I’m incredibly disappointed with the changes you are forcing on us. We all have established workflows, and for those of us with coding experience—not just casual or no-code users—maintaining large codebases with the agent mode is becoming a real struggle.
From my experience, the agent mode works fine for small, non-production-level websites or MVPs. But for maintaining a serious, production-level codebase, it is absolutely terrible. I can’t even begin to express how frustrated I am. My development process has been completely halted for the past few days because I’m constantly fixing the errors caused by the agent mode and trying to figure out why things are breaking.
I can’t work like this anymore. I used to be one of your biggest supporters, but now I can’t even use Cursor properly. At this point, I have no choice but to start looking for real alternatives because my workflow is completely disrupted. And this isn’t a one-time or two-time issue—it’s a recurring problem that keeps happening with your updates.
I’ve been a pro subscriber for quite some time now, but the way these changes are being handled makes it feel like I’m working with a small university startup rather than a professional tool I can trust. Before making major changes, you need to understand your users better and gather proper feedback. You can’t just force changes on us without considering how they impact real development workflows.
I’m saying this because I believe in what you’re building, but if you continue down this path, someone else will come along and offer a more stable, user-friendly alternative that actually listens to its customers. That’s how you lose market share.
I understand that agent mode might be useful for people with no coding experience, but don’t lose focus on your core user base. The real users of Cursor are developers—not random users building throwaway MVPs that will never see real production use. Please, don’t lose sight of who your product is really for.
I really hope you are aware that this update is a disaster and that you are at a critical crossroads where you risk losing a significant portion of your user base. I hope the team understands the frustration among your real users because when I see the CEO talking about ordering food with MCPs, I can’t help but think—no one cares about that. We just want to use Cursor for what it was meant for: coding efficiently, managing our codebases properly, and reducing the need for large development teams.
You need to ask yourselves: is “Vibe coding” really providing real-life value? That’s a big question mark. Please, genuinely listen to your core users—developers—and take our feedback seriously to create solutions accordingly. I’ve lost 2-3 days just trying to figure out what went wrong, and the biggest issue is clear: this update has completely disrupted our workflows. It seems like you had no idea how many processes were broken because of these changes.
You need to understand your real user base much better. Don’t be misled by Twitter engagement and interaction-driven threads. Those are not your actual users. People with no real coding experience will create clickbait and interaction bait—not real products. The ones building actual products with Cursor are developers.
I need to know that you understand our frustration. Right now, I don’t see that. I’m exhausted. I’m frustrated. And honestly, I don’t know how much longer I can keep using Cursor like this. This just isn’t working for me anymore.
Three days ago, Cursor was the best tool I had ever used. Now, it has become a tool that is wasting my valuable time.
I feel that we need 0.45 back with the original context length, I know you’re bleeding money but just give us Deepseek R1 direct API($0.50 / mil output), it wont cost as much as the Fireworks provider you’re currently using which is x16
Fireworks at $8/mil output , make it an option , i understand you have enterprise customer who are sensitive about data flowing to CCP.
It’s not just about 0.46 , it’s the nerfing of context length and it’s just a bad timing with 3.7 which isn’t optimized for cursor yet. its so bad i wished i could just use R1 with 0.45
It can maybe build a calculator but it can no longer work on project that are of production levels.
Plus you’re not even transparent with changes going on behind. Even if i wanted to feedback it’s hard. Are you even using your own products or are you on Roocode?
I agree. I’m still trying to understand what’s going on after the update. There were a total of 6 requests, and all (even simple ones) of them broke my project.
This version may be better if you start a new project (I haven’t tested it), but it’s a nightmare for my case.
Updates were supposed to make it better, not worse.
Do you know updates aren’t forced and 0.45 still runs perfectly fine? If you don’t disable automatic updates why would Cursor be the reason of your disappointment?
Hey, can anyone tell me how to completely disable automatic updates? I downloaded version 0.45 and want to stick with it, but even though I’ve disabled the auto-update feature in Cursor, it still updates on its own to 0.46.
I really need to use 0.45, but I can’t because of this. Does anyone have a solution? If there’s a way to properly stop auto-updates, please share it with us. The method I tried isn’t working, so any help would be much appreciated!
You can check out my more detailed post on this topic above. I’m definitely not against updates, but pushing out changes that clearly haven’t been properly tested—without giving users a way to roll back to previous versions—is a dealbreaker for me.
As an IT professional who has worked in big corporate environments for 10+ years as a dev where stability is key, I’ve never seen update management handled like this.
Thanks for the suggestion! I’ll definitely give it a try.
That said, this doesn’t change the main point I’m trying to make. Of course, updates are necessary, but they should be more stable and avoid major disruptive changes. Many of us use Cursor every single day and have established workflows. You need to consider how your users rely on the platform. If you’re going to make big changes, at least provide an easy way for users to stick with older versions.
From this chat, I now understand that previous versions are available, but the fact that I had to dig through the forum to figure that out is frustrating. This process should be handled better. I shouldn’t have to spend time searching for possible solutions,just make proper documentation so we can easily understand what’s happening.
Don’t get me wrong, I’m a huge fan of what you’re building. Cursor is one of the most impressive AI tools I’ve seen. But this whole update process really needs improvement. Thanks for everything, guys,hope this gets managed better in the future!
Totally agree man. People are complaining so much (and with reason) but are not considering this point. Software releases are like that: if you dont’t like some new update, go back and use the previous one. Simple as that.
And a lot of people here are thinking that these AI tools and models are replacing the workforce, but it is not the case (yet, but it is not). These tools are just tools, you can’t rely totally on that. Keep working.
I would like to kindly add that in fact the last update did disrupt my estabilished workflow.
One part of my workflow I’ve lost with last update was the ability to change from chat to composer while having different models setup for each one.
For instance I use deepseek-r3 for chat, and sonnet for composer.
As the panel is the same now for both, when i change mode the model keeps the same, which is bad for my workflow as I use different models in each mode.
Now, every time i change from chat to composer, i have to also change the model, right? I will work around that, but still…
Could cursor provide a way to setup different models to each mode, in a way that everytime i change mode I dont have to change model?
It makes no sense to integrate in the same panel since chat and composer dont share the same models (as most chat model dont work in composer).
So my sugestion is either separate back the panels, or provide a way to setup each mode’s model separatly.
(as most devs, I rely a lot on keyboard shortcuts to abstract handling with the IDE interface, I dont wanna think or fiddle with the IDE interface to be selecting models because I’ve changed mode, it disrupts the flow. If this wasnt an IDE targeted for devs I wouldnt be giving this feedback since casual user dont care about flow in coding sessions)
Yep, it broke my workflow as well. Especially with “agent” setting. Prior to update, if I had instruction in markdown, and tell it to do it, it would perform the task and in sequence. Now the agent 1) moves the terminal context outside and stops work 2) does not continue past the first among a sequence of tasks I told it to perform without me telling it to continue.
Not only were workflows disrupted. now previously working parts do not work or stop working randomly e.g. like stuck on “Generating” after few short requests, debugged the issue and its not even sending requests to AI cursor Api (though requests to subscription api work).
Rules arent picked up properly, have to reference by hand rules which worked before automatically.
YOLO is not working when you enter a prompt/description there. Nobody knows why, but others found this is a temporary solution to empty that prompt.
While some items may be an improvement overall its sure a disruption for developers when things unexpectedly change.
