I don’t know how anybody uses Claude these days. Its so terrible at reasoning and all it does is apologize. R1 and even o3 mini kills it.
Hey, I’d recommend starting a new Composer session, and ensuring you don’t have any Rules for AI that may interfere.
Claude 3.5 is great at out-of-the-box performance but can be incredibly sensitive to comments and how you treat it. If you start to tell it that it’s getting things wrong, it quickly falls down this hole of performing worse and apologizing again - a kind of self-fulfilling cycle.
Regularly starting new Composers, ensuring your context is precise for the changes you want to make, and being careful with your rules are all great steps to keep on top of Claude!