About agents
AI agents are autonomous systems that can evaluate their environment, make decisions, and take actions to complete tasks. Agents can break down complex tasks into steps, use various tools and resources, plan their approach, and adapt based on human feedback until they accomplish their assigned objective.
Copilot coding agent brings automation and assistance to every stage of the software development process on GitHub. You can run multiple sessions of Copilot coding agent concurrently, allowing you to efficiently delegate work items.
Utilizing custom agents you can build out a team of task-specific agents with customized system prompts to handle simpler tasks like writing tests and refactoring, giving you bandwidth to prioritize problem-solving and collaboration. See About custom agents.
Model choice allows you to choose from a selection of AI models to use with your agents, each with its own particular strengths. See Supported AI models in GitHub Copilot.
To learn more about Copilot coding agent, see About GitHub Copilot coding agent.
Managing agents
When utilizing GitHub's agentic features, you can use the Agents tab within a repository that has Copilot coding agent enabled to initiate, monitor, and manage agent sessions without leaving your workflow. You can also use the Agents page to view and start agent sessions. To learn how to enable Copilot coding agent, see Managing access to GitHub Copilot coding agent.
From the Agents tab, you can:
- Kick off new agent tasks: Select an AI model of your choice, and optionally choose from custom agents best suited for the task. See Asking GitHub Copilot to create a pull request.
- Monitor live session logs: Once Copilot starts working, you can click any agent session to open the session log and follow its progress and thought process in real time.
- Track active sessions: You can view all active agent sessions that have been started in the repository.
- Steer agents mid-session: If you realize you didn't scope a request correctly, or want Copilot to use a specific tool or service, you can step in and provide steering input without stopping the run. Steering uses one premium request per message. See Tracking GitHub Copilot's sessions.
- Open a session in VS Code or GitHub Copilot CLI: When you want to start working on changes to an agent session in your local development environment, click Open in VS Code or Continue in GitHub Copilot CLI to bring the session to your local machine.
Note
Opening a session in VS Code requires the latest versions of VS Code, the GitHub Copilot extension, and the GitHub Pull Requests extension.
- Review and merge agent code: Once Copilot completes a session, you can jump to the pull request to review the changes, request further improvements, or approve and merge. See Reviewing a pull request created by GitHub Copilot.
Next steps
To start managing agents, see Managing Copilot coding agents.