How to make sense of AI

A practical sensemaking framework for AI practitioners: ignore opinions and predictions, focus exclusively on detailed "field reports of use," and evaluate them through four questions (possible outcomes, actions, relative value, causal relationships). References concrete agentic coding examples including Microsoft's "compounding teams" — producing complex software with zero human-written code after ~6 months of scaffolding investment. Argues that engineering best practices (TDD, observability, CD) are prerequisites for high-velocity AI-first development, making this directly relevant to engineers weighing when and how to adopt agentic workflows.