I know a company that already reaches test coverage metric goals by feeding chatgpt with code that they need test for. These are tests based on mocks, running trivial things, providing little to no value to the overall codebase. But if the requirements are garbage, the delivered items will be garbage too.