How do you catch bugs in AI-generated code before shipping?

When vibe coding, it’s easy to focus on whether a feature works in the happy path

But edge cases can still break things once real users start using the app

A form might accept invalid input, an API request might fail, or a page might crash when data is missing

Fixing these issues after launch can be much harder

For developers using AI coding tools, what do you run before shipping a new feature?

Do you rely on automated tests, type checking, manual testing, or ask AI to generate test cases?