🔍 Read the full analysis: Amazon’s AI Bets Offer Engineers A Broader Definition Of Progress on ThorstenMeyerAI.com
Get privacy and security gear 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 revised guidance for software engineers on October 27, saying AI token use and other activity measures should not become performance targets. The memo says those measures currently correlate with outcomes, but argues that customer, business and team impact—not usage volume—should define progress.
Amazon published revised guidance for its software engineers on October 27, warning managers not to judge progress by AI token use or tool adoption rather than results. The memo says activity measures can drive cost and complexity when turned into targets, while defining good engineering work around impact on customers, the business and the team.
The guidance was written by Amazon’s most senior engineers and published on the company’s news site. It was introduced by Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business. The document updates role guidelines for software engineers and avoids fixing a permanent definition of AI use in the job, saying any such definition could quickly become outdated.
The memo identifies token usage, AI tool adoption, code changes and reviews as inputs that some teams have used as stand-ins for slower-to-measure outcomes. It calls those metrics “deeply flawed” when used as targets, arguing that they are not the results that matter and may encourage higher costs and more complicated systems. It cites Amazon’s leadership principles in saying that tokenmaxxing conflicts with frugality and measuring lines of code conflicts with inventing and simplifying.
The guidance does not say those measurements are useless. It says they 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.” It asks leaders to examine both low and high adoption: understand why some teams are not using the tools effectively, and test whether the heaviest users are delivering more. Engineers whose work does not benefit from current AI tools should document why, the memo says.
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 AI Progress to Outcomes
The memo draws a line between measuring AI use and rewarding it. If managers set token counts or tool adoption as goals, engineers may have reason to generate activity that does not improve a product. Amazon’s stated alternative is to use those figures as signals for investigation, then judge work by customer, business and team impact.
That distinction matters beyond Amazon because companies are trying to assess whether AI coding tools improve engineering work. Usage is relatively easy to count; the quality and value of the resulting work can take longer to establish. Amazon’s guidance acknowledges that adoption measures can track with useful outputs at present, but cautions that correlation does not make them sound individual performance targets.
For employees, the memo leaves room to report that a tool did not help a particular task, provided they explain the finding. For managers, it calls for direct experience: leaders should use the tools in production work, identify where they helped and where they did not, and explain the difference. That approach treats engineering judgment as part of the assessment rather than equating more AI activity with better performance.
Products Worth Considering
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
Jackery Solar Generator 1000 v2 with 200W Solar Panel
As an affiliate, we earn on qualifying purchases.
From Usage Leaderboards to Guidance
The memo follows earlier reporting about internal AI-use tracking at Amazon. In May, the Financial Times reported that employees gave an internal agent tool, MeshClaw, unnecessary tasks to raise token counts. The accounts were anonymous and remain reported allegations, not confirmation from Amazon that such conduct was widespread. The report also described a target for more than 80 percent of developers to use AI weekly.
At the end of May, Amazon confirmed to Business Insider that an employee-built leaderboard called KiroRank had been “deprecated.” The company said it was an informal tracker made by a group of employees 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 memo turns that message into formal engineering guidance.
The company’s broader AI investments provide a second point of comparison, though they are not the subject of the memo. Business Insider reported in July that most flagship Nova models had been moved to “keep the lights on”; Amazon’s lifecycle page also 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, against roughly $18 billion invested, according to the source material. Those are reported figures and a paper valuation, not cash proceeds. Amazon’s third-quarter results are scheduled for October 29.
““Tokenmaxxing is at odds with frugality.””
— Amazon’s engineering memo
How Amazon Will Judge Results
The memo does not provide a replacement numerical score or specify exactly how managers should compare the quality of work across teams. It also does not say that Amazon will stop collecting usage information; instead, it frames adoption and other input measures as potentially useful diagnostic signals, not targets.
It remains unclear how the guidance will affect individual reviews, team expectations or the company’s internal AI metrics in practice. The memo’s claim that high adoption currently correlates with outcomes is Amazon’s assessment; it does not establish that AI use itself caused the better results. The reports about employees inflating token counts were based on anonymous accounts, while Amazon’s public confirmation concerned the KiroRank tracker’s removal.
The Nova and Anthropic figures also do not provide a complete measure of the company’s AI strategy. A paper valuation is not a realized return, and the cited model lifecycle information does not establish the financial performance of the Nova business. Amazon’s upcoming earnings may provide additional company-wide financial data, but the source material does not say they will report a specific return on these AI bets.
Earnings and Guidance in Practice
Amazon is due to report third-quarter results on October 29, one day after the memo’s publication. Those results will show the company’s reported financial performance for the quarter, but any direct connection between earnings and the engineering guidance or particular AI investments will need to be established from the company’s disclosures.
Inside Amazon, the next test is whether managers apply the memo’s instructions: examine why adoption is low or high, verify whether heavier use produces better work, and ask engineers to explain when AI tools do not help. The company has not announced a new leaderboard or a replacement metric. For now, the published guidance sets out the principle: measure activity to learn, not to define success.
Key Questions
What did Amazon tell its software engineers?
Amazon said that token counts, AI tool adoption, code changes and reviews should not be used as performance targets. The memo says progress should be judged by impact on customers, the business and the team.
Is Amazon telling engineers to stop using AI?
No. The memo says teams with higher AI adoption are currently delivering more and that teams not adopting AI are falling behind. Its warning is against using volume of use as a goal, not against using AI tools.
What does “tokenmaxxing” mean here?
It refers to increasing AI token use for its own sake, including the reported practice of giving an internal tool unnecessary tasks to raise usage counts. The reports about that behavior were based on anonymous accounts.
What did Amazon say about the KiroRank leaderboard?
Amazon confirmed to Business Insider that the employee-built KiroRank tracker had been deprecated. The company described it as informal and said it was not intended to encourage AI use for its own sake.
What remains unknown about the new guidance?
Amazon has not specified a replacement score or explained exactly how the guidance will shape individual performance reviews. It also remains unclear how the company will test whether greater AI use directly causes better engineering outcomes.
Source: ThorstenMeyerAI.com
