📝Self-Healing Tests are automated tests that an AI agent repairs when they break, by diagnosing whether the failure reflects a real defect or a stale test and then updating the test or the code accordingly.
The idea began in user-interface test automation, where tests break when a button's selector or a page layout changes even though the feature still works. Self-healing tools detect the broken locator and substitute a matching one. Agentic coding broadens the idea: an agent reads the failure, inspects the change that caused it, and edits the test, fixture, or source code.
The key judgment is which side to fix. If behavior changed intentionally, the test should be updated; if the code regressed, the code should be fixed. An agent that edits tests until they pass can hide real bugs, so well-designed setups require the agent to explain each repair and keep test changes reviewable.
Self-healing tests usually run inside continuous integration, where a failure triggers an agent rather than a person. They reduce maintenance on large suites and are one building block of agent-driven 📝CI/CD pipelines, alongside 📝Eval-Driven Loops and automated code review.