TL;DR * In traditional software, you read the code to understand what the app does - the decision logic lives in your codebase * In AI agents, the code is just scaffolding - the actual decision-making happens in the model at runtime * Because of this, the source of truth for what