Cursor 0.50 brings Background Agents in early preview, 1.0 opens them to everyone

Think of handing a coding task to a colleague who works somewhere else while you get on with something else. You hand over the task, they proceed on their own, and you check in when you need to. That is what Cursor introduced on 15 May 2025 with the Background Agents in version 0.50.

Until then, even the most advanced AI coding tools required a developer to stay at the keyboard guiding every step. With this release Cursor moves the agent's work to remote machines: you can launch several agents in parallel on different tasks, and the local editor stays free in the meantime.

The framing is not "hit enter and come back tomorrow", though. Cursor itself describes three actions that stay with the user at any point: view the agent's status, send a follow-up, or take over. And the scope of work the company says it used the feature for internally is concrete: fixing nits, doing investigations, and writing first drafts of medium-sized pull requests. Not "writing the software for you".

At first the feature is a preview you enable by hand in settings. It becomes available to everyone on 4 June with Cursor 1.0, which pairs it with Bugbot: an automated reviewer that comments on pull requests in GitHub when it spots a possible bug and sends the developer back into the editor with the fix already queued up. Together the two tell the story better: the agent produces, a second pass reviews, the final call stays human.

The natural comparison is Cognition's Devin, the agent that first showed a whole coding task could be delegated. Cursor's difference is that the delegation starts from inside the editor the developer already uses.

For developers, the practical shift is about where the time goes: less mechanical writing, more review and judgement on what the agent produced.