Cursor 2.0 - agents say will do x, but then they don't do it

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

In Cursor 2.0, in the IDE, the agent now often says “I will now do x” but then they halt and they do not actually do x.

Also sometimes agent says something like “I am working on x and report back” but in fact that’s a hallucination - no work happens.

The agent needs to continue autonomously unless told not to.

User should not have to constantly tell the agent “ok now do it” or “don’t just talk about it, do it”

The agent should keep going autonomously until either success or hits a block needing actual user input

Steps to Reproduce

Happens in every chat in the IDE, every 3 to 5 minutes … agent does a few tasks then halts and says now I will do x instead of continuing.

Expected Behavior

Agent should keep working through the plan until completion or blocked.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Cursor version 2.0 new release

For AI issues: which model did you use?

Sonnet (auto mode)

Does this stop you from using Cursor

No - Cursor works, but with this issue

1 Like

same problem

I figured out it’s because of OpenAI. That model does it all the time.

I made a custom mode that uses Sonnnet all the time. Problem solved.