I do think, as some other posters have suggested that giving the LLM a design and asking it to follow that design is a good best practice.

Beyond that though, a really good way of testing is actually using the product you've asked it to build. In a pre-AI world that was the biggest beef I had with real developers. They would say "hey I built this thing" and never actually test it inside the app at all. I appreciate most people are going to say "those are bad devs" - but it was shockingly common. The view was always that this was something QA or the PM should do, but you never learn more than by actually using your own product (or code in this case).