Cursor’s 3.2 release ships three features that change how you manage parallel work inside the Agents Window: async multitasking, improved worktrees, and multi-root workspace support.
The headline addition is /multitask. Type it and Cursor spins up async subagents to run requests in parallel instead of queuing them sequentially. For larger tasks, it automatically breaks work into chunks and distributes them across a fleet of subagents simultaneously. If you already have messages sitting in the queue, you can invoke multitask on those too — no need to wait for the current run to finish before parallelizing.
Worktrees get a proper integration in the Agents Window as well. You can now run isolated tasks across different branches in the background, then bring any branch into your local foreground with a single click when you’re ready to test. That removes the context-switching overhead of manually juggling branches while agents are mid-run.
The third piece is multi-root workspace support. A single agent session can now target a workspace made of multiple folders, letting Cursor make coordinated changes across frontend, backend, and shared library repos without you having to retarget the agent each time it moves between codebases. For anyone working in monorepo-adjacent setups — separate repos that still need to stay in sync — this closes a real workflow gap.
TIP
If you’re already using the Agents Window for long-running tasks, try /multitask on your next batch of independent subtasks. Parallel subagents will cut wall-clock time significantly compared to the default queue.
The practical takeaway: if your current Cursor workflow involves waiting on sequential agent runs or manually coordinating changes across repos, 3.2 directly addresses both. Update and test /multitask on any task you’d normally split across multiple agent sessions.