🔍 Read the full analysis: What Amazon’s AI Strategy Suggests Engineers Should Measure on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Amazon published revised guidance for software engineers on October 27, saying AI usage metrics such as token counts should not become performance targets. The memo says those measures can help teams spot patterns, but engineers and leaders should judge AI by its effects on customers, the business and team results.
Amazon’s engineering leadership told software engineers on October 27 not to treat token counts or other measures of AI activity as goals, saying performance should be judged by impact on customers, the business and teams. The guidance comes after reports of employees using an internal AI tool for unnecessary tasks to increase their usage rankings, and as Amazon’s own investments in AI produce sharply different outcomes.
The memo, published on Amazon’s news site, accompanies revised role guidance for software engineers written by the company’s most senior engineers. It says measures such as AI tool adoption, tokens used, code changes and reviews can serve as stand-ins when longer-term outcomes are difficult to assess. But the memo calls them “deeply flawed metrics” when used as targets, warning that they can increase costs and make systems more complex without showing whether work helped customers.
The guidance does not say that AI activity data is useless. It says those input measures currently “correlate well” with outcomes that matter, and that teams adopting AI are delivering more while teams not using it are “falling behind.” Its distinction is between using metrics to investigate performance and scoring people on usage. The memo invokes Amazon’s leadership principles: “tokenmaxxing is at odds with frugality” and measuring lines of code is at odds with “invent and simplify.”
It advises managers to examine both low and high adoption. Teams near the 10th percentile can help explain where tools are not working or why adoption is low; teams near the 90th percentile can show whether heavy use is producing better results. Engineers are asked to document cases where AI does not help, while leaders are told to use the tools on production work and compare tasks where they did and did not help.
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 AI Activity Is a Weak Scorecard
The guidance addresses a problem facing companies adopting generative AI: usage is straightforward to count, but the value of that usage is harder to establish. Token totals measure model activity, not customer impact. If managers reward high counts, staff may have reason to generate more activity even when it adds little value or raises costs. Amazon’s memo frames the risk in its own terms: activity measures used as targets can work against frugality and simplicity.
For engineers, the memo sets a different expectation from a simple adoption quota. It says an employee who finds that current AI tools do not benefit their work is “demonstrating the judgement we are asking for,” provided they can explain the finding. For managers, the instruction is to treat usage data as a prompt for investigation, then assess the actual work and its consequences. Whether the approach changes evaluations or team targets is not specified in the published material.
The memo’s relevance extends beyond Amazon because employers are experimenting with AI adoption dashboards and developer productivity measures. The company is not rejecting measurement; it is drawing a line between monitoring inputs and judging outcomes. That distinction matters when AI tools can add cost or code without necessarily improving a product or service.
Popular right now
Dreame L40 Ultra Gen 2 Robot Vacuum and Mop
Philips 5500 Series Fully Automatic Espresso Machine with LatteGo
Ninja Foodi 10-Quart DualZone 6-in-1 Air Fryer, Black, DZ302
As an affiliate, we earn on qualifying purchases.
Reports Behind the New Guidance
The memo follows earlier reports about AI usage tracking at Amazon. In May, the Financial Times reported that some employees gave an internal agent called MeshClaw unnecessary tasks to raise their token counts. That account, including reporting about weekly AI-use targets and internal leaderboards, is attributed reporting based partly on anonymous employee accounts; it is not independently established by the memo.
A separate development is on the record: at the end of May, an Amazon spokesperson told Business Insider that a leaderboard called KiroRank had been “deprecated.” The company described it as an informal tracker created by a group of employees and said it was never 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 source material also compares Amazon’s AI business bets. Business Insider reported in July that most flagship Nova models had been moved to “keep the lights on”; Amazon’s lifecycle page lists Nova Canvas, Nova Reel and the first Nova Sonic as retired. Separately, the source says Amazon invested about $18 billion in Anthropic and valued its holdings at about $190 billion in a June 30 filing. Those figures describe reported investment and an accounting valuation, not cash proceeds or a guaranteed return. Amazon’s third-quarter results are due October 29, after the memo.
“Tokenmaxxing is at odds with frugality.”
— Amazon engineering memo, published October 27
How the Guidance Will Be Applied
The memo sets out principles and practical instructions, but the published material does not describe how Amazon will incorporate them into formal performance reviews, whether existing usage targets will be removed across every team, or how managers will compare outcomes across different kinds of engineering work. Amazon also has not provided a company-wide accounting of the costs or productivity effects of its AI tools in the material cited here.
Reports of employees inflating usage came from earlier media accounts and include anonymous sources; Amazon’s memo does not confirm every detail of those accounts. The source material also provides no results from the third quarter, which Amazon is scheduled to report October 29. Any comparison between Nova and Anthropic should be read with care: the cited figures mix model status and a reported investment valuation, and do not establish a complete measure of either business’s performance.
Amazon’s Results and Follow-Through
Amazon is scheduled to report its third-quarter results on October 29. Those figures may offer investors and employees a further view of the company’s spending and performance, but they will not by themselves show whether the new engineering guidance has changed AI productivity or evaluation practices. The memo’s impact will depend on how teams turn its advice into day-to-day decisions.
Managers are being asked to test the tools on their own production work, examine both low- and high-adoption teams, and record where AI helps or does not. The next useful evidence would include how Amazon evaluates those findings, whether usage-based targets are replaced, and whether measured customer or business outcomes improve without unnecessary cost or complexity.
Key Questions
What did Amazon tell its software engineers?
Amazon’s October 27 memo said engineers should not use token counts, AI adoption, code changes or reviews as performance targets. It says those measures may help teams investigate activity, but results for customers, the business and teams matter more.
Is Amazon telling employees to stop using AI?
No. The memo says teams adopting AI are delivering more and that teams not adopting it are falling behind. Its warning is against treating the volume of AI use as a goal in itself.
What does “tokenmaxxing” mean in this report?
It refers to using an AI tool in ways that generate more tokens, the text units models process, to raise a usage ranking. Earlier media reports described employees assigning unnecessary tasks to an internal tool; those accounts are attributed reports, not a full confirmation in Amazon’s memo.
What should engineers measure instead?
The memo points to longer-term impact on customers, the business and teams. It also asks engineers to record where AI helps and where it does not, so managers can assess outcomes rather than infer value from usage totals alone.
Has Amazon said how the guidance will affect performance reviews?
The published memo does not explain how the guidance will change formal reviews or whether all existing usage targets will be removed. How each team applies the advice remains unclear.
Source: ThorstenMeyerAI.com
