Three months using Claude for test maintenance — what I actually found

I've been systematically using Claude for test maintenance tasks (not generation) for 3 months. Honest findings: **Tasks where it was genuinely useful:** - Explaining what a poorly documented test is actually testing - Suggesting why a test might be flaky (given the test code + error message) - Rewriting a working test to use newer API patterns (e.g. Playwright v1.44 locator changes) - Generating test data that matches a specific schema **Tasks where it wasted my time:** - "Improve this test" without constraints — suggestions were reasonable but generic - Anything requiring knowledge of our specific application domain - Generating tests for complex business rules that aren't in the code being shown to it **Net result:** It's a good junior pair-programmer for test code. Give it specific, constrained tasks and it saves time. Ask it to understand your system and it hallucinates.