The AI-written-code debate asks the wrong question

Every "is it bad that AI wrote most of my app" thread misses the point. After reviewing a pile of AI-built Rails apps, the percentage barely correlates with quality. What correlates: whether the founder decided which zones AI was allowed to touch.

I split a codebase into trust zones. Low-stakes edges (CRUD, admin): let the agent run. High-stakes paths (money, auth, data deletion): a human owns them. AI optimizes for one thing: Making what you asked for run. Not making it safe when the input is hostile.

So a 90%-AI codebase with the dangerous zones owned by a human beats a 40%-AI one where nobody drew the lines.

Curious how other founders here decide what to hand off vs keep. (Full writeup on my site, happy to link if useful.)