What I'm finding quite interesting, is prompt creation for claude code. Opus is expensive, and still makes mistakes, I'm not just talking cost, but tokens which affect app / browser sessions. So I've been trying Sonnet on medium, it pumps things out at times a bit quick for my liking, however, I then ask it to go back over its last response, read it, think about it, and if correct, do nothing, otherwise output any corrections. Doing something twice, but each time thinking then engages and I get corrections to either mistakes OR primarily additions it didn't think about the first time writing it. Cheaper on sessions, less time-outs and hitting session limits going into paid overtime or waiting for a session to reset.