Update on that previous post… last night I gave Claude Code too complicated of a task and then tried valiantly to get it to rescue itself out of a huge mess and ultimately ended up throwing away $30 worth of API costs at the end of it all. I really need to remember that if Claude Code doesn’t get a feature right on the first try, just throw everything out and start over. (By ‘first try’ I mean up to ‘building the feature and one or two bug fix prompts to fix minor issues’.) Any time I push Claude Code hard to fix something it got wrong, it’s just a huge waste of time and money.

I just want to reiterate how much I love Claude Code. Really makes programming so much fun and takes the hardest parts out of it: the indecision; the difficulty of taking that first step. With Claude Code, that first step is just figuring out how to express what I want, then it figures out how to build it roughly, leaving me to dogfood and iterate on it. It’s… great! I dunno how many times I will repeat this, and I’m sorry.