Amazon’s engineering leaders have revised the role guideline for its software development engineers (SDEs). Amazon published the memo explaining it on 27 October, and its introduction says “what matters most is the outcome, not which tools are used.”
The memo, published by Terk Terkowitz, VP/Distinguished Engineer for Worldwide Amazon Stores, tells engineers to learn AI “as soon as possible”. It calls input metrics such as token usage “deeply flawed”, and adds: “If your work does not benefit from current AI tooling, that is a finding, not a failing.”
Short explainer · Amazon’s engineer guideline
A vertical news Short narrated from the published text of this piece and related FourWeekMBA pieces.
An evergreen Business Engineer explainer on why products built on the same model differ, because of the harness around it. It explains the general idea and does not discuss Amazon.
The key insight: As we read it, Amazon’s revised engineer role guideline does not write AI practices into the role; it keeps the measure on outcomes for customers, the business and teams, while the memo still treats token and throughput metrics as useful signals used with care.
Update, 27 Oct: what a new study says about judgement. The memo asks technical and organizational leaders to be able to “explain that judgement to your team.” The same day, Anthropic published a study of how AI agents shape experts’ judgment calls (paper). It recruited 61 analysts experienced in causal inference; the 50 who used the agent did a statistical analysis with Claude Code, running Claude Opus 4.8, for a fictional farming cooperative. On each decision, a coin flip set which option the agent leaned toward, in hidden instructions the analysts were not told about.
According to Anthropic, analysts went with their agent’s lean 86% of the time across the four analytical choices, where about 50% would be expected if the agent made no difference. In 38 cases where an analyst had stated a preference first, they switched 60% of the time when the agent disagreed, compared with 6% when it agreed; Anthropic says that rests on only 38 cases and the estimate is imprecise. Afterward, 77% repeated their agent’s argument in at least one answer, sometimes word for word, and only 15% credited the agent.
Anthropic lists its own caveats: it deliberately strengthened the agents’ lean, and when agents aren’t given clear preferences and instructions to voice them, “the effect might be smaller.” It studied one task, in one field, with one model. It also reports that analysts who used the agent finished faster (a median of 65 minutes, against 90) and that 3 of the 50 made serious methodological errors, against 6 of the 11 who worked without it, while noting that the no-AI group was small and less experienced.
The memo says Amazon has historically kept its SDE role guidelines free of explicit technology, and held to that through “two decades of technology trends (the cloud, microservices, serverless, etc.)”.
This time, it says, the Senior Principal and Distinguished Engineer communities “deeply debated whether this time is different”, because AI is changing software engineering “both more quickly and more fundamentally than these previous trends.”
Their answer was not to write AI practices into the role. The memo says any definition written now “will likely look antiquated by this time next year”, so the guideline focuses on what is not changing: impact on customers, the business and teams.
“The outcomes that matter aren’t changing, even if the inputs, tools, and techniques are changing faster than ever,” the memo says.
Where Amazon’s engineering-leadership memo recommends “special attention”: the 10th percentile of AI adoption across teams (teams not embracing AI effectively, and why) and the 90th percentile (heavy users, and whether delivery increases match). The axis is schematic; Amazon published no adoption figures.
Token Counts: Flawed, but Still in Use
The memo says Amazon has in some cases used input metrics “like token usage or adoption of AI tools”, or throughput metrics “like code changes and reviews”, because long-term outcomes are slow and hard to isolate.
It warns that, used as a target, these metrics could raise costs and complexity, writing that “tokenmaxxing is at odds with frugality” and “measuring LOC is at odds with invent and simplify”.
Yet it also says the same input metrics “correlate well with the outputs that matter”. It credits teams that have embraced AI the most, which Amazon calls “frontier teams”, and says teams and individuals not adopting AI “are falling behind.”
Its practical advice is to watch two ends of the curve: the 10th percentile of AI adoption, to find teams not embracing AI effectively and understand why, and the 90th percentile, to check whether heavy users are seeing delivery increases to match.
What It Asks of Leaders
For technical and organizational leaders, the memo sets a hands-on bar. “Sufficient understanding means you 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.”
It also says some of Amazon’s individual contributors and SDMs have become “less connected to their customers and business” as the company grew, and that Amazon wants to reverse this trend. It says SDEs and SDMs closer to their products and businesses take the most advantage of AI and agentic software development.
The memo quotes Jeff Bezos twice, from a 2012 interview with Werner Vogels and the 2016 letter to shareholders, and says: “We believe that the opportunity for software, and for software engineers, has never been larger.”
Context: Amazon’s July Nova pullback. This was reported in July, not today. Business Insider reported on 28 July, citing people familiar with the matter, that Amazon had begun deprecating most of its in-house flagship Nova models, including the high-end Premier and Omni models, the Reel video-generation model and the Canvas image-generation model. According to the report, resources moved toward a new frontier-model effort led by researcher Pieter Abbeel, known internally as Frontier Model Research, which is developing a new flagship foundation model expected to debut at this year’s re:Invent.
An Amazon spokesperson told Business Insider: “AI models remain one of the most important things we’re working on, and that hasn’t changed.” The report said the remaining Nova portfolio includes Nova 2 Sonic, Nova 2 Lite, Nova Forge and Nova Act, and that the new model could emerge under the Nova brand. Days earlier, GeekWire reported that Amazon was closing its San Francisco AGI site as part of layoffs in its AGI organization; citing The Information, it put the AGI Lab, founded in December 2024 around hires from the startup Adept, at about 80 people at its peak. Futurism later described the Nova models as “a bust”; that is its characterisation, not Amazon’s.
Today’s memo does not mention Nova or Amazon’s own models. The July reporting is about which models Amazon builds; the memo is about how its software engineers use AI tools in their work.
The Structural Read
As we read it, the memo separates two things: the role is judged on outcomes, while AI adoption is tracked through input metrics that Amazon itself calls flawed but says correlate well with results.
As we read it, the cost line is explicit: the memo names token usage as an input metric and says that pushing it as a target, which it calls tokenmaxxing, is at odds with Amazon’s frugality principle.
As we read it, the burden moves to leaders: the memo asks them to use the tools on production work themselves and to study the 10th and 90th percentiles of AI adoption.
Amazon engineering leadership, memo to SDEs and SDMs, published 27 October 2026
“If your work does not benefit from current AI tooling, that is a finding, not a failing”
Three Implications
FOR ENGINEERING LEADERS The memo says sufficient understanding means having used AI tools on production work and being able to name a task where they helped and one where they did not.
FOR AI BUDGET OWNERS The memo calls token usage an input metric and says tokenmaxxing is at odds with frugality, so spend is a signal, not the goal, in Amazon’s framing.
FOR ENGINEERS The memo says teams and individuals not adopting AI or increasing output metrics are falling behind, and tells engineers to learn AI as soon as possible.
The framework puts it this way: “Without it, AI works for you. With it, AI works for your team.”
As we read it, Amazon’s memo pushes in the same direction: it asks engineers to document where AI does not help and share it with peers, and asks leaders to work with the tools hands-on and understand how others are using them. The framework was written before this memo and does not discuss Amazon.
What Is Not Established
The memo gives no figures: no AI adoption rates, no productivity numbers, no token spend and nothing on engineering headcount. It does not publish the guideline itself, only the memo about it.
It says using input metrics “with care and judgement” is appropriate as engineers think about their careers and their organization’s performance, but it does not say how the guideline feeds into reviews or pay.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Amazon’s revised engineer role guideline is judged on outcomes for customers, the business and teams; its memo calls token and throughput metrics flawed, yet says they still correlate with results and recommends watching the 10th and 90th percentiles of AI adoption.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. All quotations are from the memo Amazon published on aboutamazon.com on 27 October 2026 under Terk Terkowitz’s name, written by Amazon engineering leadership for its software engineers and managers. Amazon did not publish the revised role guideline itself or any adoption or productivity figures. This is not investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
