How to Do QA for AI-Built Apps: A Practical Guide

Why AI-built apps break differently

LLM-generated code is statistically likely to work for the happy path and fail at edge cases. The model optimizes for code that looks correct and follows common patterns. This means AI-built apps tend to have specific failure signatures:

• → Edge case inputs trigger unexpected behavior (empty strings, null values, unusual characters)
• → State management breaks under non-linear user flows
• → Error handling is shallow — the app fails silently or crashes instead of recovering gracefully
• → Third-party integrations have stubbed or incorrect assumptions
• → Device-specific behavior (different screen sizes, OS versions) is rarely tested by LLMs

The "chaos walk" testing technique

For AI-built apps, the most effective manual testing technique is the chaos walk: navigate through the app in unexpected ways, enter unexpected inputs, and trigger flows the happy path doesn't cover. Document everything with a screen recording. The goal is to find the edges the LLM didn't predict.

Prioritizing what to test

You can't test everything in a vibe-coded app. Prioritize by risk:

• → 1. Authentication and authorization flows — these are where security bugs hide
• → 2. Data persistence — does the app save and load state correctly?
• → 3. Error states — what happens when the network fails, the API returns an error, or the user enters invalid data?
• → 4. Edge case inputs — long strings, special characters, empty fields, maximum values
• → 5. Navigation flows — can the user reach all screens and navigate back without getting stuck?

Using clip.qa to capture AI-built app bugs

clip.qa is particularly well-suited for AI-built app QA because its output feeds directly back into the AI that built the app. When you find a bug, clip.qa's AI generates a prompt you can paste into Claude Code or Cursor — closing the loop without leaving the AI-assisted development workflow.

Feedback loop: from bug to fix without leaving AI tools

The ideal workflow for vibe-coded app QA:

• → 1. Do a chaos walk session with clip.qa running
• → 2. When you find a bug, stop recording and let clip.qa generate the report
• → 3. Copy the clip.qa prompt
• → 4. Paste into the same Claude Code or Cursor session that built the feature
• → 5. Ask for a fix
• → 6. Apply the fix and re-run the chaos walk on that flow

What to watch for in AI-generated fixes

When Claude or Cursor fixes a bug based on a clip.qa report, verify:

• → The fix addresses the root cause, not just the symptom (ask the AI to explain its reasoning)
• → The fix doesn't break adjacent flows — re-test neighboring features
• → Error handling was actually added, not just the happy path fixed
• → The fix is minimal — large refactors from LLMs often introduce new bugs