Zed Adds Parallel Agents with a Threads Sidebar for Managing Multiple AI Sessions at Once

Zed shipped parallel agent support on April 22, adding the ability to run multiple AI agents concurrently in the same editor window. The team says it’s the first AI code editor to support this natively, without hacks or extensions.

The pitch is straightforward: instead of finishing one agent task before starting the next, you can run them simultaneously. Refactor in one thread, write tests in another, debug something in a third.

The Threads Sidebar

The core of the feature is a new Threads Sidebar in the left panel. It groups agent threads by project so you can see everything running at a glance, with quick controls for stopping threads, archiving them, or launching new ones.

Each thread has independent access controls. You set which folders or repositories that specific agent can read and write, which means you can isolate threads from each other when the work demands it. If you want them sharing context, that’s also supported.

You can mix models within a project: one thread could use Opus, another could use a lighter model for something less complex. Or you can span a single agent thread across multiple repositories if the task crosses repo boundaries.

The interface changes alongside the feature. The Threads Sidebar sits on the left, with Project and Git panels moved to the right. That’s a shift from the traditional file-browser-first layout, and Zed is giving existing users the option to opt in or keep their current arrangement. Panel positions are configurable through right-click menus.

Performance

Zed has leaned on its Rust foundation as a differentiator since the editor launched, and the parallel agents announcement does the same. The team reports 0.4-second startup, 2-millisecond input latency, and 120fps rendering maintained even while multiple agents execute. For a feature that’s inherently concurrent, that matters: the editor shouldn’t slow down because agents are running.

Context

Zed launched with a clear editor-first identity, positioning itself as a fast, clean alternative to Electron-based tools. The agentic push over the past year represents a meaningful broadening of scope. Parallel agents is the most aggressive move in that direction so far.

Cursor added /multitask in version 3.2, shipping a few days after Zed’s parallel agents, which lets Cursor break requests into async subagents. Cursor’s approach is more request-driven; Zed’s is session-driven. The distinction matters depending on how you think about agentic work: composing larger tasks versus running independent parallel streams.

Zed is free and open source. The parallel agents feature is available in the current stable release.

---

Sources: Zed Blog, Zed Parallel Agents, AlternativeTo