DEV Community

charlieww
charlieww

Posted on

I tested my app across 8 platforms with zero test code — here's how

#testingaiflutterwebdev

Last week I shipped a cross-platform app and needed to test it on Flutter, React Native, iOS, Android, Electron, Tauri, and web. Writing separate test suites for each platform? No thanks.

Instead, I used an AI agent that could see my app and interact with it. Here is what the workflow looked like:

The Setup

I used flutter-skill, an open-source MCP server that gives AI agents eyes and hands inside running apps. It connects to your app via a lightweight bridge and exposes 253 tools the AI can use.

npm install -g flutter-skill
flutter-skill init ./my-app
flutter-skill launch ./my-app
Enter fullscreen mode Exit fullscreen mode

Testing with Natural Language

Instead of writing test code, I just described what to test:

Test the login flow - enter test@example.com and password123, tap Login, verify the Dashboard appears

The AI agent automatically:

  1. Takes a screenshot to see the current state
  2. Discovers all interactive elements with semantic refs
  3. Taps, types, scrolls - just like a human
  4. Verifies the expected outcome
  5. Screenshots each step for evidence

The Results

Across 8 platforms, the AI agent completed 562 out of 567 test scenarios (99.1% pass rate). The failures were all legitimate bugs it discovered.

What surprised me most:

  • Zero test code written - everything was natural language
  • Cross-platform for free - same test descriptions worked on iOS, Android, web, desktop
  • Found real bugs - the AI explored edge cases I would not have thought of
  • Snapshot is 99% more token-efficient than screenshots - the accessibility tree gives the AI structured data instead of pixels

When to Use This vs Traditional Automation

Use AI testing when:

  • You need to test across multiple platforms quickly
  • You want to explore edge cases without writing explicit tests
  • Your team does not have dedicated SDET resources
  • You need fast smoke tests during development

Stick with traditional automation when:

  • You need deterministic, repeatable CI/CD tests
  • Performance benchmarking
  • Testing specific race conditions

Try It

flutter-skill is open source and free: github.com/ai-dashboad/flutter-skill

Works with Claude, GPT, Gemini, Cursor, Windsurf, and any MCP-compatible agent.

Would love to hear if anyone else is using AI agents for testing - what is working for you?

Top comments (0)