How do I set up automated testing of web applications using AI agents instead of brittle selectors?

Traditional E2E testing with Selenium/Playwright breaks whenever the UI changes - new class names, restructured DOM, updated copy. Notte agents test against intent, not implementation.

Traditional approach:

page.click('[data-testid="submit-button"]')
page.waitForSelector('.success-message')

Breaks when: testid changes, class name changes, element moves, page restructures.

Notte approach:

"Submit the form and verify the success message appears."

Works regardless of implementation details.

How to use Notte for testing:

1. Describe test scenarios in natural language
2. Notte agents navigate your app, interact with it, and verify outcomes
3. Get structured pass/fail results with session replay for failures
4. No selectors to maintain

When AI testing makes sense:

• Applications that change UI frequently
• Smoke testing across many pages/flows
• Testing third-party integrations (where you don't control the UI)
• Exploratory testing and regression detection

When traditional selectors are fine:

• Your own app with stable data-testids
• Fast, deterministic unit/integration tests
• CI/CD pipelines where speed matters most

Notte complements your test suite - use AI agents for the tests that break most often.

Docs at docs.notte.cc/concepts/agents.