I initially reacted very negatively to this article but decided to entertain the idea anyway and I have to say I'm convinced of the idea that they are overrated.
Now I'm not saying that they are unnecessary and i definitely believe they are needed but i do think they are overrated.
I've worked on many buggy systems that had very good unit test coverage. It was only with sufficient integration testing that were able to prevent constant regressions.
Of course if your integration tests are lacking you'll have problems but this does not tell you anything about the value of unit tests, just that integration tests are obviously valuable.
Unit tests have their place but in my experience there are also a lot of places in most codebases where they don't deliver enough value compared to the effort put into both writing and maintaining them (they can be a huge PITA while refactoring). I agree that in many cases integration tests often are so much more useful in catching regressions.
Now I'm not saying that they are unnecessary and i definitely believe they are needed but i do think they are overrated.
I've worked on many buggy systems that had very good unit test coverage. It was only with sufficient integration testing that were able to prevent constant regressions.