How to run parallel AI agents in the GitHub Copilot app | Tutorial for beginners
GitHub explains how to run multiple parallel AI agent sessions in the GitHub Copilot app without stepping on your own changes, using isolated git worktrees and a centralized task board to track in-flight work and review results when you’re ready.
Overview
This video shows how the GitHub Copilot app supports running multiple AI agent tasks against the same project at the same time while keeping each session isolated.
Running multiple agent sessions in parallel
- The video frames the problem as avoiding “chaos” when multiple tasks are running concurrently on the same codebase.
- It contrasts parallel workflows with sequential workflows where you wait for one task to finish before starting the next.
What an agent session is
- The presenter introduces the concept of an agent session as the unit of work/context for an AI agent operating on a project.
Using git worktrees to prevent code collisions
- The core mechanism described is isolated Git worktrees, which keep parallel sessions separate.
- The goal is to prevent changes from different sessions from colliding with each other.
Tracking and switching between active sessions
- The video describes tracking in-flight tasks on a centralized board.
- It also covers context switching between distinct active sessions so you can move between tasks.
Reviewing completed work
- The workflow described emphasizes reviewing completed work “at your own pace,” rather than being forced into a single linear flow.