Amazon tracks how often its engineers use AI and how well they do so.

Amazon’s retail operations are closely tracking how often software engineers use AI and how it affects production, overcoming resistance from some employees.

Internal documents obtained by Business Insider show that the company’s vast retail division, known as Stores, is closely measuring its AI deployment. The team monitors how many engineers are using AI each month, how often those tools are incorporated into their daily workflows, and whether it produces meaningful results.

The initiative calls for more than 2,100 engineering teams in the retail sector to triple the rate of software code releases using techniques Amazon calls “AI-native,” while a smaller group of at least 25 teams is expected to increase production by 10x this year. Progress toward these goals is closely tracked by a team of Amazon’s senior leaders, known as the S-Team, the document said.

Generative AI is revolutionizing software development, and coding tools have become particularly powerful in recent months, led by Anthropic’s Claude Code and OpenAI’s Codex. Although software production is rapidly increasing, code quality remains relatively stable.

For technology companies, these potential productivity gains cannot be ignored. Amazon has built AI into its engineering culture and is promoting widespread use of its AI tools, but this is creating friction within the company. Last year, CEO Andy Jassy told employees they needed to implement AI or risk losing their jobs.

Amazon’s internal documents state that it “treats AI like any automation investment.” “Actively look for opportunities to apply it, measure what works, and build habits around success.”

The February document, labeled “Amazon Confidential,” was created by a team tasked with evaluating and improving AI tools used by thousands of engineers across the company’s vast retail organization. At Amazon, we encourage candid feedback from within our company to surface and address issues early.

“Goodhart’s Law”

In fact, the company remains open about the potential of AI in the workplace, especially how success should be measured.

Amazon’s documentation says the tracking is designed to measure adoption rates and AI engagement while preventing what Amazon internally calls “Goodhart’s Law,” the risk that once a metric is targeted, it’s no longer a good metric. In other words, humans naturally adapt to new metrics, so sometimes what we measure is exactly what we get.

Amazon spokesperson Montana McLachlan told Business Insider that this is an example of how the company is “leading investments in employee training and adoption of AI tools,” and will continue to test and learn what works.

”Amazon’s store engineering teams are realizing the most meaningful benefits in terms of what they can invent for customers and how quickly they can deliver it by integrating AI throughout the development lifecycle, rather than just incorporating it as an afterthought. “We also identified opportunities for improvement, which were reflected in the ambitious goals we have set for some of our store engineering teams in 2026, along with our proven approach to AI adoption,” McLachlan said in a statement.

Increased adoption

The push for AI is already spreading within the company. Amazon expects 80% of its retail engineering teams to adopt AI-native practices. The documents show that as of February, about 60% of passengers were on board.

Amazon’s use of its homegrown AI tools has also increased. AI Teammate, a Slack-integrated agent that analyzes chats, documents, and tickets to automate tasks, has grown to more than 700 active teams, according to the document.

Pippin, which transforms ideas into technical designs and documentation, has become so central that several groups, including parts of the AWS cloud division, are adopting it more broadly. Other tools, such as Kiro, an AI coding assistant, are also seeing increased adoption and usage, according to the document.

measurement

The basis of Push is a detailed measurement system. Executives track everything from weekly production deployments per engineer to AI adoption and engagement rates.

Individual AI tools are closely monitored using metrics such as monthly active users, usage across small “two-pizza teams,” and Net Promoter Score to measure employee sentiment. Amazon also tracks a metric called “value-derived events,” which measures the frequency of actions such as producing output or providing feedback.

The document provides guidance for managers to “set clear adoption and engagement goals.” “Measure both access (who has the tools) and usage (who is actually using the tools).”

Amazon spokesperson McLachlan told Business Insider that Amazon looks at a variety of data to understand how its employees are adopting new technology and what they’re interested in.

Rebellion against top-down orders

The development caused friction within Amazon’s famously decentralized engineering culture, according to the documents.

Internal feedback cited “negative perceptions of top-down, centrally managed authority” and concerns about duplicating AI efforts across teams.

Engineers also took issue with the burden of tracking progress through self-reported goals and the lack of clear success metrics and implementation guidance. Some are looking for more direction. Some people want room to experiment.

There are also practical challenges. Some employees said onboarding certain AI tools was too complex, creating a barrier to adoption. Amazon is also dealing with an increase in duplicate internal tools and data, known as AI sprawl.

Amazon responds

In response, Amazon made adjustments. As of February, leadership was planning to shift guidance to “collaborative AI practices” rather than requiring the use of specific tools. The company is also working to replace manual reporting with automated metrics and give teams more flexibility in how they deploy AI. A central learning platform to integrate best practices and feedback is under development.

“Removes friction,” the document added. “Let’s celebrate early wins and share success stories to build momentum.”

Amazon spokesperson McLachlan said the company does not “centrally mandate teams to use AI tools,” but instead gives teams the flexibility to choose what works best for them. She added that Amazon encourages active discussion and that this document reflects its culture rather than “an aversion to AI adoption.”

don’t force it

Despite its ambitious goals, Amazon’s internal approach to AI is also pragmatic.

This document focuses on speed and practicality with six AI-native engineering principles. Prioritize practical solutions over cost optimization and use AI where it adds value rather than forcing it on every problem. It also emphasizes transparency and developing systems that can be scaled across the organization.

Here is a complete list of Amazon’s AI-native engineering philosophies:

• Delivery time comes first, cost comes second: We prefer practical and effective solutions over cheap solutions. This means building now and optimizing your compute costs later.
• AI Native is not just for AI. We use the best approach to solve the problems we face. Sometimes that requires an AI, and sometimes the AI ​​is an LLM, but that’s not always the case.
• Cutting edge, not cutting edge: We are not going to catch up with AI technology. We evaluate the benefits and maintain the flexibility to switch if the benefits outweigh the costs. In some cases, the latest improvements may be overlooked.
• Not for you, but with you: We rely on the expertise of our existing team and do not become domain experts in your field. Participating in a pilot requires domain expertise and an investment of time.
• Not all settings are required. Although we strive to please our customers, we cannot accommodate everyone’s preferences. Instead, optimize for hundreds of teams instead of just a few.
• No black box: Every solution we deploy must be auditable, understandable, and traceable. Performance and cost improvements are foregone to maintain human understanding and traceability.

Still, the company is increasingly convinced that AI should be integrated into daily operations.

Engineers, known internally as “builders,” are encouraged to “experiment with different tools” and “recognize that AI can accelerate when they default to manual work,” the document said. Leaders and managers will be told to set clear guidelines and make AI tools easily accessible.

“Make AI tools part of your daily workflow, not something you pick up once in a while,” the document says.

Any tips? To contact this reporter via email, please specify the following address: ekim@businessinsider.com or on Signal, Telegram or WhatsApp 650-942-3061. Use a personal email address, non-work WiFi network, and non-work device. Here’s a guide to sharing your information securely.