GitHub Copilot app for Beginners: Run several agents at once

Parallel agent sessions in the GitHub Copilot app

Running multiple AI agents against the same project might sound like a recipe for conflict, but the GitHub Copilot app keeps each agent session isolated so they can work side by side without stepping on each other. The practical payoff: you can kick off several tasks at once and spend your time reviewing results instead of waiting for one job to finish before starting the next.

Think of it as doing laundry at a laundromat. You load multiple machines, set each one independently, and none of them block the others. Start a load, walk away, and come back when the cycle is done—the machines don’t need your attention in the meantime.

How sessions stay independent

An agent session covers one task from start to finish. The app’s session view shows every active task as a card, with its title and current progress.

Each session runs on its own Git worktree, which is what makes true parallelism possible. Because the worktrees are separate, the agents don’t contend for the same files or interfere with each other’s changes. You can start a new session at any time without disturbing the ones already running.

Context is also kept per session. Switch from one task to another and back, and each one picks up exactly where you left it—no need to re-explain the task or re-orient the agent. That reduces the mental overhead of juggling multiple threads of work.

A practical example

Say you’ve got a repository called tailspin-toys and three things on your list: add a funded sort, do an accessibility review, and run the project’s tests. In the Copilot app you’d start a session for the funded sort feature, then immediately open a second session for the accessibility review. Before either finishes, you can start a third session for the tests.

In the session view, you watch all three progress independently and review each result as it completes. Or skip the monitoring entirely—the agents keep working whether or not you’re watching, so you can step away and come back to finished work.

Getting started

The easiest way to see how this works is to launch two small, unrelated tasks at the same time in the GitHub Copilot app. Low-stakes parallel sessions let you observe how the agents run without coordination and how each preserves its own state, giving you a feel for the workflow before you scale it up.