Since the last update, the Auto agent has become thinking, with dummy results and decisions every time! This is insane, and I really can not use it anymore.
The Ultra pack is not for my daily work because it is consumed. I am over $600 for now, and for small and dummy tasks, I am using Auto. Now I really can not! Could you please be so kind as to revert back to Auto as it was before this release?
Steps to Reproduce
Send a message to Auto agent
Expected Behavior
to be as it was before the relase
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)
Hey, thanks for the report. We’re seeing similar Auto mode issues after the 2.0 update from many users.
Workaround: instead of Auto, manually select a specific model (Claude Sonnet 3.5, GPT‑4, etc.) in the model selector. This should give better quality while we investigate.
To help debug, please:
Share examples of the “dummy/meaningless results” you’re seeing
Add a Request ID (Cursor Settings → Privacy → disable privacy, reproduce the issue, then copy the Request ID from the chat)
Let me know if switching to a specific model helps!
Omg, I thought it was just me. I’ve spent the last few hours wondering if I’d gone insane. The agent has literally become the most tediously inefficient temp employee. It used to understand when to take futher action, when to look for more files, when to dig deeper. It was almost self-sufficient. Now I feel like we’ve regressed months in terms of how much hand holding it needs.
The problem is the bill @deanrie. It is counting on every hit of the enter button. It cost sometimes $1 or even $2 when using Claude. Auto was good for everyday tasks but now it is like GPT3 in its early stages.
