🔍 Read the full analysis: Amazon’s AI Strategy Shifts Attention From Tokens To Outcomes on ThorstenMeyerAI.com
Get hardware and tech essentials 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 Oct. 27 telling software teams not to treat AI usage, token counts or code volume as performance goals. The memo says those measures can help track adoption but should be judged against customer, business and team outcomes; Amazon’s own AI model and Anthropic investments provide wider context, though the company’s latest quarterly results were not yet available.
Amazon published revised guidance for its software engineers on October 27, telling teams to judge AI tools by their results rather than by token counts, adoption rates or code volume. The memo says such figures can help monitor AI use, but turning them into targets may raise costs and complexity without showing whether customers, the business or teams benefit.
The guidance was written by Amazon’s most senior engineers and announced in a memo from Terk Terkowitz, a vice president and distinguished engineer in Amazon’s stores business. It revises the company’s role guidance for software engineers. Rather than prescribing fixed AI rules, the document centers the job on impact on customers, the business and the team, arguing that definitions of AI use could quickly become outdated.
The memo describes token usage, AI-tool adoption, code changes and reviews as “deeply flawed metrics” when used as targets instead of measurements. According to the memo, these indicators currently correlate well with meaningful outputs, but they are inputs, not proof of value. Setting goals around them can encourage needless usage, increase expense and make systems more complex.
Amazon’s proposed approach is to examine both low and high adoption. Leaders are asked to understand why teams near the 10th percentile are using AI less, and to check whether teams near the 90th percentile are producing better results. Engineers who find that current tools do not help their work are encouraged to document that finding. The memo also asks leaders to 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 AI Metrics
The guidance addresses a practical problem for employers adopting generative AI: usage is easy to count, but value is harder to establish. A token total can show how much text a model processed; it cannot by itself establish that a task was completed faster, that software improved or that customers received a better service. When employees are rewarded for usage alone, the metric can encourage activity that does not advance the work.
For Amazon engineers, the shift affects how AI adoption may be discussed and evaluated. The memo does not tell teams to stop using AI. Instead, it rejects the idea that high usage is automatically evidence of strong performance, while saying teams that are not adopting AI may be falling behind. That leaves managers with a more demanding task: compare adoption with actual results, and distinguish a tool that helps from one that adds work.
The distinction also matters for other companies building AI programs. Internal leaderboards can make adoption visible, but the Amazon memo warns that a score can distort behavior if it becomes the goal. It frames a documented decision not to use a tool as potentially sound judgment, rather than as a failure to meet an adoption quota.
Products Worth Considering
Apple MacBook Pro 2021 16‑inch with M1 Max, 32 GB RAM, 2 TB SSD, Space Gray (QWERTY English)
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
Ledger Nano X Bluetooth Crypto Wallet
As an affiliate, we earn on qualifying purchases.
Amazon’s Models and Anthropic Bet
The memo follows reporting earlier this year about internal AI-use tracking. The Financial Times reported in May that some Amazon staff assigned unnecessary tasks to an internal agent called MeshClaw to raise token counts. Those accounts were attributed to unnamed employees and are reported allegations, not independently confirmed details. Business Insider reported that a leaderboard called KiroRank was deprecated; Amazon said it was an informal tracker created by employees and was never intended to encourage AI use for its own sake.
Amazon’s public response at the time echoed the new guidance. Senior vice president Dave Treadwell told staff not to use AI just for the sake of using it. The October memo makes that principle part of formal engineering role guidance, while retaining adoption figures as possible indicators rather than discarding them outright.
The source article also compares the memo’s emphasis on results with Amazon’s AI investments. It cites Business Insider reporting that most flagship Nova models were moved to maintenance and Amazon’s lifecycle information showing that some Nova products had been retired. It also points to Amazon’s investment in Anthropic and a reported carrying value for that stake. Those investment figures are accounting measures, not a direct measure of customer or operating performance. Amazon was due to report third-quarter results on October 29, after the memo’s publication.
““Tokenmaxxing is at odds with frugality.””
— Amazon engineering memo
How Teams Will Be Evaluated
The memo does not spell out a single replacement score or explain exactly how managers will weigh AI use against project outcomes in performance reviews. It says outcomes can take time to measure and that usage indicators still correlate with results, leaving how teams will apply that judgment to individual roles and projects unclear.
It is also not yet clear whether Amazon will publish evidence that the new approach changes costs, delivery times, software quality or customer experience. The source material’s descriptions of Nova’s status and the value of Amazon’s Anthropic holding rely on reporting and financial disclosures; neither by itself establishes how well Amazon’s AI products perform for customers. The company’s scheduled third-quarter report may provide further financial context, but no results were available when the memo was published.
Third-Quarter Results and Follow-Through
Amazon was scheduled to report its third-quarter results on October 29, two days after publishing the engineering memo. Those figures could provide financial context for the company’s broader AI investments, but they will not, on their own, show whether the revised engineering measures have improved team or customer outcomes.
The next test of the guidance will be how Amazon’s engineering leaders apply it: whether teams track adoption without turning it into a quota, document where tools help or fail, and connect AI use to measurable work results. The memo does not announce a new companywide leaderboard or a specific timetable for reporting those effects.
Key Questions
What did Amazon announce?
Amazon published revised software-engineering guidance on October 27 that says AI tools should be judged by work outcomes, not by token consumption, adoption rates or code volume alone.
Is Amazon telling engineers to stop using AI?
No. The memo says AI adoption can correlate with useful outputs and warns that teams not adopting AI may fall behind. Its point is that usage should not be treated as proof of value.
What does “tokenmaxxing” mean in this story?
It refers to using an AI system more, or assigning it unnecessary tasks, to raise token counts. The memo argues that rewarding token totals can conflict with frugality and fail to show whether the work improved.
What remains unknown about the policy?
The memo does not set out a new numerical performance score or explain exactly how managers will apply the guidance in reviews. Amazon has not yet reported results showing whether the change improves costs, quality, speed or customer outcomes.
Source: ThorstenMeyerAI.com
