This week I ran five QA rounds on a SaaS product. 80+ test cases. 15+ bugs found and fixed. All in one day.
No Selenium. No Playwright. No QA tester.
The setup
- I describe what needs testing, in plain English.
- Claude writes a structured test suite.
- Cowork (Anthropic’s browser agent) opens the product, logs in, runs every test and returns screenshots with PASS/FAIL.
- Claude Code fixes each bug directly in the codebase.
- A fresh test round verifies the fixes.
30 minutes per round.
The wild part isn’t the speed. It’s the coverage.
The AI tester finds things a human tester scrolls right past: window.confirm() freezing in automated environments, race conditions on rapid double-clicks, missing validation on edge cases like a price of 0.
And it finds things Playwright never catches: a button that “works” in the code but is blocked by a DropdownMenu swallowing click events. An AI agent navigates the product like a user, and sees what a user sees.
The feedback loop has collapsed
Manual QA isn’t dead. But the feedback loop that used to take weeks, write tickets, wait for prioritization, argue about severity, fix, wait for re-test. Now it runs in hours.
Five rounds. One day. Zero humans in the QA chair.
The chair is still there. There’s just an agent sitting in it.