Card #347 · Cursor (AI IDE)
AI agent generates noisy output with hallucinations and self-correction loops
The Cursor agent frequently produces output containing 'noise', such as fabricated tags or contradictory numeric IDs. Users report that the AI enters self-correction loops, wasting usage tokens and producing unreadable responses mixed with internal reasoning markers.
Pain score
0.24/ 1.00 · weighted product of four components
Reach0.63distinct authors (log-normalized)
Recency1.00freshness decay, half-life months
Engagement0.83upvotes + comments, normalized
Monetization1.00willingness-to-pay cues in evidence
§ Evidence — 4 verified quotes
AI sometimes generates output that includes noise or content we don’t want.
The Cursor Agent is consistently hallucinating a non-existent tag when generating or editing React components. It injects this fabricated tag into the code and then immediately burns extra execution steps (and my usage limits) trying to fix its own mistake
the delivered message mixed correct technical statements with contradictory register IDs and procedural text meant for hidden reasoning.
the auto‑review feature doesn’t seem to trigger or produce any meaningful output.
Comments are a Pro feature — join the conversation, share what you've shipped, hear what others are building.
Upgrade to Pro