Automating software is a really hard problem. I think I can imagine a possible roadmap to it, but it's so hard to explain it in under an hour, it would require

Automating software is a really hard problem. I think I can imagine a possible roadmap to it, but it's so hard to explain it in under an hour, it would require several sequential new technologies, and some of it hinges on parts of information theory I don't know enough about, and statistical ML isn't part of the core.