If an AI model implements a feature for you with errors, and one bug fix attempt doesn’t work, I usually roll back and ask for the feature again. That stops the codebase from getting worse through dozens of partial fixes, as well as muddying up the chat context with debugging info rather than product info
If it still doesn’t work after the second attempt, try breaking it down into smaller features. For example, if you’re building an image upload system, ask for just the UI in one message and then have it upload to Vercel Blob in a second message. You can chat with v0 to get ideas on how to break features into parts
Here’s some more info on how v0 works and how to get better results