Blog / debugging
What to do when your AI-built project is stuck
Every AI-assisted project hits a point where the model goes in circles. You ask for a fix, it breaks something else, and you are three prompts deep with no progress. Here is how to get out.
Stop and read the error
It is tempting to paste the error back and hope. Read it first. Often it names the file and line that matter.
Shrink the problem
- Reproduce the bug in the smallest possible case.
- Remove unrelated code until it still fails.
- Change one thing at a time.
Go back to the last thing that worked
If you commit often, you can return to a working point and try again with a clearer request. Starting over from a good state is usually faster than digging deeper from a bad one.
Give the AI better context
- What you expected.
- What actually happened.
- What you already tried.
- The relevant code and the exact error.
Start a fresh conversation
Long conversations accumulate confusion. A new session with a short summary often resets the loop.
Ask another person
Write the problem as a clear question and put it in front of other builders. Explaining it often solves it, and someone may have seen it before.
Record the fix
Once solved, write down what was wrong and what fixed it. You will meet it again.
Keep your own build log
Write numbered entries, follow other builders, and let people watch your idea become real.
Join vibFolio