Self-Healing Test Automation Treats the Wrong Disease
TL;DR
Self-healing test automation patches a broken locator and keeps the run green. We built it, then removed it. A heal cannot tell a moved button from a broken feature, and at the churn rate of AI-written UI it becomes a tax. Write the test as user intent. There is nothing left to heal.
We shipped self-healing test automation. Then we took it out.
Not because it didn't work. It worked often enough to look good in a demo. It failed in the one place that matters: telling you whether the red test was a moved button or a broken feature.
This post is the developer-side view of self-healing test automation, from a team that shipped it, paid for it, and replaced it.


