Hidden Technical Debt in AI-Generated Codebases — Why Working Code Gets Expensive to Change

"It worked fine for six months. Then every change became expensive." Your AI-generated app is live. Customers are paying. The dashboard looks fine. But every feature takes 2-3x longer than it should, and nobody can explain why. This post explains what hidden technical debt looks like in AI-generated codebases, how to detect it, and the structural decomposition that fixes it. What Makes AI-Generated Debt "Hidden" Traditional technical debt is visible — you know you're cutting corners. AI-generated technical debt is invisible — the shortcuts happen at generation time and nobody notices. Here's the mechanism: Session 1: User service created (clean, 80 lines) Session 15: Pricing logic added to user service ("it needs user data") Session 30: Notification logic added to pricing flow ("notify on purchase") Session 45: User service is now 600 lines, handles 4 domains Session 60: "Why does adding a referral system take 3 weeks?" Each session makes a locally reasonable decision. The AI puts code