I am not sure if this is Cursor or Claude. Last month when I was using Cursor 1.2 and Claude 4 Opus and Sonnet I was able to create some firm rules for a product migration and migrate 300+ files in a matter of weeks. Now if I create a rule that tells the Model 10 items it is never allowed to do - or something as simple as follow these 5 specific steps - I can be almost assured when it process a file it will not do it. It will tell me it is successful - but then when I ask it to be truthful tell me it ignored 90 percent of the rules. I am wasting a lot of $$$ on very expensive models - and having to redo things over and over again
Steps to Reproduce
Just use cursor with latest models over a short period of time
Expected Behavior
If I tell the AI to do 10 specific things - and it actually tells me the 10 things when I ask it - then I expect the AI to do them
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
Hey, thanks for the report. Could you let me know if you’re using project rules, .cursorrules, or user rules? Also, have you tried this in a completely new chat?
I use User Rules and Project Rules and also memories.
I can provide a set of rules, ask the AI Model to explain and verify its understanding of them, and then, despite that, it ignores every single item. I am currently lucky that it does 20-30 percent correctly.
I have not changed these settings recently - I was able to migrate over 400 programs with an 85 percent success rate and then within a week or so it drops down to 20 percent.
I try resetting the chats - everything - its the same
This is exactly what you’ve been trying to correct - I understand now that these shortcuts completely undermine the migration’s integrity and create more work later when things have to be redone properly.
I keep hearing people suggest Claude Code as an alternative to Cursor, but is it really a similar service? I have only briefly looked into Claude Code, so I would be interested to hear from people who have used both.
Does Claude Code…
integrate into the IDE as intimately as Cursor
have (unlimited) tab completions
automatically create restore checkpoints made with each request
allow inline requests
show diffs and allow to manually choose which changes to accept
Context and Awareness
automatically gather context with open tabs and looks through your project folder
monitor clipboard/selection and which files and specific sections of code you are just simply looking at, let alone interacting with
allow copying a section of code into the request to give it very clear context (code section and file)
allow dragging tabs, files, sections of code, and pasting screenshots all into the request context
duplicate a chat to create a sort of branch that you can return to
easily switch between models
And (for now) unlimited requests with certain models (Auto)
Are there other alternatives than Claude Code that satisfy these features?
Claude Code is not an alternative for my business - we must use Cursor AI - but it’s really hard when the quality has gotten so bad - And I pay for the most expensive Claude Opus option with no luck
Complete
