Interesting point about how long it will take to train. It makes me think that the way we look at sprints and fast-iterations today may work in AIs favor.

Build this -> builds -> fix this -> fixes -> change this -> changes.

Essentially, this is a learning model already.