Comment by Schiendelman

Maybe this is a dumb question, but isn't this solved by publishing good API docs, and then pointing the LLM to those docs as a training resource?