In a major shift toward autonomous software development, Cursor (the AI-native code editor) has launched Cursor 3, a unified workspace that allows developers to run AI coding agents on any machine and control them remotely from a smartphone or browser.
The update effectively turns Cursor from an IDE into an orchestration layer, allowing you to kick off complex, long-running coding tasks on your powerful home “devbox” or cloud server while you’re away from your desk.
1. How the Remote Control Works
The core of this feature is a new “worker” process that bridges your hardware to Cursor’s central dashboard.
- Setup: You start a lightweight worker on your target machine (laptop, remote server, or cloud VM) using a simple terminal command.
- The Mobile Interface: From a mobile browser (a dedicated app is reportedly in development), you log into your Cursor dashboard. You can see all your connected “active” machines.
- Command & Control: You can type a task in plain language (e.g., “Refactor the auth logic to use the new SDK and run the test suite”) and assign it to a specific machine.
- Persistent Execution: The agent runs directly on that remote machine’s environment. You can close your phone, and the agent will continue working in the background, notifying you when the task is complete or if it needs a manual approval.
2. Key Features of Cursor 3
The release of version 3.0 introduces several tools designed for this “agent-first” workflow:
| Feature | Description |
| Agents Window | A new dedicated UI (Cmd+Shift+P -> Agents Window) to manage multiple parallel agents. |
| Parallel Execution | Run different agents across multiple repositories and environments simultaneously. |
| Cloud-Local Handoff | Move an active agent session from a cloud VM to your local machine (or vice versa) without losing context. |
| Design Mode | Use the built-in browser to annotate UI elements, giving the agent precise visual feedback. |
| /best-of-n | Runs the same task across multiple models (e.g., Claude 4.6 and GPT-5.4) so you can pick the best output. |
3. “Long-Running” Agents
Cursor has also moved its Long-Running Agents out of research preview. These agents are designed for “ambitious” projects that might take hours or even days.
- Self-Verification: Using “YOLO mode” and advanced planning, these agents don’t just write code; they build it, run tests, fix their own errors, and iterate until the task passes your predefined benchmarks.
- Watch Back: The system can record a video of the agent’s actions, allowing you to review exactly how it navigated your files and terminal while you were offline.
4. Security: Self-Hosted Cloud Agents
For enterprise users worried about code privacy, Cursor 3 introduces Self-Hosted Cloud Agents.
- Local Data, Remote Brain: The agent’s “thinking” happens via Cursor’s models, but the execution (file access, terminal commands, build outputs) stays entirely within your own private network or infrastructure.
- Audit Logs: Admins can monitor all agent actions and sandboxed commands to ensure compliance with security policies.
5. Why This Matters: The “Deskless” Developer
Cursor is betting on a future where developers spend less time writing individual lines of code and more time managing a fleet of agents. By allowing control from a phone, they are removing the physical “laptop-open” requirement for high-level software engineering.
“You can now run Cursor on any machine and control it from anywhere,” the company announced. “Kick off agents from your phone to run on your devbox while you’re at lunch and come back to a completed PR.”
