🔍 Read the full analysis: Amazon’s Message To Engineers: Focus On AI Outcomes, Not Token Counts on ThorstenMeyerAI.com
Get business pricing on tech for your team
- 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 Oct. 27, warning that token counts and other AI-use metrics can become harmful targets when they are disconnected from results. The memo says such metrics can still help teams understand adoption, but engineers should be judged on customer, business and team outcomes—not how much AI they use.
Amazon’s senior engineering leaders published revised guidance for software engineers on October 27, warning that token counts and AI-adoption measures should not become performance targets. The memo says engineers should be evaluated on impact for customers, the business and their teams, as the company seeks to curb incentives that reward AI usage without showing useful results.
The guidance, published on Amazon’s own news site, was written by the company’s most senior engineers and applies to its software engineering role. It is presented by Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business. The memo says the role definition avoids prescribing specific AI rules because those rules may quickly become outdated.
Teams have used measures such as token usage, AI-tool adoption, code changes and reviews as stand-ins for long-term outcomes, which are slower and harder to measure. The memo calls these “deeply flawed metrics” when treated as targets: they can increase costs and complexity without establishing whether the work helped customers or the business.
The guidance does not say those measures are useless. It says they currently “correlate well” with important outputs, while warning against confusing correlation with performance. Leaders are asked to examine both the 10th percentile of AI adoption and the 90th percentile—to understand why some teams use the tools little, and whether heavy users are delivering more. The memo also asks leaders to use AI in their own production work and explain where it helps and where it does 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
How Amazon Wants AI Work Judged
The guidance matters because internal targets can shape how engineers spend time and how managers judge performance. If token volume or tool adoption becomes a score, workers may have reason to use AI on tasks where it adds little value. Amazon’s memo argues that this can create extra cost and technical complexity while rewarding activity rather than impact.
For engineers, the message is that declining to use a tool on a particular task need not count as a failure. The memo says that if current AI tools do not benefit a piece of work, that is a finding to document and discuss. For managers, the focus shifts from tracking usage alone to asking whether it improved delivery, quality or customer outcomes.
The distinction also matters beyond Amazon. Companies introducing generative AI often use adoption figures and activity logs because they are easy to count, while business outcomes take longer to establish. Amazon’s position is that usage data can help identify patterns, but cannot substitute for evidence of value.
Products Worth Considering
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Usage Leaderboards to Guidance
The memo follows earlier reporting about internal pressure to use AI. In May, the Financial Times reported that Amazon employees were assigning unnecessary tasks to an internal agent tool called MeshClaw to increase token counts. That account, based on anonymous employees, is reported rather than independently confirmed by the memo. The source material also says the Financial Times reported an internal weekly AI-use target and token leaderboards.
At the end of May, Amazon confirmed to 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 not intended to encourage 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 October guidance turns that principle into part of the company’s formal engineering role expectations.
The memo’s timing comes as Amazon evaluates different approaches to AI. The source material reports that Business Insider said in July that most flagship Nova models had been moved to “keep the lights on,” and that Amazon’s lifecycle page lists some Nova products as retired. Those developments are separate from the engineering guidance; they do not establish that the memo was prompted by the status of Nova or by any one product decision.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering memo
How Teams Will Apply the Guidance
The memo sets out principles and review practices, but the source material does not specify how Amazon will measure individual performance under the revised role guideline, or whether any existing adoption targets will be formally withdrawn. It also does not provide company-wide figures showing how much the earlier metrics increased costs or complexity.
The claim that some employees inflated token counts with pointless tasks comes from earlier media reporting based on anonymous accounts. Amazon’s confirmation that KiroRank was deprecated does not independently verify every detail of those accounts. The memo’s statement that adoption metrics correlate with results is also not accompanied in the source material by data explaining the size or basis of that relationship.
Amazon’s third-quarter results were due on October 29, the day after the source article was published. Those results may provide financial context, but the source material does not report them, and no financial outcome should be inferred from the engineering memo.
Next Tests for AI Productivity
Amazon’s next step, as described in the memo, is for leaders to examine the reasons behind both low and high AI adoption. Managers are asked to test the tools on their own production work, identify tasks where they helped and where they did not, and explain those differences to their teams. Engineers are expected to document cases where current AI tools do not improve their work.
The practical test will be whether teams can connect AI use to outcomes rather than simply report activity. Amazon’s October 29 earnings release was the next scheduled company-wide milestone at the time of publication, but the source material does not say whether the company planned to address the memo or AI productivity during that report.
Key Questions
What did Amazon tell its engineers?
Amazon’s October 27 guidance says engineers should be assessed on customer, business and team impact, not on token counts or how often they use AI tools.
Is Amazon telling employees to stop using AI?
No. The memo says AI-adoption measures currently correlate with important outputs, but warns against turning those measures into targets. It asks teams to check whether heavy use is producing better results.
What does “tokenmaxxing” mean here?
It refers to increasing AI token usage for the sake of a higher usage score or leaderboard position, rather than because the extra AI activity improves the work.
Did Amazon confirm that employees inflated token counts?
The source material cites earlier Financial Times reporting based on anonymous employee accounts. Amazon separately confirmed that its informal KiroRank leaderboard was deprecated, but that confirmation does not verify every reported account of token inflation.
What remains unknown about the new guidance?
The memo does not spell out exactly how performance reviews will change, whether all usage targets have been removed, or how Amazon will measure the costs and benefits of AI use across teams.
Source: ThorstenMeyerAI.com
