In short: In my opinion, Cursor is a good product and I am grateful to the Cursor team for developing this tool!
I check these forums every once in a while to get updates and see community guides and other similar topics. From what I’ve seen for a few weeks now, the posts here tend to contain a lot of complaints. People are saying the AI is too dumb, this Cursor feature is useless, I liked version 0.XY more, why don’t they implement this feature I want, etc.
It gets annoying very fast.
I just wanted to say I’ve been using Cursor on Linux, with a Pro subscription for close to a year now and it’s been a very good experience.
It’s helped reduce the time I spend writing boilerplate, it’s helped me reason through complex coding problems more wisely, to find documentation for more obscure frameworks (e.g., GTK styling), and I’ve even used the chat for non-programming tasks like psychological research articles, generating Excel formulas, and so on.
It’s a very good tool, and it’s steadily improving alongside the AI models. In my opinion, many of the issues we’re seeing aren’t so much with Cursor itself, as they are with the current state of the AI models that are available. It’s a rapidly evolving domain with lots of breakthroughs and features appearing almost overnight.
I’m not saying Cursor is a perfect product, but I do want to offset the negative tone on these forums. It’s a good product and I am grateful to the Cursor team for developing this tool!
it is not about “I liked version 0.XY more”;
version 0.48 is not working
version 0.45 is STABLE
can you see the diff,
btw Cursor is the best Agentic solution on the market
I completely agree. Like many others, I genuinely believe Cursor is an exceptional editor. The rapid advancement of AI technology inevitably exposes various issues, and we’ve also witnessed the Cursor team’s quick responses to them. However, it’s equally inevitable that new challenges will emerge.
I’m convinced that everyone posting feedback does so because they want a better experience—no one wishes Cursor harm, or they simply wouldn’t keep using it. What we’re seeing is a community eager to resolve pressing issues, which isn’t a bad thing at all. But it’s crucial for both the community and the development team to courageously address these concerns head-on. Avoiding them would only lead to unchecked frustration, and that’s when real negativity takes root.
Agreed, this has been my goto IDE, I find I’m always keeping an eye on my model count which ■■■■■ but it I fill it and it gets slow there is always vscode and copilot ready to go.
I don’t think AI is getting dumber. I believe that with a good prompt, we get good results. A good engineer will be better at prompting, and AI is just a reflection of us. My concern is that Cursor removed useful features from version 0.45, like Copilot Edit, etc.
I also facing some issue while Agent just give me result without doing action in latest version which this is never happened in 0.45
The reason this is so frustrating is that these errors and bugs are costing people massive tokens without always being clear as to whats happening. I would LOVE a " enter token to continue?" pop up like old fighter games, so That i know when one is being counted.
I LOVE cursor, and don’t mind some bugs, but i work really hard on my token budget for the day lol.
Although I have seen many criticisms, this does not prevent cursor from being the best AI programming tool currently available. I hope the workers of cursor will continue to work hard and not fall behind in this great era.
Hey, just wanted to clarify, but for the last week weeks we now track errors and failed AI requests, and don’t take them from your usage, so you only lose what you use in usage / billing!
Appreciate all the love on this thread, but we still have plenty to do to keep building and improving!
I get you, but all complains are VERY VALUABLE feed back from paying customers.
Most of who complains are the ones who love cursor the most.
This clearly makes cursor better as this forum is a proxy to cursor dev team.
(after update 0.46 disrupted my stablished workflow, I’ve downgraded to 0.45 and disabled auto updates, it’s being pure bliss since then)
Cursor is really great ! Its just that people dont understand that its not a release stable version… its a beta version still .. there is still bugs and people dont like bugs they would like that everything works perfectly … which a beta isnt that lol…
This is one of the only forums I’m a member of where 90% of the posts are bug reports and complaints about the bugs.
But there’s a reason why. People aren’t making it up.
Cursor appears to be following the typical Silicon Valley program:
Gain as many users as possible, losing money on them if necessary, create hype, deliver a ridiculously good value so everyone talk you up.
Value engineer the product to squeeze more profit (or rather, lose less money) and hope users don’t notice the difference in quality.
Raise the price and use the dominant market position to force users to stick with the more expensive product because they have no other choice.
Every tech company follows these same 3 steps.
Cursor is currently on step 2, but the problem is users have noticed. We all noticed. Anyone that used Sonnet 3.7 on Cursor the first few days it went live had their mind blown at how astoundingly good Cursor was paired with Sonnet 3.7. The problem was the context window was a good size and Cursor was likely losing an enormous amount of money.
Since then, the team has done all sorts of tricks to reduce the context window while trying to maintain the same quality, and it has proven difficult. That’s what many complaints are about on here, and they are legit complaints. I think users get overly upset because these bugs and reduced context waste valuable time and money. This isn’t a social network, it’s a tool many use for important development. The many bugs that come along the path to completing step 2 are also frustrating and harmful to Cursor’s brand. But the frustration is real, and it manifests in the forum.
What could Cursor do?
They could raise the monthly price in exchange for a defined, longer context window. I would certainly pay $100/mo for 500 requests if a) the context was a good size, b) it didn’t do this new thing where it calls a tool 20 times to read different parts of the same file, c) doesn’t exhibit the editing and failed tool call errors that have been prominent lately.
They could partner with Deepseek and really engineer their agent to work well with it. Then offer it as a separate plan with either a low price or a large context window, as Deepseek’s fees are far less than the big players.
Wait it out for new SOTA models to come out that are cheaper. Though I’m worried that OpenAI, Anthropic, and Google may collude to keep the token pricing at a certain level and avoid a race to the bottom. Hi Deepseek.
I don’t have all the answers but the number of complaints in the forum should speak volumes to Cursor team on expectations and what needs to be done to turn sentiment more positively. When Cursor is on its game - like when Sonnet 3.7 first came out, it is an extremely impressive product. I hope it can get there again.
It doesn’t seem to be an issue, as there’s Sonet 3.7 Max, and I can confirm it’s fantastic! Whenever I have a task that requires a large context window, I switch to Max, and it often handles it effortlessly, even with a narrower context window.
Overall, I agree that there’s too much negativity, especially from people who don’t know how to code but want to build a complex product or feature for $3, then throw temper tantrums in every discussion, accusing Cursor of stealing money.
I absolutely like Cursor. There are multiple projects I wouldn’t have been able to start working on without it, and I can literally begin contributing within minutes. The more recent versions are also much more stable!
Of course, there are people for whom it crashes on startup, but that’s not the type of negativity I believe the original poster was referring to.
Yes Max is quite good, but it also seems to fall into the same pattern: for each new release it is shockingly bad as the prior stable version. Then it stabilizes. In the meantime, the developer’s project do not change. These waves of instability are deeply frustrating. The complaints are valid.
For example, V.47 + claude 3.7 sonnet MAX was unreal! With the new V.48 release it’s like we went back to the highly verbose and over-eager claude 3.7 vanilla (not understanding the code it just wrote, going on insane joyrides, overwriting a ton of stable code).
As others have mentioned there are forum issues. Im not dismissing Cursor bugs but things go way out of hand here. Perhaps thats something where Cursor can look how to improve too, though i see so much more answers from their team recently and very proactive which prevents the bush fires often.
Lots of separate posts on the same issue instead of searching: expect a single post to get instant support as if they pay a 24/7 private concierege, wastes Cursor staff time and delays handling critical issues.
Lots of ‘i have the same problem’ replies on solved issues without giving any more info or context: pollutes forum and is not helping the user at all as it reopens months old threads that often not related to new issues or the causes they face.
Many are reposting half baked imagined causes like cursor trying to rip everyone off or cursor being abusive
Yes non-coders but especially users not familiar with prompting face hurdles they contribute to Cursor AI is bad.
The list goes on but lets be positive.
There are also good things
Users collaborating on prompts, approaches, solutions
Users helping users with debugging or suggesting possible causes
Great variety of issues that were attributed to only Cursor are being reported by the user themselves as having been an issue with: extensions used, os level issues, networking issues, etc..
Its normal that people with less issues post less as they are not in need of an solution that bad.
A more positive community helps everyone: less frustrated users, more likely to contribute and innovate, the product which is widely used would get better initial impression as some are wary of subscribing due to such a prevalence of negativity.
For over a decade i used a top of the line IDE makers paid software which is now at best legacy compared to Cursor.
Dont get me wrong, i cant stand VScode and its flimsy unprofessional extensions, the lack of professional tools for specific programming languages and frameworks etc that the legacy IDE has. Thats not even a fraction of whats wrong with vscode. Having the choice i still use Cursor and appreciate the improvements.
And yes i would like it to be better including the services Cursor offers as part of package (Agent handling, tools, …)
