The biggest problem of the Agents window is not the new features it has (like having multiple workspaces at the same time, or the shift in paradigm where agent comes first and code comes later, even visually in the UI). It’s not like I’m missing the extreme performance drops Cursor has and Agents Window doesn’t have.
Those are great features. Those should be part of Cursor.
The biggest problem of the Agents Window experience to me is that it’s missing (quite literally) all the things that made cursor great.
I can’t speak about your business direction because you guys know much more than anyone else, but since you might be reading, I might as well just tell you this:
Agents Window seems to be an attempt to steal some attention from Claude Code, OpenCode, Codex, etc.. And as it stands right now, it might work for that purpose.
But it also poses the risk of losing the current user/customer base.
And maybe the risk is worth it for you.
As a paying user for quite some time though, I do not feel Agents Window is going to make me use cursor more, but definitely something that made me look away from it.
Agents Window is an early addition, not a replacement. Classic editor mode is still here and works the same as before. The idea is that both options can coexist, an agent-first workflow for some tasks and a full editor for others.
It’d help a lot if you could share which specific things you’re missing. General feedback is useful, but specific items are easier to pass to the team for prioritization. Let me know what exactly is blocking you and we’ll check what’s already on the radar.
Some of the things that are useful and missing from the Agents Window, features that made cursor great:
Ability to fire up an agent to do some of the tasks in the to-do list created by “plan mode”.
It’s there in the editor view, and key for any serious work as the ai models tend to have a rather stupid behavior when doing many tasks on a relatively big projects.
One of the most useful features in Cursor is the ability to review the code and go change by change seeing not only what the AI did, but actually accepting or rejecting changes.
That ability seems to be gone in the agents window. Maybe on purpose or maybe because it’s more geared towards people who don’t even know how to code.
But for people with money on the line that want to review agent-led changes easier, this feature is a must.
While AI-autocomplete sort-of works, actual intellisense is nowhere to be seen. Not good when trying to tell between hallucination and an actual, valid function/method/variable/constant/etc.
Less important, probably a bug: Can’t save the plan to workspace. I don’t have the option.