Came here to make the same comment. If the first AI model is buggy, then instead of debugging it with Software 1.0 techniques, just use Software 2.0 to debug it with AI! And when there are issues with the AI debugger, use a 3rd model to debug the debugger. And then a 4th. It's AI all the way down! I can see whole empires of consulting firms becoming very rich.