While I get it, the really wild thing is that you’re better off defining tests than you are code and letting the AI define tests that fit the code.
Following the proposed pattern, the tests are only as valid as the code and still prone to inaccuracy.
