DEV Community

Cover image for Stop Explaining Bugs to AI - Show It the Bug
Brady Stroud
Brady Stroud

Posted on • Originally published at bradystroud.dev

Stop Explaining Bugs to AI - Show It the Bug

#webdev #ai #programming #productivity

I found a better way to debug bugs with AI.

AI is brilliant at code-level reasoning, but sometimes it's painful to provide enough context of a bug to get a useful output.
So if you're trying to debug by describing the bug in chat, you can waste a lot of time.

A better approach is to stop explaining and start showing.

The Pattern That Works

When the AI can't reliably drive your UI, do this:

  1. Tell the UI to add targeted logs around the action path. It knows what to log and where to put it.
  2. Reproduce the bug yourself in the app.
  3. Write the logs to a file.
  4. Let the AI read the file and debug from facts, not guesses.

This works well, lately I've fixed heaps of bugs this way.

My prompt:

There is a bug with xxx feature where yyy happens.
1. Add logs in that area and log to a local file.
2. Ask me to reproduce the bug a few times.
3. Read the logs and find the bug.
4. Fix the bug, then I'll test again.
Enter fullscreen mode Exit fullscreen mode

UI issues are usually timing + state + sequence problems:

  • "Clicked save twice in 400ms"
  • "Request B completed before Request A"
  • "State changed after unmount"
  • "Token refreshed mid-flight"

Those are hard to communicate in plain English, but obvious in logs.

When you give AI a real event trail, it can:

  • reconstruct the execution path,
  • identify ordering/race issues,
  • map symptoms to likely code paths,
  • and suggest fixes grounded in actual evidence.

Can AI Just Use the UI?

Sometimes, yes.

With MCP-based tooling, you can give agents browser automation abilities.
For example, Playwright MCP exposes browser actions to AI agents.

So UI-driving is possible, but even with good tools, it's not always the fastest path for tricky bugs.

And if you're using Aspire, there is a built-in MCP server that can give agents rich app context:
https://aspire.dev/get-started/configure-mcp/

  • aspire mcp init can configure MCP integration for supported assistants.

Top comments (0)