AI will be trained on the code it wrote, our feedback on that code, and the final clean architecture(?) working(?) result after that feedback.