I had a very specific problem I wanted to solve, Opus kept being entirely sure that it knew what the problem was, and that it could fix it. Next iteration, when

I had a very specific problem I wanted to solve, Opus kept being entirely sure that it knew what the problem was, and that it could fix it. Next iteration, when the previous solution didn't work, Claude was sure it was some other thing. So on and so forth, until I was $35 poorer and had a bunch of code that didn't really seem to do anything, and certainly not fix my problem. Planning and discussion wouldn't have fixed that.