Cursor 3 introduced a new agent-first workspace that runs beside the IDE and hands tasks between local and cloud agents. Try the new flow if you want tighter multi-agent dispatch and review, but expect a bigger change to existing IDE-heavy habits.
You can read Cursor's official launch post, skim the new Agents Window docs, and dig through the launch threads for the new interface, Agents Window, and Worktrees & Best-of-N. The weirdly concrete bit is that Cursor is not just talking about cloud delegation in the abstract, it says cloud agents now return demos and screenshots for verification in the desktop app, and the handoff works in both directions according to the blog post and forum announcement. Dan Shipper's team also called out the auto-generated demo video as a standout moment after a week of testing Dan Shipper reaction.
Cursor 3's headline feature is the Agents Window, a separate workspace built from scratch around agents rather than files. In the official post, Cursor says the interface is inherently multi-workspace, which means humans and agents can work across different repos from one place.
The forum launch thread adds the sharper product framing: run many agents in parallel across repos and environments, locally, in worktrees, in the cloud, and on remote SSH. That is a bigger shift than a UI refresh. It turns Cursor into a dispatcher and review surface for concurrent agent sessions.
Cursor also kept an escape hatch. The company says users can switch back to the IDE any time, and the forum thread says both windows can stay open simultaneously. That helps explain why early reactions describe Cursor 3 less as a replacement editor and more as a new operating mode sitting beside the editor Ryo on Cursor 3.
The most concrete workflow change is session handoff. According to the launch post, a cloud session can move local when you want to edit or test on your own desktop, and a local session can move to the cloud so it keeps running while you close your laptop or move to another task.
The Agents Window announcement expands the same model across more sources. The sidebar is supposed to show agents started from mobile, web, desktop, Slack, GitHub, and Linear, alongside local, cloud, worktree, and SSH sessions.
That cross-environment handoff is the part reviewers latched onto first. Dan Shipper's team said the local-to-cloud flow felt promising, especially when a cloud agent finished a feature and sent back a demo video. Cursor's own post says cloud agents produce demos and screenshots for verification, which makes the launch feel unusually focused on review artifacts, not just code generation.
Cursor did not stop at dispatch. The official post says the new diffs view is meant to speed up review with a simpler UI, then carry the same work through staging, commits, and PR management.
That matters because Cursor is trying to collapse a familiar multi-window pattern into one surface:
One early reaction captured the opening Cursor sees here. Zeev wrote that Cursor 3 might be the thing that pulls them off Claude Code style TUIs and back into a GUI, while also calling competing desktop agent tools rough around the edges. That is probably Christmas come early for coding agent nerds who want orchestration without living in terminals.
The last notable launch thread is not about the main window at all. Cursor also redesigned worktrees around two explicit commands, described here:
/worktree starts an isolated Git checkout for the rest of a chat/apply-worktree brings the result back to the main branch/best-of-n runs the same task across multiple models at once.cursor/worktrees.json can define project setup steps like dependency install, env-file copy, and migrationsThis part broadens the release beyond a new shell around old agent behavior. Cursor is also making model comparison and isolated execution more explicit in the product surface, right down to dedicated slash commands and a config file for worktree setup. That is separate from the Agents Window story, and it gives Cursor 3 a second identity: not just a prettier agent inbox, but a more opinionated runtime for parallel code generation.
We’re introducing Cursor 3. It is simpler, more powerful, and built for a world where all code is written by agents, while keeping the depth of a development environment.
cursor 3 launched today. the focus here is an agent-first workflow with seamless local and cloud handoff. really nice to see the cursor team build the new interface from scratch and move away from the more traditional setup. but you can still jump between the two.
We’re introducing Cursor 3. It is simpler, more powerful, and built for a world where all code is written by agents, while keeping the depth of a development environment.
cursor 3 is here, and its incredible its my daily driver where i can switch between cloud/local easily go try it: 1. upgrade 2. cmd+shift+p > agent window
We’re introducing Cursor 3. It is simpler, more powerful, and built for a world where all code is written by agents, while keeping the depth of a development environment.
Cursor 3 is here. Where power meets simplicity. Works across all your projects, local and cloud. It starts simple, then unfolds more tools when you need them – so you stay in flow and in control. Enjoy!
We’re introducing Cursor 3. It is simpler, more powerful, and built for a world where all code is written by agents, while keeping the depth of a development environment.
this might be what gets me off of claude code/tuis and back into a gui ive enjoyed bits and pieces of conductor and codex's desktop app, but they're really early and have a lot of rough edges
We’re introducing Cursor 3. It is simpler, more powerful, and built for a world where all code is written by agents, while keeping the depth of a development environment.
We’re introducing Cursor 3. It is simpler, more powerful, and built for a world where all code is written by agents, while keeping the depth of a development environment.
