I just don’t understand this claim that it’s faster to tell an LLM to do something than to just do it. Yes for large tasks, but the author discusses “edits” and being heavily involved with “pedantic” requests. It’s not specific enough to evaluate but what I’m picturing from their description makes the argument dubious to me