Then you build a helper tool that analyses the differences between good code and bad code.

Then you set it loose on its own source - and you have an AI that takes over the world. (Or at least github, which is more or less the same thing.)