Testing websites and GUIs

I’m at a point where I would feel comfortable vibe coding some website (though the core logic is carefully human-written), but I still don’t know how to write tests for a website. If the LLM implements feature B while breaking feature A, that’s not much progress forward. How do you all deal with this? The LLM itself seems to recommend (normal unit tests and) using playwright. Playwright looks popular, but it has apparently never been mentioned on the julia discourse, does anyone have experience...