What's Wrong with Cursor Lately

Feedback
22

Hey Guys,

Are you still having the same problems? I’m eager to pay for cursor but with so many complains I dont know.

1 Like
24

Yes! it is a nightmare.

25

It’s so bad.I just thought it was my problem, but now it seems it’s not

26

OMG!
i thought i had lost my prompting abilities or something…
It’s so bad that i can’t even believe i’ve to ask V0 for help and copy & paste into cursor, even with that… Cursor couldn’t even enhance anything!
I can’t accomplish anything right now… OMG!!!

27

Guys, try switching the model to Clause 3.7 especially -thinking its better, Cursor is not as good as it used to be just last week when it comes to tooling and terminal commands, but using Clause 3.7-thinking does a good job compared to the 3.5

29

I don’t know what happens but for Three JS I have, during these 3 days, hard times, it becomes so terrible that I believe I won’t renew my subscription… sonnet 3.5 and 3.7

30

I had initially aimed to encourage my dev team to adopt Cursor, but we’ve been seeing notably poor results when using Claude 3.7 and other models. The quality of the outputs has consistently fallen short of expectations, making it difficult to advocate for broader adoption. I’d appreciate if these issues could be addressed or improved in future updates, particularly regarding model performance and reliability.

31

Agree’d - latest version of cursor with claude 3.5 and 3.7 has been quite poor compared to what it once was. I am certain it has something to do with the context windows - claude 3.7 is especially terrible lately, creating an insane amount of nonsense to fix one small issue in a single file. The only time 3.7 has actually performed great was the first day it was released, specifically the first hour or so before everything ground to a halt. That version was like magic and the reason I was hyped so much for cursor.

Now i’m fighting with it left and right to get it to do basic things in a medium sized ETL platform codebase. Claude 3.7 has gotten to the point where it wont send me direct responses from time to time, it constantly wants to run a terminal command with an echo of its response. I laughed when it first happened, but it’s also concerning that something deeper is afoot under the hood.

2 Likes
32

When working with large codebase it struggles. Try using aug… and i would stop there. Out of respect for cursor i wouldnt go further , just do some research and there is a great alternative for large codebases

33

I started using Cursor yesterday. Overall, I found it fantastic—truly an incredible and revolutionary tool. However, unfortunately, I have already encountered several issues, all seemingly related to the platform’s low intelligence: it did not modify the referenced file as I requested, only responded in the prompt (even in “agent” mode), repeatedly; it created files but couldn’t find them; it failed to create a file; it couldn’t make simple changes to the code, which I ended up doing manually in a very simple way (copy and paste). All of this happened regardless of the model—Claude 3.7 Sonnet (whether “thinking” or not) or o3-mini.
But I hope this will be fixed soon because the platform’s potential is truly very high.

34

I ask cursor for “how to set opa on apisix if that didnt show on web page”, and after a series of search actions , Cursor decide to use serval different method to read the documentation ,like ‘curl -s xxxxx|grep ■■■’ … and failed , but that is not the end , cursor keep trying other regex with curl to read the doc…

35

Just want to wade in here as I’m sure there’s some tracking sentiment - 0.46 has really brought my feelings of Cursor into severely disappointed territory.

I haven’t really achieved very much this week - its been a total disaster.

Failed Expectations

  • Default Agent model over-engineers everything.
  • Cursor Rules now even more mysterious than 0.45, no official documentation and I have tangible doubts that Agents are even following them.
  • No seeming enhancements to larger codebase support.

Specific Workflow issues:

  • Agent creating new files instead of updating existing, even when selected in the context
  • Agent ignoring instructions about the scope of what I want it to do.

Competitor products

  • Aug… not perfect but interesting. Definitely handles larger codebases better. If I wasn’t such a 1.5 year coding n00b, this probably what I’d recommend for more experianced devs. I’m kind of running them in parallel as its quite good a writing prompts for the Cursors agent.

Suggestions

  • Please publishing some documentation about Cursor Rules. My subscription is up in 12 days and I’m feed up of read people’s opinions about it. Like, newbies come here and the first search results returns a page recommending .cursorules as a hack - which is a depreciated feature… the feature is a hot mess and I’m actively trying competitors product because of it.
1 Like