I was gonna say that, probably the higher the abstraction level the best it is for LLMs to reason about the code, because once learned it’s less tokens.