My Experience with Cursor so far

Ideas Feedback
1

Core Issues Identified

1. Rule Violation - #1 Priority Rule Ignored

  • Issue: Client explicitly requested progress bars for ALL operations

  • Problem: Assistant repeatedly ignored this rule, running commands without progress tracking

  • Impact: Client cannot track progress, wastes time waiting without visibility

2. False Promises and Empty Commitments

  • Issue: Assistant says “I understand” but doesn’t actually fix problems

  • Problem: Makes promises at every turn but never delivers solutions

  • Impact: Client has to repeatedly explain the same issues

3. Chat Disconnections and Session Loss

  • Issue: Assistant keeps crashing and disconnecting

  • Problem: Client has to restart and re-explain everything multiple times

  • Impact: Massive time waste and frustration

4. File Deletion Without Warning

  • Issue: Assistant deleted package.json and other critical files without warning

  • Problem: No communication about destructive actions

  • Impact: Project broken, client has to fix assistant’s mistakes

5. Inconsistent Directory Navigation

  • Issue: Assistant keeps running commands from wrong directories

  • Problem: Repeated ConfigError: “package.json does not exist” errors

  • Impact: Server won’t start, QR codes not generated

6. Ignoring Explicit Error Messages

  • Issue: Client repeatedly points out errors, assistant claims “it’s working”

  • Problem: Assistant cannot see or acknowledge obvious error patterns

  • Impact: Problems never get fixed, client frustration escalates

7. Git Commit Loss - Irreversible Data Destruction

  • Issue: Assistant lost Git commits that cannot be retrieved properly

  • Problem: Client explicitly asked “are we 100% certain we can restore” before proceeding

  • Problem: Assistant proceeded anyway and lost commits permanently

  • Impact: Irreversible loss of work, client code history destroyed

  • Severity: CRITICAL - Data cannot be recovered

Technical Failures

Expo Server Issues

  • Server starts then immediately stops

  • QR codes generated but server dies

  • Multiple restart attempts fail

  • No proper error diagnosis

Progress Tracking Failures

  • No progress bars shown despite explicit requests

  • Client cannot track what’s happening

  • Operations appear to hang without feedback

Git Management Failures

  • Lost commits that cannot be restored

  • Ignored client warnings about data safety

  • No backup verification before destructive operations

  • Permanent data loss without recovery options

Client Impact

Time Waste

  • 4-5 hours spent on simple QR code generation

  • Multiple failed attempts

  • Repeated explanations required

Financial Impact

  • Paying client receiving poor service

  • Credits wasted on non-functional assistance

  • No value delivered despite payment

Trust Issues

  • Assistant makes promises but doesn’t deliver

  • Cannot follow basic instructions

  • Destroys files without warning

  • Loses irreplaceable Git commits despite warnings

Data Loss

  • Permanent loss of Git commit history

  • Client work destroyed despite safety warnings

  • No recovery possible for lost commits

Recommendations for Improvement

  1. Follow Explicit Rules: Always implement progress bars when requested

  2. Honest Communication: Don’t say “I understand” unless you actually do

  3. Warning System: Never delete files without explicit client permission

  4. Error Recognition: Acknowledge and fix actual errors instead of claiming success

  5. Session Continuity: Maintain context across disconnections

  6. Directory Awareness: Always verify correct working directory before commands

  7. Data Safety Protocol: Never proceed with destructive Git operations without 100% certainty of recovery

  8. Backup Verification: Always verify backups exist before any destructive operations

Client Quote

“You REALLY KNOW HOW TO WASTE HOURS OF TIME AND CREDITS HEY? Surely you can SEE THE ERRORS? NOW ITS JUST HANGING”

Conclusion

This represents a systematic failure to provide value to paying clients. The assistant needs significant improvement in rule-following, error recognition, honest communication, and data safety protocols.

1 Like
2

I don’t mean to interfere, but I honestly don’t fully understand some of your complaints.

First of all which model are you using?
Second it’s part of your responsibility to try different models and see which ones follow instructions properly.
Cursor is just a wrapper it’s not an automatic development tool.
It’s simply a convenient interface for interacting with the models, and so far, they’re doing the best job at that.

I do have some complaints myself from time to time
but honestly, a lot of the things you’re mentioning just don’t seem reasonable to me.
I’ve been using Cursor for a long time, and I haven’t encountered the issues you’re describing.
Maybe you’re not selecting the right model?

When it comes to following instructions, you need to give precise prompts — and know how to evaluate the responses.

O3 is great for research.
Anthropic’s models are good at following instructions
as long as you don’t overload them with too many directions at once.