Be intentional about how AI changes your codebase

Proposes a two-tier function taxonomy — "semantic" (pure, minimal, unit-testable) vs "pragmatic" (orchestration, integration-tested, expected to change) — as a framework for keeping AI-generated codebases maintainable. Extends the argument to data modeling: enforce correctness via branded types and tight model names so wrong states are impossible at construction time, not discovered three layers deep. The underlying thesis is that explicit naming and structural discipline reduces the surface area where AI-assisted changes can silently break things.