AI slop starts with the codebase itself.

The author argues that AI coding assistants perform dramatically better with clean, well-structured codebases than with inconsistent, bespoke ones — and that this gap changes the economics of software rewrites.

Two factors drive AI code quality: the model's training data (popular stacks win) and the context the codebase itself provides. In a clean system, the model picks up patterns from a few files and delivers good implementations with minimal prompting. In an inconsistent codebase, it fumbles unless fed examples and extra documentation — you spend time teaching it your dialect before it can build anything.

That teaching time is a competitive disadvantage. The rewrite question now carries an AI-productivity dividend: rebuilding around patterns models already understand isn't just modernization, it's a compounding speed advantage. Every minute AI spends deciphering proprietary conventions is a minute competitors are shipping.