I’ve been asking for pinnable & starrable chats for a little while now. This seems like what you might be talking about. Something comparatively simply, and extremely useful if it were implemented.
As a pure vibe coder, I quite like the AI Pane–I just think it needs improvements. As for 2.0, I think they either rushed it to deployment or they are totally blindsided by all the bugs we’re experiencing. Or both.
Just know–Cursor, we love you. A lot. We’re just a tough crowd since we’re paying customers.
As for me, I’m camping out on 1.7 until a bird tells me that 2.0 makes sense.
I really the hate the old AI Pane. It’s too small and makes it really hard/annoying to manage multiple agents running in parallel.
For a few earlier beta Cursor 2.0 releases, we had Agent and Editor as two completely separate windows. For me, that on a 49” monitor was the perfect solution. I can still dig into code in Editor while keeping track of Agents.
Hey John! Thank you for this feedback, as well as others here.
I can definitely empathize with your feelings around wanting things to remain editor-centric versus agent-centric. I also get very attached to how my editor is configured, because it’s how I’m getting my work done, and I want to have a lot of control over that.
Something that was really important for us with this release was to not disrupt existing workflows. If you prefer the classic layout with the file tree, that is great and we want to keep that for you! That’s part of the beauty here, the agents view is just an abstraction on top of the rest of the Cursor editor. We want to keep all of the advanced power for those who choose Cursor to get their job done every day.
As more folks start coding with agents, we’re continuing to evaluate the best way to bridge this gap. But it’s very very important to us that as we do this, we’re not leaving behind those who have committed to Cursor for using Tab or other editor-first features.
We’re really proud of this release, having not only added new features, but critically also made many improvements to memory usage, performance improvements (like LSPs), and other quality-of-life and UI polish fixes.
@Pablo11 20 prompts to get a single line of code? I would be interested in what you were trying to have it do. Cursor isn’t perfect, but it should be able to write a single line of code. If it couldn’t, why would any of us be here. Not saying its your fault, just wonder what your circumstances were.
@vibe-qa Name checks out. What kind of work do you do? I think some types of work are more practical for vibe coding than others?
@JAMES_OLAWALE What elements can you not attach to the chat input field? They did remove active tabs, which some people liked and some didn’t. Dragging the tab manually or just in your prompt saying “for the active tab:” does basically the same thing (or creating a custom command).
I also think the agent is the next step in future. Code is fading to background while planning and testing is taking more role in human coder’s life. I hope you guys keep improving the planning features, as they are now quite limited.
yeah, that ai agent is absolulty terrible.
is there an option to go back to the old one? the new one is so unusable, that i currently have to use other tools
I only just updated it, but for now I am just going to use Editor mode and see how the agent performs vs before.
I had a quick test of the new Agent mode and it’s not for me, I use agent only when needed and tab autocomplete. I like to code myself so that I can really understand the code. For some developers who use agent for all their coding then yes this is a great new feature, but for us who use it alongside then the editor mode is there and I am hoping that this will never go away. Cursor is a great tool, I just hope it continues to serve developers who only use it as a supplement to their coding needs.
I think people tend to post more about their negative experiences than the positive so I wanted to write something to the former to counter balance.
As developers, we are/will be shifting more towards age tic coding in the future wether we like it or not. Wether that’s screaming with resistance or embracing it is up to each of us.
So I think this is a great step in the right direction. I’ve been using it for a couple of days and I like it. I havent had many issues that others report ( thought not dismissing them) and I really like composer agent - it’s super fast and has worked well for me for code changes. Maybe I’m just getting lazy but I now use agents for almost everything. I barely touch code. And having a fast tool that can do the nitty gritty is great!
Keep up the good work. I was actually going to cancel my subscription this month as I’ve been using codex for a while and it’s crazy good. But I’m reconsidering now.
I like v2.0 as well. I don’t mind the Agent view because I just don’t use it.
I see where you are coming from with this statement, but I don’t think that’s genuinely where we are at right now. There are many reasons why people would not want a purely agentic workflow:
personal preference
the project requires a human to be more intimately involved with the code
lack of model training on certain frameworks or industries
cost since some people can’t afford to have agents do all their work.
The Editor view is a great hybird for people who can’t be 100% agent for one reason or another. The industry may be heading towards 100% agentic workflows, but it is at different rates depending on many variables. Cursor is trying to play to both audiences, which is smart, but I don’t think its as clean cut as you say where people should just be aiming for as much agentic workflows as possible, and any resistance is looked down upon.
I assume others can agree, there are times where I can produce code manually faster than walking some of these models through several iterations of the problem, where by the end I still need to take time to understand what the model wrote to review, maintain and ask follow up questions and correct any design flaws. The agents 100% help, but anyone who can do all their work with agents either has a lot of money which is a fine strategy, or doesn’t mind some sloppy code, or is developing something that AI can spit out with high confidence because its heavily trained on it and in a more saturated industry (like web development).
Fair points. I agree things are shifting at different rates and I don’t mean to look down on people with resistance. I had plenty of that too. I was more referring to the long term view that as models get better and cheaper, then more work will shift to agentic and trust will gradually build with the agents. We are in a highly experimental and new phase at the moment so of course the editor is still important and I use it plenty. But keep it in place while investing in agentic flows and features is future proofing the product for whats to come. I think cursors doing quite a fine job of bridging that gap between the two. Granted there’s bugs and not every release is perfect but the intention and direction I think deserves a praise.
Rust everything, frontend, backend, microservices, image generation, neworking, APIs, cryptography, webservers, except where not applicable such as embedded, mobile apps, bash scripts and browser extensions.
Explorations such as personal generative knowledge “grokipedia” library, microcontroller firmware, Zig migration, SIMD and web scraping have been very frustrating and expensive.
Some say around this time 90% of the code is written by AI. For me this is false, 99.999% of my code is written by AI.
The main problem is I keep clicking the new buttons looking for FILE - when you click it, all your editor windows are closed, and when you go back they have to re-initialise, and if they were showing webviews from extensions - they get reset. So frustrating!
I wouldn’t go as far as to say it’s a step in the wrong direction. I think it’s just too early.
I mean, I’m pretty sure there are lots of use cases where fire-and-forget agent-only mode works great, and where it is not necessary to steer the agent very deliberately.
But personally, working with a monorepo codebase of several hundred thousand lines of code, it often feels a bit like a slot machine when trying to let the agent do everything by itself. So, curating context is usually still one of the best predictors of agent performance afterwards. It is still possible now to drag files into the editor (which I do). But previously, I just opened all the tabs, described the task, and then added all open files via a single slash command.
I like the multi model death match mode. But PLEASE move the open worktree in terminal from menu item to a dedicated button on each model.
This is probably user error but I can’t figure out how to use embedded browser inside cursor. When I use @browser it fires up a chrome window like selenium, puppeteer or playwright. Still cool I had to use mcp before but seems basically just like an mcp short cut + now showing the screenshots in the stream. I’ll go do more homework on it.
Whenever you come up with new features or ideas, make sure to get user feedback first before implementing anything.
A lot of companies skip this step and end up building based on developer assumptions rather than actual user needs. Think of it like Ocean’s Eleven, before the real heist, they replicate the entire infrastructure, test it, get feedback, refine, and then execute. That’s the mindset we want here: simulate, validate, then build.
This will help ensure what we build actually matches how people use it — not just how we think they should.
Today was the first time I joined the forum to share questions and frustation… I’m happy to see that it’s not just me going nuts about 2.0.
You had a product that was already validated: helping developers (not vibe coders)… You just had to make it BETTER (less errors, more context comprehensive), not fancy… You made it worst…
I’m joining the crowd. Yesterday I had to use Cursor and it got lost in basic things, specially due the lack of context (“@”)… I closed Cursor and went back to Intellij + Copilot Free, and got the job done in a few clicks.
Right now: downgrading to 1.7, and thinking about trying Copilot or other options.
Tip for developers and product owner on Cursor: Whatsapp was not the first instant messager, but was the first one that JUST WORKED RELIABLY. You don’t need the be the first to launch fancy stuff, you just need to be the best.
