cursor auto models often appear where no code is written, but the conversation has already been completed.
The user experience is getting worse and worse, with all kinds of charges, cloud services can’t be used, and changes to billing rules are not notified.
Hey, thanks for the report. A couple of things are going on here, so let me address them separately.
About auto models showing up where you did not write any code, can you share a screenshot of what you are seeing? From the description, it is hard to tell if this is tab completion showing up in unexpected places, the model label showing in the UI after a conversation ends, or something else. A screenshot and a Request ID from Chat context menu then Copy Request ID would help a lot.
About billing and usage, Auto mode is included in the Pro plan with usage limits. Once you go over the included usage, on demand charges may apply. You can check your current usage on https://cursor.com/settings. If you have questions about charges, share what you see there.
48c11658-267b-42a6-8b32-a885457bbd77
ad277b7c-d577-4348-8c4f-9a569650194c
——————
My question is why Cursor unilaterally changes the billing rules without notifying users, and how this issue will be handled for existing users, rather than me just checking the new billing rules. However, I haven’t seen your new billing rules’ solution for existing users, and there wasn’t even a notification.
About the billing changes, I get why this is frustrating. Updates to pricing and limits are posted on Changelog · Cursor and Models & Pricing | Cursor Docs. I agree we could’ve communicated these changes more clearly. That’s fair feedback, and I’ll pass it to the team.
On the auto models showing up when you’re not writing any code, a screenshot would still really help. Request IDs tell us about specific requests, but without the visual context it’s hard to see what’s happening in the UI.
If you have specific questions about charges on your account, check the details at https://cursor.com/settings. You can see a usage breakdown there. If anything looks off, tell me and we’ll dig into it.
From what I can see, the agent found a truncated build.gradle, tried to fix it by rewriting the file (+58 lines), then deleted and recreated it, and eventually ended up modifying 82 files. That’s definitely not ideal behavior.
A couple of questions to help us investigate further:
What was the original prompt you gave the agent in this conversation? Knowing what you asked it to do would help us understand if it went off-track or misunderstood the task.
Were those 82 file changes all unexpected, or were some of them related to what you asked?
As a general tip, if the agent starts making changes you didn’t expect, you can hit Undo or reject changes in the Review panel before they’re applied. For large projects, breaking tasks into smaller, focused prompts also tends to give better results than broad requests.
