Cursor 2.0 introduces Composer, its first coding model, and parallel agent workflows

Cursor 2.0 introduces major updates for developers with Composer, its first in-house coding model built for fast, low-latency agentic coding. Most turns complete in under 30 seconds, and the model runs about four times faster than similar ones. Early testers noted smoother handling of multi-step coding tasks, supported by training tools like codebase-wide semantic search for better understanding of large projects.

The update also debuts a redesigned interface centered on agents instead of files, helping developers focus on results while agents manage implementation and navigation. Users can still open files or switch back to the classic IDE layout if they prefer traditional workflows.

Cursor 2.0 supports running multiple agents in parallel on separate git worktrees or remote machines, reducing conflicts and improving output quality when several models tackle the same task. To streamline review and testing, it adds a tool to inspect agent-made changes and a built-in browser that lets agents test and refine their own code. The update is available for download, with full details on the official changelog.