Cursor 3 Agents Window: A Hands-On Tutorial

Cursor 3 shipped on April 2, 2026, and it's not a minor bump. The old Composer pane is gone. In its place: a full Agents Window where you run multiple AI agents in parallel — across local, cloud, worktree, and remote SSH environments — all from one screen.

Cursor 3 is a rebuilt version of the AI-native code editor that replaces the Composer interface with a dedicated Agents Window for running and managing parallel AI coding agents. It adds Design Mode for visual UI annotation and a /best-of-n command for comparing outputs across models.

What You'll Need

• Cursor 3.0+ — download from cursor.com
• A Cursor Pro subscription ($20/month) or higher. The free Hobby tier gives limited agent requests. Pro gives you a $20 monthly credit pool for premium models, cloud agents, and parallel execution.
• Any codebase to work with
• Basic code editor familiarity

How Does the Cursor 3 Agents Window Work?

The Agents Window is Composer rebuilt for parallel agents and cloud execution. Instead of a single chat sidebar, you get a full workspace where each agent runs in its own tab. You can arrange them side-by-side or in a grid layout, and since version 3.1 (April 13), your tiled layout persists across sessions.

Each agent operates in one of four environments: local (your machine), cloud (Cursor's infrastructure), worktree (an isolated git branch), or remote SSH. Cloud agents can run while your laptop is closed and produce screenshots and demo videos so you can verify their work before merging.

Here's how to set it up.

Step 1: Update to Cursor 3

Open Cursor and check for updates. If you're on a version prior to 3.0, you'll see the update prompt. Download and install it.

Installing fresh? Grab it from cursor.com. Available on macOS, Windows, and Linux.

Step 2: Open the Agents Window

Open the command palette with Cmd+Shift+P (Mac) or Ctrl+Shift+P (Windows/Linux) and type Agents Window. Select it.

You'll see a workspace with a sidebar for managing agents and a main area for the active agent's output — file diffs, code changes, and conversation history.

Step 3: Launch Your First Agent

Start a new agent and type a prompt. Something real:

Refactor the auth module to use JWT tokens instead of session cookies.
Include unit tests for the token validation logic.

The agent picks up your project context, starts editing files, and streams diffs in real time. You can click into any diff to jump to the exact line in the editor — full LSP support (go to definition, references) works inline.

Step 4: Run Agents in Parallel

While agent one is working, create another agent tab. Give it a separate task:

Add dark mode support to the settings page.
Use the existing design tokens from theme.ts.

Since Cursor 3.1, you can divide your workspace into tiled panes and drag agents between tiles. Both agents work simultaneously in separate worktrees so their changes never collide.

Step 5: Use Design Mode

This is the feature that changes UI work. Toggle Design Mode with Cmd+Shift+D (Mac) or Ctrl+Shift+D (Windows/Linux).

A browser view opens connected to your running dev server. Use Shift+drag to select a UI element, then press Cmd+L to add it to your agent chat. Now you can type "make this button wider with rounded corners" and the agent knows exactly which element you mean. No more describing CSS selectors in plain text.

Design Mode works with anything your browser can render — React, Vue, Svelte, Django templates, plain HTML. Just have your dev server running first.

Step 6: Try /best-of-n

Type this in any agent tab:

/best-of-n Implement a caching layer for the API responses

Cursor runs the same prompt across multiple models in parallel, each in its own worktree. You get a side-by-side comparison of each model's approach — the diffs, the file structure, the logic. Pick the one that makes the most sense for your codebase.

Step 7: Review and Merge

Each agent works in an isolated worktree — a real git branch. When it finishes, review the diffs. You can accept all changes, cherry-pick specific files, or reject and re-prompt.

With cloud agents, you also get screenshots and demo videos of the running code before you merge, so you can verify UI changes visually.

How Does Cursor 3 Compare to Other AI IDEs?

FAQ

Is Cursor 3 free? The update itself is free. The Hobby (free) plan gives limited agent requests and tab completions. Parallel agents, cloud agents, and premium model access require Pro ($20/month) or higher. Pro gives a $20 monthly credit pool — roughly 200-250 Claude Sonnet requests. Pro+ ($60/month) gives 3x credits, Ultra ($200/month) gives 20x with priority access.

What's the /best-of-n command exactly? It runs the same prompt against multiple LLMs simultaneously, each in its own isolated worktree. You see the results side-by-side and pick the best output. You don't need to specify a number — Cursor handles the model selection.

Does Design Mode work with backend-rendered pages? Yes. It works with anything your browser renders at localhost — React, Next.js, Rails, Django, plain HTML. The shortcuts are Cmd+Shift+D to toggle, Shift+drag to select elements, and Cmd+L to add the selected element to your agent chat.

What happened to Composer? The Agents Window is Composer's replacement — same concept, rebuilt for parallel agents and cloud execution. Your old Composer history is still accessible.

If you found this useful, subscribe to CodeBrainery for more step-by-step tutorials like this — new guides drop daily. You can also follow us to get notified when we publish something new. Got questions? Drop them in the comments.