If you write code that is not testable, how do you test it? What you propose is false dichotomy really. Code that isn't testable is almost always confusing, large, hard to work with etc.