E2E Testing with agent-browser

Hauke Jung

Hauke Jung

— 1 min read

I've been experimenting with agent-browser in Claude Code, and it's changing how I think about end-to-end testing.

The Old Way

await page.click('#submit-btn-v2-new');
await page.waitForSelector('.modal-content > div:nth-child(3) > button');

Selectors break. Tests become maintenance nightmares.

The New Way

Write test cases that read like requirements:

TEST-ONBOARD-001: New User Redirect

Priority: Critical
Preconditions: User with no organizations

Steps:
1. Create test scenario with user having no orgs
2. Login as that user
3. Navigate to /dashboard
4. Verify redirect to /onboarding
5. Verify plan selection cards displayed

Expected Results:
- Users without orgs redirected to onboarding
- Plan cards show Starter and Pro options

Hand this to agent-browser. It navigates, clicks, validates. Done.

Why This Matters

  • Tests read like specs. PMs understand them. No translation layer.

  • Resilience to UI changes. Button renamed? Class refactored? AI adapts.

  • Faster authoring. Describe the flow, run it, ship it.

Read more