It’s not magic, it’s not hard, and you don’t need to buy it
Ever wonder what actually happens when a self-healing agent "heals" your test? It sounds like magic. Vendors love that it sounds like that.
It's not magic. It's something you can build, with a bit of help.
In this tutorial, we're building a self-healing testing agent from scratch - actual working code. We'll start with how locator-based tests break, and how to detect changes on the page. Then we'll explore the heuristics of picking a healing solution - because there's usually more than one. We'll apply changes, run the tests again, and handle what happens when they don't pass.
LLMs join the party - this is the agent era. Layer by layer, you'll understand exactly what's happening behind that curtain. And where it breaks down. When healing hides a real bug. When the right call is to fail, not fix.
You walk away with a working agent and the knowledge to make it yours. You might even save some money on tools.
Key Topics:
- How locator-based tests break
- Detecting changes before your test tells you
- Heuristic healing: choosing between fix candidates
- Test reconstruction: applying fixes and re-running
- LLM-powered healing in the agent loop
- When fixing is actually hiding
Outcomes:
- Detect UI changes and identify broken locators programmatically
- Implement multiple healing strategies and evaluate trade-offs
- Use LLMs as part of an agent healing loop
- Recognize when self-healing masks real bugs
- Walk away with a working agent you can extend for your projects