I am open to new ideas especially if they can work in VSCode so I don’t have to have a separate editor open just for AI
It’s a hot mess right now and for some odd reason lately, it’s been getting worse rather than better. And I mean that it’s odd cause it can have moments where it behaves followed by moments of stupidity of such magnitude that you almost feel embarrassed for even trying to use a pile of junk like this. The frequent time outs in the composer. There are incredible slowdowns over extended sessions and the constant need to reload. And don’t even get me started on Sonnet’s sadistic edge to make a dedicated effort to misunderstand you at moments when it’s least convenient. This toy can make you feel like the clever nerd in 6th grade elementary till the teacher punches you in your face for a solid wake-up call.
Works fine for me, I’ve noticed no degradations in the model performance and its bizarre to me how many people complain on the forum as if Cursor is the one that owns the models and not Anthropic or OpenAI. The suggestion that OP and other people have made that the devs are basically committing fraud against their customers by not using the models the user selects on their backend seems very implausible to me as well.
There are problems with Cursor itself that can get frustrating, namely how slow apply is sometimes but other than that my experience has been good. I’ve been using it since the end of August 2023 so quite a long time and I haven’t regretted it at all.
6 Likes
Think most of us are quite aware of who owns what. Fact of the matter is, Sonnet (Claude, Antrophic yes) is the most capable LLM to interact with right now and the fact is also we do this inside cursor. So sure, people who don’t want to specify every corner in their frustration that things fail to materialize sometimes, just container the whole thing to make a point. Other than this, congrats on not sharing our difficulties. That’s an envious position.
Well it just seems misdirected, what’s the point in seething about Claude getting worse on the cursor forums as if the devs have any control over that?
2 Likes
Stop with the features. And make the thing stable.
It can literally cure give you mounds of gold, but if it doesn’t work or its not stable, then its wothless.
2 Likes
I’m not here on the forum every day to read all the complaints, so I could be wrong about this, but I would think that it isn’t really a matter of people coming to the Cursor forum to complain about the quality of Claude AI (which I agree would be silly). I think rather it’s a matter of Cursor being the middleman between the user and Claude. So while they have no control over Claude itself, they have full control over the prompt that ultimately gets sent to Claude.
When things work great for a time and you become familiar with a particular LLM and how it behaves, and then suddenly it’s not behaving the same way anymore and it’s being far less helpful, I think the assumption or best guess is that it’s more likely that Cursor is at fault than Claude.
That is to say, perhaps something has changed and Cursor is doing something different internally than it was doing before, like a change in the way it processes a request or whatever occurs behind the scenes for it to construct the final prompt that it sends off to Claude. And because it’s a black box (as far as I’m aware), the user has no way of knowing what prompt or context was actually sent to Claude in the end, making it harder to know where the problem truly lies.
If I’m off base here, let me know. Eager to learn 
1 Like
While the topic is a bit harsh I have to agree (On 0.41.3 @ MacOS):
- My chat context has not been applied about 50% of the time (file picker selection not put in context) and probably have wasted lots of requests for that reason.
- Composer just randomly omitting existing code while writing new and breaking things
This inconsistency typically leads into unsubscribing when I get frustrated enough. But, eventually I have returned due to excellent Cursor Tab model, automatically-scoped Cmd-K inline edits and couple of niceties that work better than the competition (e.g. Supermaven… I have tested the Pro and just don’t find it useful at all).
My message to the Cursor team: It seems that there is way too much of moving fast and breaking things. Focus on your core value-providing strengths and hone those to the perfection! Ensure that things become stable because it is one thing to become frustrated due to LLM capabilities themselves but another thing when IDE gets broken every now and then.
1 Like
You pay 20 bucks per month. For that you get one of the best software out there where highly specialised developers work 24/7 to provide you with value.
Also, you get free access to LLM models usage - models that cost over $100 million to develop.
@blackwidow76 and the rest of you that complain. I think you should leave Cursor behind and don’t return. Whiners contribute nothing to the product or to the general community around it.
2 Likes
May I suggest we keep our eyes on the prize, i.e. a world where:
-
Everyone has shelter, food, good health and relationships
-
Our creations heal, sustain and inspire
Right. So paying customers discussing issues they’re having with the software. I think anyone paying for a product should also be welcome to discuss their problems with it. Granted some of those posts can be more constructive than others, and some can get a bit more emotional or aggressive than others, but don’t expect that whole “stop complaining or leave and don’t come back!” mentality to get anywhere with people, either.
I’m sorry, I have to agree, I understand the frustration. While the testing period works fine, I subscribes now to the anual plan (yes, sorry, I’m stoopid) and they seem to change someting, so it becomes absolutely unreliable. Applying changes often ends in a error message, sometimes it ruines working code.
Sorry folks, I want my money back or this needs some yt videos showing what is it like working in real life with cursor.
Of course you should be welcome to discuss the product. But, just as you mention, swearing and being aggressive in forum is never okay, especially not when you pay peanuts for something delivering a great deal of value.
I think people are frusturated that we can’t get an insight look at what cursor is actually sending the LLMs
they add prompting from what I understand. so it could be possible the actual cursor prompts are not being used efficiently
I am just about done with this failed experiment. I am tired of fighting this idiotic AI to try and get things done. It would be just as effective to find a discarded toaster in a dumpster and have it write code. I feel like we have been duped into paying for a beta, because this is clearly not ready for prime time. How do we cancel our subscriptions and get a refund?
I think all of you that are “breaking your keyboards” out of frustration have not obviously been coding “the old way” - if you can’t handle the stress, don’t do it. Life is short. Yes, it’s one thing to have a bug vs a tool you use and have no control over malfunctioning. But hey - you are not a brain surgeon operating on your son and the operating device stops working mid surgery, ok?
I view Cursor as a tool to improve my UX. LLMs are here to save me from typing the code myself. Sometimes they really solve a problem I can’t. 3 years ago I would pay $2k a month for 10% of what I get now for $50 per month.
So I get the frustration - but writing elaborate posts on how the Cursor team baited and switched their users and how they must have done this to raise a lot of money - c’mon dude… If you are that frustrated - go do something else.
Btw, I advise anyone to watch Lex Fridman’s pod with the Cursors guys. They are young geeks and seem likeable. I’m rooting for them to make Cursor even better and they have my support and money. Good luck!
2 Likes
This is great advice! I’ll just add (for the non-coders here) - leverage git and branching. Learn to use git as a pro. Before doing any new feature - commit before hand. Branch out and develop on a branch. It’s crucial to not get frustrated. I sometimes make 2-3 branches and try to achieve the same but with different models. Then merge back and repeat. That’s the practical and calm way to iterate quickly even if you know LLMs will mess up 20% of the time. sometimes just repeating the prompt will get you a better result. Remember they are non-deterministic models.
Happy coding guys!
3 Likes
Hello Everyone
I’m Using Cursor at all generally in his usage.
Let’s say a bit, about Composer:
I do guarantee you allso, that Composer is a mighty tool, that eats more context,
as any - literally > ANY < - Ai agent has it done for me ever before.
so what we can say over cursor 'n composer at all:
Frustrations ?
Frustrations come and go,
that’s right, persistence in bug variety’s are unlikely.
but the cursor team said and will reply often that, because they working harder as we think. and the idea to put an > On Window > output editable ai editorian is a awesome, more like a brilliant idea you had done dear @cursor team 
really love it. Finally ive found my Devplace and Tool where i can get my productivity more done.
Thank to you guys !
4 Likes