Q: What metrics does Amazon use to measure AI adoption among its retail engineers? A: Amazon measures AI adoption with indicators such as weekly production deployments per engineer, adoption rates across “two-pizza” teams, monthly active users for each AI tool, engagement and “Value-Deriving Events,” Net Promoter Scores for sentiment, and comparisons of access versus actual usage. Senior leaders review these signals to focus on outcomes and whether AI use produces meaningful results. Q: What velocity goals has Amazon set for its engineering teams using AI? A: The company asks more than 2,100 retail engineering teams to triple software release velocity and expects a smaller group of at least 25 teams to boost output tenfold this year. Progress against those goals is closely tracked by the S-Team to ensure gains come with steady code quality and customer value. Q: How does Amazon guard against metric gaming when tracking AI use? A: To reduce Goodhart’s Law risks, Amazon tracks both deployment rates and Value-Deriving Events, cross-checks tool usage with shipped results, and is moving from manual reports to automated telemetry. That mix is intended to show whether AI actually speeds delivery and creates value rather than just increasing superficial usage. Q: Which internal AI tools are highlighted in the article as being adopted at Amazon? A: The article names AI Teammate, a Slack-integrated agent that automates tasks, Pippin, which turns ideas into technical designs and documents, and Kiro, an AI coding assistant, while also noting industry coding tools like Claude Code and Codex. Those tools are monitored through metrics like monthly active users, engagement, and Value-Deriving Events. Q: How has Amazon responded to employee pushback over top-down AI mandates? A: Amazon shifted guidance toward collaborative AI practices, automated metrics to reduce self-reported tracking burdens, allowed teams to pick the tools that fit their work, and created a centralized learning hub to share playbooks and feedback. Leadership also emphasized removing onboarding friction, celebrating early wins, and consolidating duplicate tools to address AI sprawl. Q: What are the AI-native tenets guiding Amazon’s engineering decisions? A: The tenets prioritize delivering working solutions before optimizing cost, using AI when it adds value rather than forcing it, favoring proven over bleeding-edge approaches, partnering with domain experts, designing for many teams instead of one-off preferences, and avoiding black boxes by requiring auditable, traceable solutions. These principles guide when to deploy AI and how to measure success across the development lifecycle. Q: What lessons can other leaders learn from how Amazon tracks AI adoption? A: Studying how Amazon tracks AI adoption shows that leaders should tie metrics to shipped customer value, measure both access and real usage, and bake AI into planning, design, coding, testing, and operations rather than treating it as a one-off pilot. Automating telemetry, sharing playbooks, and giving teams flexibility on tools while consolidating duplicates are practical steps to reduce friction and sustain long-term velocity gains. Q: How does Amazon balance speed with safety and explainability when using AI? A: Amazon accepts trade-offs to keep systems auditable and understandable, explicitly requiring solutions that are traceable even if that costs some performance or compute savings. The company also monitors engineer sentiment with measures like Net Promoter Scores and uses those signals to guide adoption and maintain trust in AI tools.