however, i think there is a continuum of methods, from TDD to proving correctness. it is often practical in the real world to do up front design before hacking out some code. i believe super smart hackers actually make lots of smart design decisions when "hacking" out code, whether that is realizing the appropriate loop invariants to make an algorithm work or figuring out the right abstractions for reusable code. there's lots of ways that good forsight produces correct code faster. that's what intelligence is: thinking, not brute force.
the smart coder, especially after gaining a variety of experiences and learning from different mentors, takes pieces of different methods in order to simultaneously make headway on different goals, not just correctness and extensibility, but refactorability, fast-rampup for new team members, fun, etc.