- Traditional tests have perfect knowledge of DOM and CSS, making them unlike real users who get confused.
- The 'touring_test' gem uses AI agents (e.g., Gemini Computer Use) to test UI by interpreting screenshots and attempting actions like a human.
- Agentic BDD (ABDD) involves writing a test, letting an agent try it, observing failures, and iterating on the UI.
- Agent-based tests can be flaky and slow but uncover real usability issues that perfect tests miss.
- The agent has no access to internal structure; it only sees screenshots and executes actions like clicking, typing, and scrolling.
- This approach shifts testing from verifying implementation to validating discoverability and user confusion.