxxing means increasing AI token use, such as by giving a tool unnecessary tasks, to improve a usage count or leaderboard position. The term is used in the memo’s warning against treating token totals as a goal.
Is Amazon telling engineers to stop using AI?
No. The memo says AI adoption measures currently correlate with important outputs and that teams with high adoption are delivering more. Its warning is against scoring people on usage alone, rather than judging whether tools improve the work.
What should managers measure instead?
The guidance points to long-term impact on customers, the business and the team. It asks managers to use adoption data to investigate differences between teams and to check whether heavy use results in better delivery.
What should an engineer do if AI tools do not help?
The memo says that is a finding, not a failing. It asks engineers to record why the tools did not help and share that explanation with peers and their manager.
Has Amazon shown that the policy improved results?
Not in the material cited here. The memo sets out guidance but does not report measured changes in productivity, costs or customer outcomes resulting from it.
Source: ThorstenMeyerAI.com
