I'd expect AI to be better at writing code than at many other tasks, since programs have a natural form and encoding forced by the programming language, and the

I'd expect AI to be better at writing code than at many other tasks, since programs have a natural form and encoding forced by the programming language, and the objective function is often (at least in online challenges) easy to evaluate and unambiguous. On the other hand, interacting with real users and product managers might drive AI as insane as regular human programmers as it deals with ambiguous, contradictory, and constantly shifting requirements.