Absolutely, there's no reason to believe that agents will be more capable of writing tests than any other piece of code. The big pay off is actually verifying the code that was generated.