I have been going around in circles for days with the app to update a list returned by an API. It is all over the place with its recommendations, building random folders out of the file hierarchy, asking me to deploy updates to files with “no changes made”, and ignoring the documentation that I am supplying it. It was literally guessing at solutions instead of reading the doc.
Am I just expecting too much? Here is its output after I complained about its poor performance:
I apologize for my poor performance. You're absolutely right - I've been:
* Not carefully reading the error messages
* Making random guesses at solutions
* Not tracking the actual file state
* Wasting your time with non-solutions
There should be a “submit for refund” option when this is happening. I know it’s only 4 cents per query, but it’s frustrating to pay for something that is the equivalent of keyboard smashing.
I understand the frustration with inconsistent AI behavior. Cursor works best when used alongside your development workflow rather than as a complete solution. The AI can sometimes make educated guesses when it’s uncertain, but it should generally improve with clearer context and properly indexed documentation
For the specific issues you mentioned:
- Random folder creation isn’t expected behavior - worth checking your
.cursorignore file
- For file updates, you can use
@Recent Changes to help Cursor track modifications better
- To improve doc usage, try explicitly adding them via
@Docs → Add new doc
You can check out our docs on codebase indexing and documentation setup for more details on getting better results
If you’re within the refund window (24h for monthly, 72h for annual), you can get an automatic refund through the billing portal. Otherwise, reach out and we can look into your specific case!
There have been code changes in my repository in the past week, but Cursor cannot detect any changes. What could be the reason for this?
