> Something that is hard to test is badly designed.
In my experience, TDD has a natural tendency to favour decoupling at all costs, and TDD zealots will push de-coupling units as an unqualified positive. And if your metric is "good design = easily testable and low coupling" it certainly looks that way.
This mindset fails to take into consideration the fact that tightly coupled things are often simpler to understand and reason about. They're more closed to extension, sure, and harder to test in isolation, but a straightforward process that acts like a "black box" and performs a job simply in a few lines of code is often better than the class explosion that rigid TDD often encourages. And what's the value of keeping things open to extension if the only thing that is going to reasonably extend them are tests?
A good illustration made by DHH as part of one of his anti-TDD tirades is here: https://gist.github.com/dhh/4849a20d2ba89b34b201
Anyone who can honestly say the latter example has a better design has a very different opinion on "good design" than I do.