I find that most of the time it's not that agents make mistakes as much as they're not doing it how I wanted. MIstakes are not in code but in interpretation and the blame is only partly mine — how can it reproduce something it never saw? We still need humans to innovate.
commit 1fff764