A Java engineer walks through how AI (Claude) now writes unit and integration tests fast enough that "no time to write tests" is no longer an excuse — but shows why green tests aren't enough. Using PIT, a Java mutation-testing library, the team intentionally breaks code to check whether existing tests actually catch the change, exposing boundary-condition and "check exception but not return value" gaps that 100% line coverage hides. The whole thing is wired into reusable Claude commands ('of test', 'of PR review', 'of feature') stored in a shared repo/skills library so any team can install the plugin and get consistent testing conventions without re-explaining context every time.
