A well-tested codebase has a number of examples, but testing certain aspects of the system results in a codebase that is resistant to some types of change.
An example is testing for specific output--e.g., text or HTML. Tests are often (naively?) written to expect a particular block of text as output for some input parameters, or to search for specific sections in a block.
Changing the behavior of the code, to meet new requirements or because usability testing has resulted in change to the interface, requires changing the tests as well--perhaps even tests that are not specifically unit tests for the code being changed.
How do you manage the work of finding and rewriting these tests? What if you can't just "run 'em all and let the framework sort them out"?
What other sorts of code-under-test result in habitually fragile tests?