You should only write tests when that part of the codebase has real users relying on it not breaking, real contributors who might inadvertantly break it, and everyone involved has reason to believe that further experimentation is pointless.