https://eml.systems/blog/Perspectives on AI products, engineering, and the people around them.enhttps://eml.systems/blog/ai-quality-needs-ownership/https://eml.systems/blog/ai-quality-needs-ownership/Product quality crosses team boundaries. Give consequential decisions clear ownership while making room for the people who understand different kinds of failure.https://eml.systems/blog/a-smaller-useful-experiment/https://eml.systems/blog/a-smaller-useful-experiment/Before committing to an AI feature, identify the uncertainty that could change your direction. Then choose the smallest investigation that can credibly address it.https://eml.systems/blog/the-interface-is-part-of-the-ai-system/https://eml.systems/blog/the-interface-is-part-of-the-ai-system/People need to inspect, correct, and steer AI-assisted work. The interface determines whether those interventions are possible and what happens when a suggestion becomes an action.https://eml.systems/blog/a-faster-coding-agent-changes-the-work/https://eml.systems/blog/a-faster-coding-agent-changes-the-work/When implementation gets faster, specification, review, and project understanding may become the constraint. Measure the accepted change and improve the workflow around it.https://eml.systems/writing/measure-the-whole-job/https://eml.systems/writing/measure-the-whole-job/An impressive answer can still create more work. Evaluate the complete workflow to understand whether an AI product makes a useful difference.https://eml.systems/writing/from-demo-to-production/https://eml.systems/writing/from-demo-to-production/A prototype helps a team learn. A production product creates obligations. Make those obligations explicit before people begin depending on the system.https://eml.systems/writing/context-is-part-of-the-system/https://eml.systems/writing/context-is-part-of-the-system/Useful AI behavior depends on the information available at each step. Treat retrieval, memory, tools, and working state as decisions you can inspect and evaluate.