The AI guesses at anything it can't see. Learn which file to open, which error to paste in, and which example to give it.
Take a moment to think about it this way
The context window has limits. Telling the AI to "read the whole project" sometimes makes things worse, not better. The right context is small and relevant: the file to be edited, a similar example file, the full error message, and a description of what the current UI looks like. The more unrelated files you throw in, the more confused the AI gets.
Let's connect this to everyday life
Want to change a button style? Just open Button.tsx and the one page that uses it. Got an API error? Copy the request URL, status code, and response body. "It's not working" on its own isn't context. In Cursor you can attach a file with @. In chat, just paste the relevant function. Never include secrets, .env files, or user data.
Let's try it hands-on, together
Paste this, not “it is broken”:
File: apps/web/src/app/api/notes/route.ts
Action: POST /api/notes
Status: 400
Body: { "error": "title required" }
Expected: empty title shows inline error, not a blank pageBe able to put together a context bundle the AI can actually use to understand a bug.5-minute try-it
Pick a bug you've actually run into. In 8 lines or fewer, write down the file, the action, the expected result, and the actual result. Double-check that no secrets are included.
One quick warning
Never treat AI output as the final, correct answer. Read through code, secrets, and user data yourself, test it, and only then put it to use.