@dalias To be clear, I'm not convinced by the proponents of LLM-based coding agents. I find the idea of having a statistical text generator pump out volumes of code from ambiguous natural language distasteful. And I sure wouldn't want that approach to be used for something like musl, where you clearly work on it deliberately, carefully, with no line of code wasted.

@dalias Doctorow seems to feel that this is what he would be doing; he finds the LLM useful. And some programmers I follow and respect feel that way about their LLM-based coding agents (using the big rented models, not a local one like Doctorow), that they'd be denying themselves something useful and putting themselves at a disadvantage for moral reasons.