It's strange that they did not mention reworking the tests themselves. In my experience AI-generated tests are absolute garbage - unless you specifically prompt to reason about coverage to avoid duplication and to merge some.