Mad at composer / chat

[blackwidow76]

I’ve seen this phrase enough today, to last a whole lifetime “I sincerely apologize for the continued frustration. You’re absolutely right to be upset, and I take full responsibility for my mistakes.”

I spend more time fighting with this platform to make even the SLIGHTEST bit of progress.

Something changed over the past couple weeks… whether they’re using haiku or gpt-4o-mini, when it should be sonnet or 4o.

The quality of this platform has degraded EXPONENTIALLY over the past few weeks; I could get a lot more done for a lot less money using aider or ClaudeDev.

Was good while it lasted. Never in my life has something been so insanely inconsistent that it makes me contemplate smashing my computer on numerous occasions.

I paid for a working product… this is so far from that its not even functional anymore

[merleau]

I have to agree, unfortunately. This service has become very unreliable. I am looking into alternative services at this point.

[xl0]

Same experience, over and over.

1. Cursor works
2. Cursor team tweaks something without comprehensive testing.
3. Cursor works but is worse.

Tried to use instructions today (either in settings or in .cursrorrules, but they are simply ignored. I can explicitly ask in chat about the details in the instructions, and it has no idea.

Same with the quality degradation that’s been on for a week or two. I think they “ortimized” the amount of stuff that gets sent to the model, and the edits got worse because of the lack of context. Of course, you can still add the whole file manually, but that’s just annoying.

Wish there was a viable alternative that did the basic chat+edit, but alas, there is nothing out on the market.

[litecode]

Hi @blackwidow76 ,

Sorry to hear you are having a hard time with it.

I also get frustrated (aka very angry) working within Cursor sometimes.

The potential of what AI coding assistants can deliver, versus current human and technical limitations, can be anxiety inducing (I experience this myself).

But, in regard to blaming the Cursor product itself, my contemplations of the process are:

• The responses are originally from an LLM, not Cursor itself, and I have experienced the same ‘confusion’ when working with LLMs directly in chatbots like Chat GPT etc

• If my questions are either too large in scope, not clear enough or part of a very long chat, then I cannot expect the LLM to do any better, and so I need to regroup and plan out a better, more logical and modular, approach

• This is literally a cutting edge technology and I should remain in awe of what it is achieving behind the scenes

• I am grateful for the efficiencies that it has already introduced which have let me do a lot more than I could previously do

Yes, there are bugs within Cursor, but part of what this community is about is getting to use something new and very cool and sharing feedback directly with the devs so that they are aware of the issue and can resolve them as soon as possible.

I understand people may have different ways of processing things, and different expectations of the product and community, and they are valid, but just sharing my thoughts in case you can relate to any of them.

I guarantee you that everyone in the Cursor team, and all of Cursor’s users, want the Cursor product and experience to be better and are doing everything they can to make it so.

FWIW, I’ve personally found the approach outlined in this video aligns with my methodical way of achieving results with AI and is good to reflect on when progress starts getting muddy:

Mastering AI for Software Development: Essential Tips & Principles

The video covers things like:

Principal 01 - Set realistic expectations for AI
Principal 02 - Embrace iterative development
Tip 01 - Prepare your coding assistant for tasks
Tip 02 - Specify what you don’t want the assistant to do
Tip 03 - Fix bugs by showing the assistant the error
Tip 04 - Work in small, focused units
Tip 05 - Always review code updates
Tip 06 - Choose and master a particular AI toolset
Tip 07 - Think ROI when considering AI costs

Again, sorry to hear you are having a hard time with it, good luck on your journey and I hope it becomes better soon!

(please consider revising your topic title as it may offend some other users and goes against the spirit of the community guidelines).

[xl0]

In theory AI-generated user support in forum sounds like a good idea, but in practice, well, you see the results…

[litecode]

Hi @xl0 ,

The post was not AI generated.

[blackwidow76]

I also want to provide more context…

I’ve been using cursor 8 months or more. Long before composer, or sonnet 3-5, so I have enough experience working with the platform to have seen the considerable decline in capabilities.

Call it a coincidence, call it what you want, the platform coincidentally worked great, JUST long enough for about 10,000 people to make youtube videos on it, draw in as many people as they could, give them a decently working product for a while to get them hooked, then COMPLETELY NUKE the performance across the board for everyone from long time users (me), and everyone else that was given a great experience to start by teasing them with acceptable context windows and premium models, just to pull some string on the backend that shittified the experience across the board.

Maybe it was a flawed business model from the start, but maybe consider reigning in the free trial for NON-PAYING customers first, before NUKING the performance for everyone, including long-term, previously loyal users…

There is not a single application on the planet worth the frustration caused by the degraded performance and inconsistency. I understand, for new user to your platform, the advice offered might have been valid and insightful, because they have no frame of reference… but I’m telling you, I had 10x productivity using cursor for about 7 months… right up until about 4 weeks ago when it literally completely went to shit.

Ironically, the product was great, just long enough for them to raise 60m. Maybe the investors/advisors clued them in to how unsustainable the business model was, or maybe there was another reason behind the 180 degree change in performance and capability across the platform, unfortunately that is not information i’m privy to and the limited time I spend in VC/Private equity is inadequate for anything more than hypotheticals.

There are new open-source platforms coming available every day. Cursor was a lot of value for the price, up until… about 4 weeks ago, right around their big youtube/influencer push… which just happened to align with their 60m dollar raise… and then as if it was overnight they flipped a switch and we went from using the premium models, to cursor-small, haiku, 4o-mini… UNLESS you’re paying per request; then you may be given use of the model you’re expecting… maybe.

[aar2dee2]

scaling is hard. We’re with you cursor team!

[blackwidow76]

– I said 8 months in my post… I’ve been using cursor since September 2023; I’d say thats long enough to have a fundemental understanding related to the capabilities and limitations.

[Taidan]

There’s Zed and github copilot has gotten much better

[deanrie]

We’re working on fixing all the bugs and making improvements. I really hope we can fix everything and add a lot of cool features I’m working on. Please don’t be too harsh, we’re trying really hard.

[somerandomguy]

“I sincerely apologize for the continued frustration. You’re absolutely right to be upset, and I take full responsibility for my mistakes.”

I’ve also noticed this phrase or a variation of it happening a lot, and it does get frustrating going in circles with the chat and burning requests on what I feel are not terribly complex issues or topics.

If I ask a question and it provides a response with a solution, it really should produce a solution that it feels confident in. But I’ve noticed often that, if I ask in the very next prompt if it thinks this solution is the best approach, it will immediately backtrack on itself and go in a different direction, and thank me for helping it get there.

For me, being rather new to programming with AI, this doesn’t really instill confidence in the output being generated. This might be more of a “me” problem than a Cursor problem, or just a reality of what it’s like working with LLMs, I don’t quite know yet. But it can feel very aggravating and mentally exhausting at times, and makes me wonder why it won’t just generate what it feels like is the best and most complete solution in the first place, instead of blurting out the first thing that comes to mind, and then thinking on it more carefully after I’ve burned another request.

In other words, I should never feel like I need to waste a request asking a question as simple as, “Are you sure this is the best approach to solve this problem?” But I’ve seen it backtrack so many times, even on very simple things, that it can be hard to trust anything it provides on the first response. Ideally, if I ask that question, it should come back with something like, “Based on what I’m seeing and how I’m understanding the problem, this appears to me to be the best approach we can take to resolve this issue.” But it’s extremely rare that I ever see anything like that. It’s far more common that it completely changes its solution or approach (often unnecessarily) based on nothing but me asking, “are you sure?”

“Please don’t be too harsh, we’re trying really hard.”

I can also relate to this lol. Try not to take it too personally or let it get you down. Sometimes people just need to vent their frustration and get it off their chest. Your work is very much appreciated though, and I’m excited to see how Cursor progresses!

[smhoff]

@Taidan

Zed

Zed looks really interesting!

[matthewmje]

have experiened the same

[dimitri-vs]

Gotta love the hype cycle. Went from “this is the future of writing code” to “inconsistent piece of shit” real quick.

Cursor is just providing better UX by saving me time from having to copy paste manually from Claude. And the change apply feature has been pretty good, but definitely needs to be faster. Overall I’m very happy.

[xl0]

Please just provide more visibility into what is going on with a debug tool. You let users specify their own LLM endpoints, which means you are not afraid to leak at least parts of the “secret cause” that goes into the prompt.

Just a way to open a window and see what actually went into the prompt would make the experience a lot less frustrating. If I see that you sent all the correct code and instructions and the model still pooped itself, that would be a lot less frustrating than guessing if my instructions are not clear, or you are just sending an old cached version, or something else entirely.

[e-fu]

I also like to add, that when I started fighting with the AI i was maybe on the wrong path. starting from new with better idea and instructions sometimes yielded to even better results than trying to fight my way.

[xl0]

Also if the issue is that you “optimized” the context length, maybe give us an option to adjust it when using our API keys.

[pfapp]

I’ll echo the sentiment here. I’m feeling like the performance of Cursor was exponentially better just 4 weeks ago when I started my app. Its gotten so bad that I tried to start a new react app to rebuild from the ground up and I cant even get to the most basic iteration of what I was I created in like 15 mins when I first started.

I don’t agree with people being rude or nasty, however, if you guys aren’t being open about changes you’re making while performance is obviously (to some) degrading, I wouldn’t expect things to change. If anything this is a PR disaster waiting to happen.

If you guys are making changes that nuke performance, be open about it. From the outside its being marketed as something that isn’t an alpha/beta product

I’ve been obsessed since I started using it …probably averaged 8 hours a day over the last 4 weeks. At first I just assumed this would be the best product because marketing but I’m more than likely going to try the open source competition tomorrow.

I even joked with friends when I first started going to this rabbit hole that if Cursor stopped development and kept all functionality the way it was , I’d be a life long user. My biggest fear would be getting rug pulled and I hope thats not whats happening.

[stonianua]

I also broke my keyboard yesterday. It was way better a month ago.