On October 27, Amazon (NASDAQ: AMZN) posted a memo to its software engineers on its website that says AI "will be at the center of software engineering practice," while calling token usage, AI tool adoption and code-change counts "deeply flawed" ways to measure it. Amazon's post carrying the memo is written by Terk Terkowitz, VP/Distinguished Engineer, Worldwide Amazon Stores, and the memo accompanies a new version of the company's role guideline for software development engineers (SDEs).
The memo was circulated inside Amazon in September before it went up on the company's site, CNBC reported. It names "tokenmaxxing," running up AI usage for its own sake, as a cost Amazon does not want, and tells managers to judge teams by what they deliver.
If your work does not benefit from current AI tooling, that is a finding, not a failing.Amazon's memo to software engineers, published October 27, 2026
What the memo asks of engineers and managers
The memo is addressed to SDEs and software development managers (SDMs). Its instruction to engineers is plain: "If you haven't already, you should take the time to learn and apply AI into your daily work as soon as possible." Amazon chose not to define good AI practice in the guideline itself, because "any definition we write now will likely look antiquated by this time next year," and wrote the role around impact on customers, the business and teams instead.
On measurement, the memo says Amazon has sometimes used input metrics, such as token usage or AI tool adoption, and throughput metrics, such as code changes and reviews, because long-term outcomes are slow and hard to isolate. Used "as a target instead of a measure," it says, those metrics could drive up costs and complexity. It also says they "correlate well with the outputs that matter" for now, and that using them "with care and judgement" is appropriate.
Measure | Memo calls it | Its warning |
|---|---|---|
Token usage | Input metric | Tokenmaxxing is "at odds with frugality" |
AI tool adoption | Input metric | Check both tails |
Code changes, reviews | Throughput metric | "Deeply flawed" |
Lines of code | A risk | Adds complexity |
Long-term outcomes | The goal | Slow to show |
Source: Amazon's memo of October 27, 2026; SaasBuzz summary of its wording.
For managers, the memo sets a test of hands-on knowledge: a leader should have "used these tools on your production work, can describe a task where they helped and a task where they did not, and can explain that judgement to your team."
A SaasBuzz count of the memo's text finds it uses "should" three times and "recommend" three times, and never "must." It names no product, Amazon's own coding tools included, and sets no number for anyone to hit.
How Amazon says it measures today
Marc Brooker, a vice president and distinguished engineer at Amazon Web Services, told CNBC that the company looks at "feature velocity," which he described as "how many launches are we doing, how many successful changes are we making in production." An Amazon spokesperson told CNBC that the company has not capped employees' token usage. CNBC also reported that Amazon shut down an internal leaderboard in May after finding employees were tokenmaxxing to climb it.
The guideline arrived the same week Amazon reported its third-quarter results. The company presents the memo as more than an internal note: its introduction says the vision "can serve as a model for our customers."
What it means for engineering managers measuring AI adoption
The memo keeps the adoption metrics but changes how they are read. Amazon tells leaders to look at the 10th percentile of AI adoption, "finding those teams who aren't embracing AI effectively and understanding why," and the 90th, checking whether heavy users are "seeing delivery increases to match." Read that way, the heaviest users face as many questions as the lightest.
The "finding, not a failing" line turns a low reading into something a team can answer for. An engineer who documents where AI does not help is, in the memo's words, "demonstrating the judgement we are asking for." For a manager copying the approach, that means recording those explanations next to the usage numbers, so the two are judged together.
The memo also says teams and individuals "who are not adopting AI or increasing output metrics are falling behind," which sits beside its warnings about the same metrics. Corey Quinn argued in an opinion column for The Register on October 31 that managers may still end up rewarding tool activity, because it is cheap to measure, over the long-term outcomes the memo asks for. "Because if you give me a metric, I will turn it into a target," he wrote. Amazon's answer, in the memo, is judgment: "Working together, over time, we will find better metrics and set smarter goals."
