Cursor /multitask: Parallel Async Subagents in Agents Window

Parallel Agent Execution with /multitask

Cursor 3.2 introduces the /multitask command, a new capability in the Agents Window that fundamentally changes how users interact with multiple concurrent coding tasks. Rather than adding each new request to a sequential queue — where work waits behind whatever is currently running — /multitask instructs Cursor to spawn a fleet of async subagents that execute in parallel.

How /multitask Works

When a user invokes /multitask, Cursor does two things. First, for independent requests, it routes each to a separate async subagent that runs concurrently with all others. Second, for larger, more complex tasks, it automatically decomposes the work into smaller units and distributes them across multiple subagents simultaneously. The result is that what might previously have taken multiple sequential agent runs can now be completed in a single, orchestrated parallel pass.

Multitasking on Queued Messages

A particularly practical aspect of the new command is its retroactive applicability. If a user has already queued up several messages, they no longer need to wait for the current agent run to finish before those queued items begin. Users can invoke /multitask on messages already sitting in the queue, converting them from a sequential line into a parallel batch that starts running immediately.

Why This Matters for Developer Workflows

The traditional sequential model creates an implicit bottleneck: one agent works while all others wait. For developers juggling multiple features, bug fixes, or exploration tasks, this means long idle periods. With /multitask, Cursor treats the Agents Window as a true parallel processing environment, bringing agent orchestration closer to how experienced developers actually structure their work — running multiple threads of effort simultaneously and merging results.