How Zed's AI and Collaboration Work

Zed's AI and collaboration features are built around two ideas: agents that run in parallel at native speed to edit files, navigate code, and run tools, and real-time multiplayer sessions where you chat with teammates, code together, and share your screen and project. Both are available in the same editor, which is written from scratch in Rust to use multiple CPU cores and your GPU. You can try them by downloading Zed for macOS, Linux, or Windows.

What the AI side actually does

Zed describes its AI as "agentic" and designed to "work the way you code." The practical behaviors listed on the site are:

• Run agents in parallel — more than one agent can work at the same time rather than queueing behind a single chat thread.
• Edit files — agents make changes to your code directly.
• Navigate code — agents move through the codebase to find and read relevant files.
• Run tools — agents invoke tools as part of a task.

The site frames the speed claim specifically: agents "run tools at native speed," which it ties to Zed being written in Rust and able to leverage multiple CPU cores and the GPU. That's a performance claim about the editor's architecture, not a separate AI model.

What an agent session looks like in practice

The homepage shows a live example of an agent working in a project. The task given was to add AccessKit support to GPUI so screen readers can traverse the element tree. The agent's visible steps were:

1. Figure out where the accessibility tree should be built and how elements currently expose their roles.
2. Read crates/gpui/src/element.rs.
3. Search for "accessibility" in crates/gpui/.
4. List crates/gpui/src/platform/.
5. Explain its reasoning — that it would look at the GPUI element trait and the window's paint cycle to find where to hook into the tree, and check for existing accessibility scaffolding.

That sequence is a useful picture of the intended workflow: you describe a goal, and the agent reads files, searches, lists directories, and narrates its plan before making changes. The example also shows a model selector in the interface (the screenshot labels it "Claude Opus 4.6"), so model choice appears to be part of the agent UI.

Parallel agents across projects

The homepage's activity feed shows agents running against multiple repositories at once — gpui-accesskit, rope-panic-fix, text-shaping-opt, tree-diff-refactor, axum-billing, sdk-pagination, parallel-agents-page, api-route-cleanup, tailwind-v4, theme-builder-dark, and others — each with a branch name and a timestamp. This is the concrete meaning of "run agents in parallel": separate workstreams progressing simultaneously rather than one at a time.

What the collaboration side actually does

Zed lists three collaborative capabilities:

• Chat with teammates — talk inside the editor.
• Code together — edit the same project with other people.
• Share your screen and project — others can see your screen and work in your project.

The site's framing is "crafted for speed and collaboration with humans and AI," so collaboration and AI are presented as one combined experience rather than separate products.

How the two fit together

The unifying claim is that Zed is "a minimal code editor" for collaborating with both humans and AI. In practice that means:

If your goal is to have an agent make changes across several parts of a codebase while you and a teammate work in the same project, Zed positions both as first-class and simultaneous. If you only need one of the two, each works on its own.

What to check before relying on it

• Platform availability: Zed is available for macOS, Linux, and Windows, so confirm your OS is covered.
• Pricing and account requirements: the site links to a Pricing page and a Subscribe page, which indicates paid options exist. The available material doesn't state what's free, what a subscription includes, or whether an account is required for AI or collaboration — check those pages directly before committing.
• Model access: the example shows a specific model selected in the agent UI. What models you can use, and under what terms, isn't specified in the material here.
• Extensions: Zed highlights a "growing extensions ecosystem," so if you depend on a particular language or tool integration, verify it's available.

The fastest way to judge whether the AI and collaboration features fit your workflow is to download Zed and run one agent task and one shared session on a real project.