The idea of removing tests on private methods after using them to testdrive an implementation (I've heard of this called scaffolding tests). Focus on tests of public API at that point. (what about then moving one step further and eventually removing those public-API tests to test only user-facing functionality)
This is somewhat related to spike-then-stabilize
#person-sandi-metz talked about removing tests when she was on the #podcast-ruby-rogues