Same, something weird has been going on since approximately Tuesday for me. During the weekend and before, Claude worked impeccably. Then, since Tuesday or around that it felt like going back to gpt-4o for coding. I’m wondering whether my expectations have gone up unrealistically and I somehow misremember it being much better (or if it was just a lucky few days), or if its quality has indeed declined…
Same here with Claude. Went from nailing it on the first try to failing after a full day of trying to debug some scripts. Won’t say it’s Cursor because it’s probably not but there is a noticeable degradation in just the past couple of days.
2 Likes
new around here. That sucks! No, I am certain you guys aren’t crazy and the degradation is very real from the sounds of it. I was previously using sourcegraph cody and claude sonnet was working amazing for me. I hope they figure out the regression in intelligence soon, because that’s a show stopper for me. I was looking at cursor due to the new composer feature and while cursor is a significant jump in price especially for a hobby coder like me, I would pay it just for the bleeding edge cool features I see going on here. I’ll be keeping an eye on the situation and if it improves again, you’ll have a new customer. 
How is this degradation fix going?
Right now I’m on supermaven while wait for a fix for this.
I was just about a start a thread for this.
I came across cursor on YouTube and tried similar prompts that are shown in cursor demos of older versions in those videos.
The new versions of cursor is not even close to creating similar code. It’s messes up so much taking in a credits. And also the prompt responses are more chatter than code itself.
Plus most of those videos people are able to attach images screenshot, now that that is not working it getting off track.
And these are not complicated apps.
Given that the response will not be the same it will need to be close and it not.
Time to ask those video creators and issue updates. Thinking of creating a video about these issue my self.
Any response from the cursor creators on what is going and what to expect in the coming days?
Can we cancel pause our subscription or issue refunds ?
I have also noticed that composer (at least with claude-3.5-sonnet) has become dumber, compared to ~7 days ago.
For example, I asked the composer to do refactoring to one file based on three similar tagged files, and it failed miserably. Somehow it can not follow the structure and style of the tagged files. Maybe a context issue? I did similar tasks just one week ago and it worked perfectly.
I can’t share a code example since its customer’s code.
Cursor team, please let us know what is your take on this issue.
Update: Using CMD-K to do a similar task using Claude worked while it failed in composer. Something is wrong here.
Agreed here, its gone from usually-helpful to unusable for anything but the simplest questions
This is why Cody is now better than Cursor. Cody gets it right the first time with Sonnet. Cursor give lazy results now, the apply feature is completely broken. Codys apply feature works awesome and I dont have to worry about it deleting valid code like cursor.
Seems cursor’s team is losing it. Every release they fix something that breaks something else. It is like they are pushing releases out of pressure rather than fully testing it