I would think an LLM would be useful for explaining behavior, whereas visualizing code references can be done with no AI at all.