đź§  Vibe Testing AI Agent

Inspiration

AI lets teams ship in ˜˜hours˜~minutes what used to take weeks. But that same speed breaks things faster than humans can test.

I wanted to build something that could feel the pulse of an app instantly. A way to test the “vibe” of quality, not just the code.

What it does

Drop a URL.

The agent analyzes your web app (the URL screenshot), identifies the top 3 most critical user flows, runs them instantly, and generates a clear, human-friendly report written by Anthropic’s LLM.

Testing that matches AI speed! Not slows it down.

How we built it

  • Playwright for quick screenshot
  • Browser-use for live execution
  • Anthropic LLM for describing, prioritizing, and reporting
  • Daytona sandbox to run tests safely (feature idea, not implemented)
  • Galileo.ai to self-critique and refine test prompts (feature idea, not implemented - MVP implemented only with Antropic call)
  • Lovable.dev and Cursor - for quick iterations

All wrapped into one tight feedback loop: URL → tests → report.

Challenges we ran into

  • Simplicity was the hardest part.
  • Every new idea tried to grow into a full platform. I had to keep cutting until it felt effortless.
  • Another challenge: getting the AI to pick meaningful tests, not random ones — teaching it to think like a QA engineer, not a code generator.

Accomplishments that we’re proud of

  • We made testing feel light.
  • You can go from zero to your first meaningful test run in under a minute.
  • Watching the agent analyze, execute, and explain without scripts or setup feels like the future of QA.

What we learned

When speed ↑ and coverage ↑ together, confidence follows.

Formally: $$ \text{Testing Confidence} = f(\text{Speed}, \text{Coverage}) $$ Our goal is to make $$ f \text{ grow with AI speed, not against it.} $$

What’s next for Vibe Testing Agent

We’ll add:

  • Export to Playwright or Cypress code & run in Daytona
  • Deeper Daytona integration for CI/CD
  • Adaptive testing using Galileo-powered feedback loops

The mission stays the same: Make testing as instant and creative as building. 🚀

Built With

Share this project:

Updates