I wrote this little piece as part of a larger article, but I thought it might be an interesting contribution to the discourse around LLM pair programming. Not meant to be a full argument, just a note!