They work for one specific problem, locator drift, and are oversold as a cure for all maintenance. Self-healing fixes cases where the element still exists but its selector changed. One vendor analysis of real test runs put selector problems at roughly a quarter to a third of failures, with timing, data, runtime and rendering problems making up the majority. The main risk is a false pass, when the tool heals onto the wrong element or hides a deliberately removed feature. Practical controls are to review and approve heals, track how often each test heals, treat repeated heals as a sign the test is drifting from the app, and measure more than pass rate.

Sources: TestMu AI, Shiplight