Amazon’s retail business is closely tracking how often software engineers use AI and how that influences output, all while navigating resistance from parts of its workforce.
Loading audio narration…
An internal document obtained by Business Insider shows the company’s vast retail division, known as “Stores,” measuring the AI rollout in granular detail. Teams monitor how many engineers use AI each month, how frequently those tools are embedded in day-to-day workflows, and whether this produces meaningful results.
The effort calls for more than 2,100 engineering teams in the retail arm to triple software code release velocity using what Amazon calls “AI-native” practices, while a smaller group of at least 25 teams is expected to boost output tenfold this year. Progress against these goals is closely tracked by Amazon’s senior leadership team, known as the S-Team, according to the document.
Generative AI is revolutionizing software development, with coding tools becoming especially powerful in recent months, led by Anthropic’s Claude Code and OpenAI’s Codex. Software production has skyrocketed, while code quality has held relatively steady.
For tech companies, potential productivity gains like this are impossible to ignore. Amazon is embedding AI into its engineering culture, pushing for broader use of its own AI tools, which has created internal friction. Last year, CEO Andy Jassy urged employees to adopt AI or risk losing their jobs.
“Treat AI like any automation investment,” the internal Amazon document stated. “Actively look for opportunities to apply it, measure what works, and build habits around the wins.”
The February document, labeled “Amazon confidential,” was prepared by a team tasked with assessing and improving AI tools used by thousands of engineers across the company’s sprawling retail organization. Amazon encourages candid internal feedback to surface and address issues early.
“Goodhart’s Law”
Mark Felix / AFP via Getty Images
Indeed, the company is keeping an open mind about the promise of AI in the workplace, and especially how success should be measured.
The Amazon document states that tracking is designed to measure deployment rates and AI engagement while guarding against what Amazon internally calls “Goodhart’s Law,” the risk that once a measure becomes a target, it stops being a good measure. In other words, humans naturally adapt to new metrics, so sometimes you just get what you measure.
Montana MacLachlan, an Amazon spokesperson, told Business Insider this is an example of how the company is “leading in investing in employee training and adoption of AI tools” and that it will continue to test and learn what works.
“Amazon’s Stores engineering teams found that integrating AI across the full development lifecycle — not just bolting it on as an afterthought — delivers the most meaningful gains in what we’re able to invent for customers and how quickly we can deliver it,” MacLachlan said in a statement. “We’ve also identified opportunities for improvement, and those results, along with our proven approach to AI adoption, informed the ambitious goals we’ve set for some Stores engineering teams in 2026.”
Rising adoption
The AI push is already widespread at the company. Amazon expects 80% of its retail engineering teams to adopt AI-native practices; as of February, about 60% had got on board, according to the document.
Use of Amazon’s homegrown AI tools is also rising. AI Teammate, a Slack-integrated agent that automates tasks by analyzing chats, documents, and tickets, has expanded to more than 700 active teams, according to the document.
Pippin, which turns ideas into technical designs and documents, has become so central that some groups, including parts of the AWS cloud division, have adopted it more broadly. Other tools, such as the AI coding assistant Kiro, are also seeing increased adoption and engagement, according to the document.
Measurement
Bloomberg/Getty Images
Underlying the 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 monitored closely, with metrics including monthly active users, usage across small “two-pizza teams,” and Net Promoter Scores to gauge employee sentiment. Amazon also tracks a metric called “Value Deriving Event,” which measures the frequency of actions such as generating outputs or providing feedback.
“Set clear adoption and engagement targets,” the document said in guidance for managers. “Measure both access (who has tools) and usage (who’s actually using them).”
Amazon looks at a range of data to understand how employees are adopting new technology and what interests them, MacLachlan, Amazon’s spokesperson, told Business Insider.
Pushback to top-down mandates
David Ryder/Bloomberg via Getty Images
The rollout has created friction inside Amazon’s famously decentralized engineering culture, according to the document.
Internal feedback points to “negative perceptions of top-down, centrally controlled mandates” and concerns about overlapping AI efforts across teams.
Engineers also flagged the burden of tracking progress through self-reported goals, as well as a lack of clear success metrics and implementation guidance. Some are asking for more direction; others want room to experiment.
There are practical challenges, too. Some employees said onboarding for certain AI tools was too complex, creating barriers to adoption. Amazon is also dealing with an increase in duplicate internal tools and data, known as AI sprawl.
Amazon responds
In response, Amazon has made adjustments. As of February, leadership planned to shift guidance toward “collaborative AI practices” rather than requiring use of specific tools. The company is also moving to replace manual reporting with automated metrics and give teams more flexibility in how they adopt AI. A centralized learning platform is in the works to consolidate best practices and feedback.
“Remove friction,” the document added. “Celebrate early wins and share success stories to build momentum.”
MacLachlan, Amazon’s spokesperson, said the company does not “centrally mandate that teams use AI tools” and instead gives teams the flexibility to choose what works best for them. She added that Amazon encourages vigorous debate, and the document reflects that culture rather than any “aversion to adopting AI.”
Don’t force it
Despite its ambitious goals, Amazon’s internal approach to AI is also pragmatic.
In six “AI-Native Engineering Tenets,” the document lays out a focus on speed and practicality. It prioritizes working solutions over cost optimization and using AI where it adds value rather than forcing it into every problem. It also emphasizes transparency and the development of systems that can scale across the organization.
Here’s the full list of Amazon’s AI-Native Engineering Tenets:
Delivery first, cost second: We prioritize working, effective solutions over cheap ones. This means we will build now, then optimize for compute cost later.AI-native is not AI-exclusive: We will use the best approach to solve the problem we face. Sometimes that will require AI, and sometimes the AI will be an LLM, but not always.Cutting edge, not bleeding edge: We will not try to keep pace with AI technology. We will evaluate and retain flexibility to switch if the benefits outweigh the costs; sometimes foregoing the newest improvements.With you, not for you: We will rely on existing teams’ expertise and will not become domain experts in your area. Participating in our pilots requires bringing your domain expertise and time investment.Not all preferences are requirements: Although we will aim to delight our customers, we will not accommodate all their preferences. Instead, we will optimize for hundreds of teams, not just a few.No black boxes: All the solutions we deploy must be auditable, understandable, and traceable. We will forego performance and cost improvements to maintain human understanding and traceability.
Still, the company is doubling down on the idea that AI should become embedded in daily work.
Internally, engineers, known as “builders,” are encouraged to “experiment with different tools” and “notice when you default to manual work that AI could accelerate,” according to the document. Leaders, or managers, in turn, are instructed to set clear guidelines and ensure easy access to AI tools.
“Make AI tools part of your daily workflow, not something you reach for occasionally,” the document said.
Have a tip? Contact this reporter via email at ekim@businessinsider.com or Signal, Telegram, or WhatsApp at 650-942-3061. Use a personal email address, a nonwork WiFi network, and a nonwork device; here’s our guide to sharing information securely.
