My Cursor IDE was working fine a week back. but since yesterday, I have tried the Agent pane for Agent, Ask, Debug and Plan mode and it is stuck in the ‘Planning Next moves’ for so many hours.
Tried the following:
Switched to couple of different LLM models
Tried to add new agents
N/w diagnostics looks good so issue highlighted there
My terminal can compile and run the code so no issue with Terminal. Only the Agent Mode is not working.
Anyone else with the same issue and know any workarounds that I can try?
Steps to Reproduce
Click on the Cursor IDE ‘New Agent’
Type any instructions there and xlixk to execute it.
Its stuck in ‘Planning Next Moves’
Hey, thanks for the report. We’re seeing this issue for a lot of users on version 2.4.x. Plan mode and Agent get stuck on “Planning Next Moves”.
Here are a few things you can try:
Quick workaround (works for other users):
Roll back to 2.3.41 or 2.3.34. A few users confirmed this fixes it. You can download an older version from Download · Cursor and pick the version you need.
Troubleshooting (if you want to try staying on 2.4.21):
Settings > Network > Run Diagnostics. If anything fails, try switching to HTTP/1.1 in Network settings
Check your settings. Is indexing running. One user fixed it once indexing finished
Run cursor --disable-extensions from the command line. If it works, one of your extensions is blocking it
The team is aware of the issue. For now, the most reliable option is rolling back to 2.3.41.
This issue, if I remember correctly, has existed since the first day I started using Cursor, and it still hasn’t been resolved! I really don’t understand what the Cursor team has been doing all this time—such a simple problem has been left unfixed for so long! They’ve released countless updates in between, yet they still refuse to address this. I strongly suspect it’s intentional. If this were a Chinese company, Cursor would have been criticized to death long ago.
The Cursor team keeps refusing to acknowledge this issue and just tells people to “try, try, try…”—but isn’t it precisely because we’ve already tried everything that we’re reaching out to you in the first place? Your own software has a problem, yet you’re pretending to be clueless. Don’t you even know where the issue is? Can’t you just try it yourself?
Thanks, Dean, for the follow-up. I tried bullets 1 and 3 to no avail. It was very elusive, and I couldn’t correlate it to anything that I could duplicate. It sometimes just worked.
I finished my last project, but I’ll give it a try when I start my next one.