Cursor 3.0 and the agent experiences are additive. They’re meant to expand what’s possible, push the frontier, and offer a glimpse of where we think agent management is headed. However, the point of the new Agents View is not to replace the core editor experience you know and love. That’s not going anywhere. It is a primary focus and will remain so.
Hi George! Thank you for sharing your feedback. We are not dropping VScode or the IDE experience. The Agents View is a look at the future and is an option that exists along side the IDE editor. It’s still early in the lifecycle and development so expect the Agents View to evolve and improve.
There are a couple themes offered in the Agents view now actually, System, Light, Dark, and High Contrast. You can also choose a tint color and tweak the transparency etc.
There are in-line Diffs displayed in the Agents View as shown below. Let me know if you can’t see these in your Agents view.
Yes, but if you’re relying on the Prettier extension, it doesn’t work right now in the Agents view but as a workaround you can have Prettier run as a pre-commit hook, and this would be respected by commits in the Agents view (you may need to install Prettier differently, I know there are a bunch of ways you can install it).
IDE is still a primary focus. Agents view exists alongside the IDE
Hmm, I don’t quite follow. Can you share a screenshot or give more detail? You can hide the sidebar with the button next to the magnifying glass in the upper left-hand corner if you don’t want to see the conversation sidebar.
Multi-root workspace support is being actively built out now for Agents view.
Vscode extensions currently not supported in Glass but there are lots of options for example with Prettier you can just set up a precommit hook and have Prettier run before you commit.
Themes are limited but there are a few options you can choose from and some color / hue/ transparency changes you can make in Agents View settings.
I don’t have a public answer to share for the plugin question yet, so I won’t comment on that at this time.
Previously, I reported some issues here:
Many of them have already been fixed in the new version, such as “not being able to see which modified files are uncommitted” and “having to click twice to enter tabs of different categories.”
There are still a few issues that feel inconvenient to me:
-
The tabs opened on the right side are isolated for each agent conversation.
In other words, if I open files A, B, and C in one conversation, they disappear when I switch to another conversation. I understand the intention—each agent represents an independent workspace. However, in many scenarios this is not very convenient. For example, I might run one agent to execute tasks while forking one (or even several) others to discuss certain issues. These agents often focus on similar topics, so the tabs on the right are effectively shared context (at least from my perspective). Having to reopen those files every time I switch agents is quite frustrating. -
There is no way to fork from a specific message.
The new interface only allows forking the entire conversation (via the top-right corner), but I strongly prefer being able to fork from a specific response. Whenever I’m not satisfied with the current direction and want to try something new, I usually pick a node in the conversation and fork from there to explore a different approach. This was easy to do in the old interface, but is very difficult in the new one. -
Clicking a subagent opens it in the agent panel and covers the main conversation.
This design is extremely, extremely frustrating. Normally, when an agent calls a subagent, I might click it just to see what it said—nothing critical, just out of curiosity. In the old version, I could view the subagent content on the right side, glance at it for a few seconds, and move on without any disruption. I would casually check each subagent.
In the new version, however, clicking a subagent replaces the main conversation view, forcing me to click again just to go back. This extra step is annoying enough that I now avoid clicking subagents altogether, let alone casually checking each one like before. -
One more thing that might sound a bit odd, but I do feel a difference: the input field in the new version is too low.
In the old interface, there was a status bar at the bottom, and the input box itself occupied an extra line, so the actual typing position was roughly around the third line from the bottom, which felt fine. In the new version, since there is no status bar and the input box has been simplified to a single line, I constantly have to look at the very bottom of the screen, which feels tiring.
Imagine if, when using any app on macOS, you had to keep your eyes fixed on the Dock at the bottom the entire time—it’s a bit exhausting.
Yes sorry, I see I can hide the sidebar. But just consider a wide screen, where I want to have 2 open Cursor Windows each working on a specific project.
3.0.13, don’t see any update / ways to update to the Cursor 3. Settings from the video/documentation are not present.
Currently, after completing plan mode, it is not possible to create new chat windows and execute the corresponding TODOs from the plan within a fresh agent window context.
Love the concept of the new UI but seems pretty buggy right now. I open a workspace from the recents list and it does not show up on the left sidebar. Now I have nothing showing up whatever I pick from my recents. It does change the main agent window and the right panel with the new workspace files. Also I noticed that sometimes the terminal sessions just ends and resets on some workspaces that I have.
Also would be nice to have a find or search when viewing changes.
Thank you for the answer, this means that we can’t use this new view until VS Code extensions support is added. Our biggest pain point is our language tools not working making code exploration impossible which is the point of using Cursor for us. If we didn’t want to explore the code we’d use the competition tools that are much cheaper to use.
Will the legacy agent view be maintained until extensions are properly supported?
Things are moving fast these days, and filing bug reports isn’t something I really have time for. But I’m seeing insanely degraded performance across a wide swath of functionality in the Mac OS Cursor 3.0. Overall, the switch to this new agents window and elimination of side-panel agent control in the cloud has destroyed my workflows and productivity. I’m now considering switching platforms completely as each new agent I spin up has some fresh new bedevilment. Is there an announcement, a mea culpa, a blog post… anything, somewhere, that the Cursor team is acknowledging? You guys have really degraded productivity I imagine for many thousands of developers.
- Cannot start new agent - both CMD+Enter and click the button to kick off work do not function - potentially fixed in a newer update
- Agent pauses or stops executing without having done the basic task
- Plans either disappear after navigating to another agent, were never created, or aren’t possible to preview
- Agent requires re-authenticating GitHub, or loses context of previous allow listed commands
Hi I am no longer able to see the agents window or access design mode within the built in browser. I was able to access it earlier today but after quitting Cursor, Cmd+Shift+P -> Agents Window does not work. I tried checking for new updates and restarting both Cursor and my computer but I still no longer see it. Any ideas of how to fix this?
I have the same issue (and version). Also, can no longer spin up agents in the cloud or worktrees.
Hi! I’m really enjoying the new Cursor 3 Agents Window. It has solved many of the issues I had when working across multiple projects. That said, I’d like to share a couple of things:
-
I really miss the auto Agent Review feature that exists in the Editor Window. I find myself making commits there just to trigger the agent review as expected. It would be great to have this feature available in the Agents Window as well.
-
I’d also love the ability to reference another chat, even from a different project. Something like “I did it there, do it here as well,” where the agent can use the full context from that chat, including images, context, bugs, and decisions.
Thanks!
Hi, you can access it by going to File → New Agents Window. If the option is grayed out, it may be because you already have an agents window open somewhere else and it’s currently hidden in the background. You can only have 1 agents window open at a time, unlike the regular view which can have dozens.
Can you give a bit more info? After completing the plan, do you press Build and it fails, or are you saying that you’d like to only implement pieces of the plan at a time, one by one, without using the Build button?
Can you also please let me know your cursor system info? Version and OS?
Seems fixed. I tried on both Windows and Mac, and they both work fine.
Could you provide a bit more detail here? What model are you using and what triggers the pause / stopping? Any error messages?
I have a solution! Yes, sometimes it’s unclear where the Plan went. However, it’s not gone! It’s just that the Plan is stored outside of the project repo (default is something like:
/Users/username/.cursor/plans/my-project.plan.md) You can always ask Cursor, something like “Hey where’s the plan.md” and it will surface it for you. I realize this is not obvious.
Allowlists should remain the same across Agents View and the regular IDE. MCPs I know need re-auth frequently but I use the Github CLI and I haven’t noticed log outs. How do you use Github with Cursor ? I’ll try to replicate
Are you on an Enterprise account by any chance? Do you want to DM me your account email and I can take a closer look? With enterprise plans the rollout is controlled by your admin so there is a chance that Agents View was disabled by your admin.