Since the LLM needs more context around mocks for the unit test, it is better in generating test cases above existing test cases. Not good for generating from scratch for complex applications.