The GitHub Copilot app introduces the capability to manage multiple AI agent sessions simultaneously. This functionality addresses the challenge of coordinating several agents working on the same project, preventing interference and maximizing efficiency. The app’s architecture facilitates parallel execution through independent Git worktrees for each agent session. This allows agents to operate concurrently without impacting each other’s progress.
Each agent session operates as a distinct task, tracked within the sessions view. This view displays the title and progress of each session, providing a clear overview of ongoing activities. The isolation of each session within its own Git worktree ensures that changes made in one session do not affect others, maintaining a stable development environment. This parallel execution significantly reduces the time spent waiting for individual agents to complete their tasks.
Consider the ‘tailspin-toys’ repository, where three distinct tasks – adding a funded sort, performing an accessibility review, and running tests – can be managed concurrently. The GitHub Copilot app allows initiating these tasks independently, with each running in its own session. Engineers can then monitor progress and review results without interruption, streamlining the workflow.
Furthermore, each session maintains its own context, enabling seamless switching between tasks. This minimizes context switching and reduces feelings of distraction. The GitHub Copilot app also provides tools for viewing diffs, running terminal commands, and previewing web apps side by side within the app interface. Project HydraFusion, available as a research preview, demonstrates the potential of this approach, matching or exceeding Opus 5 baseline performance while reducing workflow costs.
