It's similar to us. We can reason about certain things but we often have to look up additional information in a book or ask someone. In the same way I see LLMs being augmented with traditional code. It doesn't really matter if the result is called AI or not.