AI Code argues for cleaner codebases

This is less a product launch than a style guide for the age of coding agents, and the core message is right: if AI is going to write more of the code, humans need stronger structure, not weaker standards.

• –The semantic-function vs. pragmatic-function split is a useful way to keep agent-generated code modular and reviewable.
• –The model advice is the strongest part; fewer optional fields and sharper types reduce the chances of AI creating messy, ambiguous states.
• –Shipping the guidance as a skill makes it actionable, not just philosophical, which is exactly what most AI-coding advice lacks.
• –The risk is dogma: if teams turn “self-documenting” into naming theater, they can accidentally add abstraction without adding clarity.
• –For AI-heavy teams, this reads like governance for codebases: constrain the surface area where agents can improvise, then test the rest hard.