Amazon Updates SDE Guidelines for AI Era: Focus on Outcomes, Not Tools

Amazon engineering leadership — including Distinguished Engineers and Senior Principal Engineers — has revised the Software Development Engineer (SDE) role guidelines to address how AI is reshaping software engineering. The memo acknowledges that AI is changing how SDEs work at a faster and more fundamental pace than previous technology waves such as cloud, microservices, or serverless, yet it argues that the core attributes of a great SDE remain unchanged.

High-performing engineers at Amazon demonstrate that neither the developer role nor its output depends on whether AI is used. The company envisions a "renaissance developer" who blends AI tools with human judgment and domain expertise to build new products. Because Amazon employs tens of thousands of SDEs and leads in agentic AI, it shares this vision for the broader industry.

Why the Guidelines Avoid Prescribing AI Tools

Historically, Amazon has avoided mandating specific technologies in SDE guidelines to reflect the diversity of its businesses and the breadth of SDE work. This principle held through cloud, microservices, and serverless. However, senior leaders debated whether AI is different: it is altering what a software engineer is, and what software engineering itself means, more rapidly and profoundly. Despite the speed of change, the leadership believes it is still early.

Every engineer must embrace the shift and continuously learn, or risk obsolescence. AI will remain at the center of software engineering practice. Yet defining "great AI engineering practice" in a way that stands the test of time is difficult; any definition written today may be outdated next year, and outdated definitions can fossilize progress. To avoid this trap, the updated guidelines focus on what does not change: impact on customers, the business, and the team. Even as inputs, tools, and techniques evolve faster than ever, the core outputs that matter stay the same.

"Everyone asks me: 'What will change in the next 5 to 10 years?' But almost no one asks: 'What will not change in the next 5 to 10 years?' Actually, the second question is the more important one." — Jeff Bezos, 2012 interview with Werner Vogels

Measuring AI Adoption: Outcomes vs. Proxy Metrics

Focusing on unchanging outcomes simplifies the guidelines but raises a practical question: how should teams use AI and how should AI usage be measured? The memo states that the best measure of individual and team performance is long-term outcomes. However, outcomes are lagging, hard to measure precisely, and easily confounded with other trends. As proxies, organizations sometimes look at input metrics (token usage, AI tool adoption rates) or throughput metrics (code changes, code review counts).

The leadership openly acknowledges the flaws in these proxies. If treated as targets rather than reference measures, they can drive negative behaviors — e.g., maximizing token spend violates the "Frugality" principle; maximizing lines of code violates "Invent and Simplify." Trust in engineers' ownership and focus on delivery is expected to prevent such distortions.

On the positive side, input metrics currently show good short-term correlation with key outcomes, and this is expected to persist for a while. "Frontier teams" that most aggressively embrace AI show significant productivity and delivery gains, while teams and individuals not adopting AI or not showing output growth are falling behind. Therefore, prudently referencing process metrics for career planning and organizational performance is reasonable.

The memo recommends paying special attention to the 10th percentile of AI adoption (identify teams failing to embrace AI and understand why) and the 90th percentile (identify high-AI-use teams and verify whether their delivery outcomes have grown proportionally). Over time, better metrics and smarter targets will emerge.

"Great creators and designers deeply understand their customers. They spend enormous energy developing that intuition. They study and understand many anecdotes and details, not just the averages from surveys." — Jeff Bezos, 2016 Letter to Shareholders

In other words, great leaders must deeply understand their team's work, cultivate intuition by studying concrete anecdotes and tracking actual outputs, not just surface metrics.

Leadership Expectations in the AI Era

If an engineer's current work does not benefit from existing AI tools, that is a "discovery," not a "failure." The engineer should document the specific reasons, share them with peers and managers, and collaborate on solutions. An engineer who can clearly explain why AI cannot help in certain areas is demonstrating the judgment Amazon values.

Technical leaders and managers are responsible for ensuring their teams extract maximum utility from AI to deliver better customer and business outcomes. This requires personal, hands-on understanding of the tools — not hearsay — and knowledge of how others are using them. "Deep understanding" means having used the tools in real production work, being able to cite specific tasks where they helped and where they fell short, and communicating that judgment to the team.

Applying AI in an organization demands balancing exploration of newer, better ways of working with doubling down on proven approaches. Judgment and prudence are critical. Leaders must own the link between their team's inputs and outputs.

Across the company, the most impactful leaders are those who double down on "Learn and Be Curious" — they invest time in learning AI themselves and create time for their teams to learn. This investment yields outsized returns. Teams that stay curious while balancing "Deliver Results" and "Insist on the Highest Standards" capture the greatest positive gains from AI.

Amazon also finds that SDEs and SDMs who are closer to the product and business are best positioned to leverage AI and agentic software development. As the company has grown, some ICs and SDMs have become less connected to customers and the business than before. Leadership wants to reverse this trend. ICs at all levels are encouraged to maintain close contact with customers and deeply understand their team's product direction; ownership scope and depth increase with level. SDMs should likewise stay close to customers and business and help their ICs do the same.

Core Principles Remain Unchanged

This is a transformative era for the industry and for the definition of a software engineer. Amazon believes the opportunities for software engineers are unprecedentedly large and expresses excitement about the future of software development. The company will continue evolving the SDE role definition as AI's impact on business and engineering practices grows, but the core principles will not change: customer obsession and a long-term focus on delivering results.

Author: Terk Terkowitz, VP Worldwide Stores, Distinguished Engineer, Amazon.

Source: https://www.aboutamazon.com/news/company-news/amazon-software-engineer-role-guideline-ai