How do you handle broken dependencies in AI-generated code?

When vibe coding, the UI can look perfect while the app still has problems under the hood

A package update breaks something, an API call fails, or the AI changes a function that other parts of the app depend on

Fixing one issue can sometimes create two more

I’m curious how developers handle this when working with AI coding tools

Do you usually debug the generated code yourself, ask the AI to trace the issue, or roll back to the last working version?