Technically an LLM is a tool for extracting candidate responses to plain-text requests. Since (textual) programming languages are languages, they can create passable candidate responses to queries about those. Certain LLMs such as Copilot and Claude have had their training focused a bit more towards programming tasks, but saying that LLMs as a class are for coding assistance is a little narrowly stated.

It would maybe be handy to feed the responses from an LLM through a computational reasoning engine to grade a few of them.