I’ve been using Cursor for a short while as a developer with intermediate skills. While I appreciate what the tool is trying to accomplish, I’m encountering a frustrating pattern that’s costing me both time and money.
After exhausting 500 fast requests in just four days, I’ve noticed that roughly 70% of my time is spent in an inefficient loop: the Agent (I’m using Claude 3.7 Sonnet) frequently tells me that a bug is fixed or a feature is implemented, but when I check, I discover this isn’t actually the case.
This creates a costly cycle where I:
- Identify an issue and ask for help
- Get confirmation that it’s been resolved
- Verify and find it’s still broken
- Queue the LLM again to retry
- Repeat steps 2-4 multiple times
I’m burning through credits quickly and spending more time managing this back-and-forth than actually accomplishing coding. Has anyone else experienced this issue with false confirmations? How are you handling it? Any strategies to reduce these occurrences or make the process more efficient?