Cursor、マルチタスク・Worktrees・マルチルートワークスペースに対応 Multitask, Worktrees, and Multi-root Workspaces

Cursor, the AI-assisted code editor, has rolled out three updates aimed squarely at developers who juggle multiple codebases and concurrent streams of work: multitasking for agents, native Git worktree support, and multi-root workspaces. Together they push the editor further from a single-file assistant toward an orchestration layer for parallel agent work.

The multitasking feature allows several agent tasks to run simultaneously, with progress for each surfaced in a consolidated view. Previously, developers had to wait for one task to finish before kicking off the next. Now a more asynchronous workflow becomes practical: while reviewing the diff produced by one agent, a user can dispatch another to start on an unrelated change, and check back when it has made progress.

Git worktree support addresses a subtler but increasingly common pain point. Worktrees let a single repository have multiple branches checked out in separate directories at the same time. By giving an agent its own worktree, its in-progress edits no longer collide with the files the developer is actively editing, and long-running refactors or experimental branches can be isolated cleanly. The pattern has been gaining traction across the AI coding tool ecosystem, with projects such as Cline and Aider also leaning on worktrees as a way to keep autonomous edits from stepping on human work.

Multi-root workspaces, meanwhile, let users bundle several distinct repositories — a frontend and a backend, or a set of services that live outside any monorepo — into a single workspace, and expose that combined context to the agent. The underlying capability has long existed in VS Code, on which Cursor is built, but extending it so that an AI agent can reason across project boundaries rather than within a single root is the meaningful shift here.

The broader context is a clear industry move toward parallel, longer-running agents. Anthropic has publicly championed workflows in which multiple Claude instances work side by side, and competing products such as Devin and Claude Code have made concurrent task execution a central selling point. Cursor's incremental additions appear to position it along the same trajectory, repackaging the IDE as a control surface for fleets of background agents rather than a tool focused purely on inline completion or single-file edits.

There are practical reasons this matters beyond marketing posture. Parallel agent execution stresses assumptions that traditional editors were never designed for: file locking, branch state, dependency installation, and test runs all become contested resources when several autonomous processes are touching the same checkout. Worktrees and multi-root workspaces are, in effect, plumbing that makes parallelism safe to use in everyday development rather than a demo-only feature.

For teams already invested in Cursor, the immediate appeal is likely the ability to keep one agent grinding on a tedious migration in an isolated worktree while another handles a quick bug fix and the developer continues hand-editing a third branch. For those evaluating AI coding tools more broadly, the update is another data point suggesting that the line between IDE and agent runtime is steadily blurring. Whether that convergence ultimately favors editor-led products like Cursor or platform-led offerings from model vendors remains an open question, but the direction of travel looks consistent across the field.