Cursor didn't ship an update. It shipped a thesis: the IDE is no longer about editing files — it's about orchestrating agents that edit files for you. Cursor 3 introduces the Agents Window, a new interface built from scratch that replaces the editor-first model with an agent-first workspace. Run parallel agents across repos, hand off between local and cloud, compare models side by side, and drop back into the full IDE when you need to.
The framing from Cursor's team is direct: "In the last year, we moved from manually editing files to working with agents that write most of our code." Internally, more than 30% of the PRs Cursor merges are now opened by agents running in cloud sandboxes.
The Agents Window
The centerpiece of Cursor 3. Every agent — local, cloud, worktree, remote SSH — appears in a unified sidebar. Kick off agents from desktop, mobile, web, Slack, GitHub, or Linear. View them side by side or in a grid. Cloud agents produce demos and screenshots of their work for verification without needing to pull the code down.
Two new commands ship alongside the window:
/worktree— creates an isolated Git worktree so agent changes happen in a separate branch/best-of-n— runs the same task in parallel across multiple models, each in its own worktree, then compares outcomes
Local-Cloud Handoff
Moving agents between environments is now near-instant. Start an agent in the cloud from your phone, come back to your desk and pull it local for testing. Or push a local session to the cloud so it keeps running while you step away. It's the same idea as Anthropic's Dispatch, but built into the IDE workflow rather than as a standalone feature.
Design Mode
For frontend work, Design Mode opens an integrated browser panel where you can annotate UI elements directly — click on a button, a card, a layout — and target code changes visually instead of describing them in text. Chrome-based browsers only for now, but it eliminates a real pain point in UI iteration.
The Cost Question
The reaction has been polarized. Reviewers scored it 92/100 and called parallel agents a "paradigm shift." Product Hunt gave it #3 for launch day. But the cost model is raising flags. One early adopter reported spending $2,000 in two days running cloud agents with premium models. Others reported switching from $1,800/month on Cursor to roughly $200/month on Claude Code and Codex for equivalent output.
The competitive picture is shifting too. Developer surveys from early 2026 put Claude Code at 46% "most loved" versus Cursor at 19%. Claude Code reportedly consumes 5.5x fewer tokens for equivalent complex tasks. Cursor wins on visual orchestration and IDE depth; Claude Code wins on cost efficiency and terminal-native simplicity. The choice comes down to whether you want an agent workspace or an agent runtime.
Still an IDE at Heart
Cursor 3 doesn't abandon the editor. Full LSP support, go-to-definition, file inspection, an integrated browser, and a plugin marketplace with MCPs, skills, and subagents all carry forward. You can have the Agents Window and the IDE open simultaneously, or switch between them. Enterprise features include self-hosted cloud agents (code stays on your infra), audit logs, and admin controls for team environments.
Cursor's bet is clear: developers won't abandon the IDE even as agents do most of the authoring. The terminal-native camp — where Claude Code holds a commanding lead — disagrees. Cursor 3 is the strongest version of the IDE-centric argument yet. Whether orchestration UX beats raw cost efficiency is the question the market is actively answering.