AI Code

A manifesto and guide for maintaining code quality when working with AI coding agents. Covers two function archetypes: semantic functions (minimal, pure, self-describing, highly unit-testable) and pragmatic functions (complex orchestration wrappers with doc comments). Emphasizes data modeling discipline — making invalid states unrepresentable, avoiding optional field sprawl, and using branded types to prevent silent ID-swap bugs. Warns that AI agents can rapidly degrade a codebase if not guided by intentional patterns, and that both functions and models tend to decay in predictable ways over time.