When we started building Cursor, we forked VS Code instead of building an extension so we could shape our own surface. With Cursor 3, we took that a step further by building this new interface from scratch, centered around agents.
This part sounds to me like they are starting to move away from VS Code into their own thing? Unless they only add additional things to the UI with that view, it seems a bit unclear to me.
Another thing they advertise is that they have cloud and local agents and now switching between them, or moving them, is now a better user experience. Not something I am looking for, because I don’t use that functionality. Giving agents free reign over stuff while I am not there is not something I am comfortable to do.
This new view however reminds me a bit of their CLI. In the CLI you don’t really write it yourself, you let the AI do its thing and you review and commit. This view will give you the same experience now.
Dive deeper anytime by viewing files, and go to definition in the editor with full LSPs.
This part I find confusing. LSP and stuff existed before, why is this advertised as something new?
Either way, I will be trying out this new Agent Window!
Video by Fireship about it, which also talks about moving away from vs code and writing their own editor. So my guess from the other time was correct. Kind of exciting.
So today I actually opened the agent window and I was shocked at the interface. It is the same chat interface you see when you open Gemini or ChatGPT on the web. It does feel like they’re trying to move towards the more casual or common consumer, instead of developer. Thst makes me a bit sad.
That sad, I also saw they now have a marketplace. With some already interesting plug-ins. That part I like a lot. Create a plug-in with custom skills that can be used company wide. Or team wide. However you like!
I think I’ll find myself more familiarised with Cursor in the near future. Bit the bullet and asked for a license at work; a lot of my peers are using it for their day-to-day and sounds to me like it can easily operate while keeping multiple codebases in its context window (I hope I’m using the right term), which is entirely what I’ll be up to.
Have you used Cursor for a bit in the past? How has it been working out for you so far?
I’ve switched to cursor when it was still version 1. I’ve been using it since then and I’ve enjoyed it. I’m also using it with Composer 2. Generally, I’m happy. I do wish I’d knew how it works under the hood, because with the Claude Codes leak I’ve just became curious of the inner workings and its loop.
Never tried Cursor myself, I’m more used to working with Antigravity, since I like seeing what the AI does in real time, and then manually approving/denying the changes, maybe I should give it a go.
You do that in cursor as well. Maybe you’re mistaken it for doing everything autonomous? Which indeed is an option since cursor 3.0 with cloud workers and such. Personally not a fan of.