Cursor launches Projects: One Agent acts as project manager, overseeing thousands of Agents beneath it.

Beating AI News Flash: Cursor launches Projects, upgrading from a single Agent doing work to "Agent managing Agents." Each project has a coordinating Agent that doesn't write code itself, but is responsible for breaking down tasks, making plans, and distributing work to thousands of sub-Agents for parallel completion.

Projects retain code repositories, research results, and user preferences long-term, and can pick up work months later. Tasks run in the cloud by default, continuing even when the computer is turned off; when local testing is needed, it automatically spins up a local Agent. It can also monitor Slack and PRs, automatically fix CI, and start working on new tasks on its own without waiting for someone to send another prompt.

This is essentially the official productization of Cursor's previous multi-Agent experiments. In January of this year, it already demonstrated hundreds of Agents collaborating simultaneously, even writing a browser with over 1 million lines of code from scratch. Projects has now entered beta and is gradually being rolled out to all users.