vibestacklab, AI coding tools, sourced in publicvibestacklabAI coding tools, sourced in public
Follow @devdudejeremy on X
Menu

Guide

beginner

Why your AI-built app broke and how to debug it

Updated 2026-10-15. Sourced from vendor docs · verified 15 Oct 2026 · hands-on test pending wherever a sentence depends on a vendor page.

Most fresh AI apps break for a short list of boring reasons. Read the error, change one thing, and do not start a new chat for every typo.

Answer

When an AI-built app breaks, copy the actual error, say which page you opened, and ask for one fix. The usual causes are a missing environment variable, a package that was named but not installed, an edit in the wrong file, or a free-tier limit. This guide does not claim a success rate for any tool.

Beginner path

Read this before you open a new chat

The app worked in the builder and failed when you clicked around. That is normal. Fix the thing in front of you.

Common causes, in plain language:

  • Missing env var. The code expects STRIPE_SECRET_KEY and the value is only on your laptop, or only in the builder, not on the deployed site.
  • Wrong file. The agent edited a copy, or a file the page does not use.
  • Package named, not installed. The import is there and the install step never ran.
  • Free tier. v0 stops at 7 messages a day on the free plan. Bolt free has a daily token cap. Replit Starter has a daily Agent cap. The button that "does nothing" may be the plan.
Paste the error
I opened the homepage and clicked the tip button.
The browser showed this error:
[paste the error, with any keys removed]
Find the file that renders that button.
Make the smallest change that fixes this error.
Do not refactor unrelated files.
  • I copied the error instead of describing the vibe.
  • I removed any secret from the paste.
  • I asked for one fix.
  • I tried the same click again.
  • If the plan might be out of credits, I checked usage before a second hour of prompts.

Advanced path

A tighter debug loop

Reproduce with one command you can hand the agent: a test name, a curl, or a browser step. Ask it to add a failing test only if the repo already has tests. A new test framework in the middle of an outage is how the diff doubles.

If you use Gemini CLI on a paid API key, /stats model shows that session's token use. That is a spend check, not a diagnosis. On quota plans, hitting the daily request cap looks like a broken agent and is actually the plan.

When the agent proposes a fix, read the diff for drive-by edits. Revert those before you commit. Two agents debugging the same error will fight. Pick one.

Next step

Questions

Should I delete the project and start over?

Not on the first error. A new chat throws away the files the agent just explained. Start over only when you cannot tell which files are yours.

The preview spun forever. Is that a bug in the model?

Sometimes the free allowance ran out, and sometimes the command never started. Check the tool's usage screen and the terminal before you blame the model. This site has not measured how often either happens.

Sources