“who actually types anymore?”… The people staying with Cursor who are building things that still require a person to design, innovate, review, and often still code. Most people moving to Claude Code are building very common applications that AI has trained extensively on and its only a matter of time the “prompt guys” will just be replaced by AI or whatever they are building will just be flooded with competition because you can build it so easily with AI. But yea Claude Code and Cursor are entirely different tools, and its clear Cursor is the wrong tool for you because you don’t need it.
No, they both just SWE tools with some pros and cons.
Pro on Cursor is: all models, and tabtab.
I do.
We’re not even building something complex and unusual. Just another CRM and just another personal cabinet for clients. Nothing fancy.
Still, AI can’t do ■■■■ about all the bizarre business processes, all the undocumented legacy and project knowledge living somewhere between our work chats and our brains… And it can’t be trusted, even if it could handle all that without human effort. It’s often much faster and safer to just do it manually with Tab.
Agents are still a handy tool, don’t take me wrong. It’s just that for professional work autocomplete IMO is more essential and useful atm. And Cursor still has the best autocomplete, it’s not even close.
1 Like
Auto mode is an absolute digrace lately, this is UNUSUABLE. Get stuck in loop, delete scopes that it shouldn’t delete, use really bad practices. Its getting worst and worst.
Now there is 2 solutions left : Use Claude Sonnet 4 –> very expansive
Use grok code fast 1 → good and fast but dont use best practices, so its a bit tiring to correct it all the time.
Also gpt-5-mini which as of now doesnt apply to monthly usage.
Did you really renew your annual membership for the auto mode? Don’t forget, the Cursor team can change the billing rules at will.
Not just for Auto, but also for the $4/mo discount and to lock in if they increase any prices. I am satisfied with the product, and am using Auto less anyways with cheaper/free models now available (gpt-5-mini, grok-code). But I still get use out of Auto, esp with boilerplate stuff. It’s fast and just works. I am not worried about not getting my $16/mo worth out of Cursor, and they seem open to giving refunds if you ask, so worst case I get a refund if they change policies down the road. Best case scenario, I lock in $16/mo for a year and get unlimited requests with Auto inside a tool I have grown accustom to and get $20+/mo in credits to higher end models. If you need just an AI programming assistant, for $16/mo cursor is a no brainer decision. If you need tons of prompts for vibe coding, and are considering Ultra or something, then maybe Cursor isn’t the best value. Depends.
Once we understand the power of LLMs - how they work, how context is processed - then you will realise that Auto is just a bad idea for doing meaningful work. Every model has its strengths and weaknesses, when you use Auto - you are randomising to the most available model.
I’d suggest creating a custom mode, select your model of preference - go to that model’s documentation and see what works best for that model and set custom instructions for it and a role for it. This should give you the best the LLM model has to offer.
2 Likes
Paying yet more money to a company that can’t decide how to charge to gain access to something “free” seems counterintuitive. Cursor is an editor, it really shouldn’t need this much monitoring drama to know what I’m being charged.
I just canceled, even though I had 500 requests, end of the day between claude code, and work paying for ChatGPT, and access to a service with qwen coder, just not worth it. And has anyone checked on tab complete? I’ve been disabling it, keeps trying to delete huge swaths of code.
i just did, no other product on the market makes sense atleast for my usecases, i need Cursor honestly!
1 Like
Yea, lol why would you keep Cursor if you had CC, chatgpt, and qwen coder. I don’t think the point you were trying to make was made… Cursor is so good, that only after you have access to three alternatives, is it worth then canceling from.
Yes, tab complete works great. I do disable it temporarily when I am doing a lot of manual work, but generally it does help.
How would you describe your “need” of Cursor and why do other products fall short?
You do a lot of defending of cursor, I get if you’re a huge fan, some aren’t, some where big fans and now are absolutely not (like me). But if people can only make agreeable comments about a product maybe this place needs to close the forum.
You don’t need access to three, only 1, (work was reimbursing me for cursor, so they pay for chatGPT now as I find using the cli better.) tab complete is no longer great it was the reason I kept it for so long, now its suggestions are delete what I just wrote, delete 5 lines of code, suggest changing the code to something else (suggest python in C# code for some freakin reason) or make pedantic whitespace changes. Sure, it still would pick up some changes and make suggestions across the entire file but it also gets in the way more then it helps and escape key, or continuing to type, aren’t enough as you just get a stream of suggestions as windows open over your code repeatedly.
add in the constantly changing drama over what the cost is, will be, I really don’t want to waste my time having to look back at the forum or blog for yet more damned changes that were quietly announced to hide they made changes.
Pick something else, go with Claude Code, ChatGPT5 (and use the VSCode plugin, or cli), use Kiro. Kilo Code, Roo, Continue, there are tons of them. Go try them all out. Cursor had good tab complete, that was it.
1 Like
The forum is like 90% people complaining and saying they are leaving. Cursor is allowing these negative posts to stay up, so I am not sure what you are getting at. I have my issues with Cursor, but I find the forum has become primarily a place to complain without anything productive out of it. You are explaining yourself well, so there are obvious a lot of exceptions.
I agree the tab complete can be annoying, and I hope it gets better or some way to control it better, but I have tried others (trae, windsurf to name a few), and I just don’t find that the IDE works as well, especially with C#.
So if I am understanding you correctly, you basically don’t use tab complete anymore, just cli? Are you doing a lot of manual programming or mostly prompts? I find the most people that complain about cursor are looking for solutions where they basically no longer program, and that is great, but not all projects or workflows can have that luxury. That is where Cursor surpasses most tools that are leaning more into vibe coding and not an AI programming assistant.
The price drama could 100% have been handled better, and it is one of my biggest complaints, but not sure what my options are when I try others and they struggle to even get c# working correctly because there isn’t a lot of supported extensions on these vs code forks. Cursor manages their own C# extension for their VS code fork, and the only reason they would do that is because there were none that were available on the extension marketplace for these forks. That’s just an example that I have noticed, but there are others that make Cursor better for those that are actively programming.
I wish but no, my day job has me doing a ton of different things only a small portion is writing code, so we’ve only started using LLM’s in testing. What I do write at work I mostly write manually, but using a cli I can let Claude document code which is historically poorly documented and hard to work through as it written by devs who didn’t think much of anyone else needing to do anything with it.
Cursor was okay with tab complete especially in larger files, but has been going down hill. So where I change a method, which needs a change in how its called/handles responses aka ad await which I had to do last week, the model suggested removing the calls, which isn’t uncommon but usually after the first manual edit gets it to understand what I want is to add something and make a change and it handled it well, but for the last 3 weeks no.
With Claude (and ChatGPT), I point it at the method, at the first change I made, tell it to find the rest and apply the same or similar change and it does it. But I’m initiating the work, not getting prompted constantly by a floating window.
You are right that the forum has turned into people complaining, but, I think it’s a bigger issue with Cursor, happy users tend to be quietly doing things they enjoy. When I first started using cursory (just for my own personal coding, at home, not work) Cursory was incredible, I started programming in the late 90s, Cursor was a pipe dream with no reality so using it felt like writing code was suddenly a dream where a helpful little clippy actually was helpful, it pointed out an error, something unclosed, suggested the variable types and remembered the name of it for you and realised you needed to call it here and suggested it.
Now, its a company that can barely communicate with its paying customers, the 1 feature I really liked, and felt made it worth the money has turned into clippy the unhelpful and worse still, to find out whats going on I have to go dig through a forum to find clarification on what the damned post meant that changes the pricing structure.
Kiro, owned by Amazon, just sends an email, I’m not even a paying customer I tried it out, they’ve given more information on the state of costs, changes, and issues, then Cursor has ever given, I don’t even use it and their transparency has been so much that I know exactly where it all stands. Cursory on the other hand, I might have been grandfathered into the 500 requests because of when I signed up, maybe I’m not and my time would have ended, or maybe I was having some grand dream, I haven’t a clue I might have missed a blog post or a forum post or a smoke signal I’ve no idea, it seems most of the users have no idea, and with tab complete going up in smoke, and figuring out how to work well with Claude and ChatGPT, its not worth the irritation of dealing with Cursor.
So people who are unhappy with Cursor should go use other tools, take the time to figure them out and how they work into your processes, I did, I kept cursor until I did, now I don’t want it.
4 Likes
I did but now I m kind of regretting it already . Ability to assist or resolve existing issues is degraded a good degree in last few weeks .
It writes 10 lines and all 5 new errors then floors to fix it and add more errors , re writing complete file is very common .
Added Claude code pro even with rate limits hit in one hour of 5 hour window it’s still get work done 10X faster than auto in cursor .
Using cursor only for documenting and figuring out small edits or error finding
Auto is a free model. Of course it’s gonna be less efficient compared to sonnet 4. Compare apples to apples. Call Sonnet 4 in Cursor. Also if you don’t need something to help you code, then just use Claude Code which isn’t based around an IDE.
I love when you guys call me a bot just because I don’t jump on your cursor bashing train. Why are you guys all hanging out on the Cursor forum anyways when you have found so many better alternatives? Gotta come back and make sure everyone knows how Cursor has broken your trust and you are so hurt by it.
Mentioning “Void” lol, come on man. Recommending openrouter free model api keys? You obviously are just a hobbyist looking for free stuff, and are just mad that Cursor cost money, not because it’s a bad product.
You wish you were on Cursor’s payroll like me. $$$ (this is sarcasm if you can’t tell. Cursor is on MY payroll. 
)
Proof I have been critical of Cursor:
Its always cool to have someone answering to any questions like you. BUT I also understand people coplaining and still using Cursor. I’m one of them, the reason is that when I joined a few months ago it was amazing (with Auto), and I’ve been able to achieve things that have been pending for years on my own. That’s why I got hyped and paid for the yearly subs. Now We clearly see that the quality is going banana and that’s why we are getting frustrated. We still hope to get that quality that we all experimented.
And yes of course, now we don’t have a choice, we have to adjust and use Cursor differently…
edit buttons - it was just a bug, a visual one, and they fixed it in the next update 
to-do lists work fine for me, it depends on a model and on your luck (as LLMs are probabilistic and we can’t really do anything about it)
pricing changes - aight, they really just promise one thing and short after doing the opposite. the changes are fair IMO but their communication was really bad and pretty unprofessional, I can agree with you
differences between low and high are listed in the model selection on hover - i.e. “low reasoning effort”, “high reasoning effort” and are pretty obvious for anybody who understands how modern LLMs work. for people who don’t - well, should they give you a f-ing crash course on machine learning then? “this knowledge is only known on the cursor forum when it first came out” - many of us knew this even before GPT-5 was announced as there were many other thinking models before in Cursor and in other tools and services where you could choose the reasoning effort.
auto mode misbehaving - well it’s a router, it can route you to a model that does the task poorly or doesn’t really listen to instructions. there’s a lot of room for improvement ofc, but for a long time auto were treated like a quick and dumb model that is “unlimited” (even if it’s not a model and sometimes it routes to “smart” ones). I wonder if they’ll route us to better models when auto starts count towards our usage…
claude models being dumbed out is probably not a Cursor issue. I’ve seen complains on anthropic subreddits about models degradation in CC as well…
so… you just seek validation for your choice to switch tools? that’s why you’re coming back to this forum over and over? aren’t you just an obsessed hater then? or did I misunderstand you and your real goal is to check if it’s worth to try and come back? otherwise, it sure looks like the prior…