GitHub Copilot now offers cloud‑based AI agents, the ability to run multiple agents at once, a new model‑selection tool called Copilot Fusion, and experimental Computer Use and Agent Host features. These changes let developers delegate larger tasks, manage them from any device, and even have agents interact with desktop apps, all within VS Code and other Copilot clients.
Visit the official GitHub Copilot website →
What changed
The 2026 update shifts Copilot from a simple code‑suggestion assistant to a full‑blown AI agent platform. Key additions include:
- Cloud agents that run on GitHub‑hosted systems, freeing local machines and allowing work to continue in the background.
- Multiple agent sessions that can be managed simultaneously through the new Agents window, letting one agent tackle a bug while another drafts a feature.
- Copilot Fusion, a general‑availability tool that selects the most appropriate AI model for each task, balancing speed, reasoning quality, and cost.
- Computer Use, an experimental feature that lets agents read desktop screens, click controls, and type text in supported desktop applications.
- Agent Host, which keeps agent sessions separate from VS Code windows and supports SSH and development tunnels for remote work.
- Expanded plugins and connectors that bring services like Slack, Notion, and Microsoft Learn into the agent context.
- Enhanced testing, review, and security controls, including automated routine tasks, Agent Merge for pull‑request prep, and local sandboxing to limit agent access.
Who gets it and when
These features are available to Copilot users who have access to the latest VS Code extension and the Copilot plan that supports AI agents. Some capabilities, such as Computer Use and certain plugins, remain in preview and may require specific account settings or platform support. The update was announced at GitHub Universe on October 28, 2026, and has been rolled out to the general public since then.
How to use it
To start a cloud agent:
- Open the Agents pane in VS Code.
- Click “New Agent” and enter a task description.
- Select the desired model or let Copilot Fusion choose automatically.
- Monitor progress in the Agents window and review the generated code when the agent completes the task.
For multiple agents, repeat the steps above for each task; the Agents window will list all active sessions. To enable Computer Use, grant the required desktop permissions in the agent settings and select an agent that supports the feature. For remote work, connect via SSH or a development tunnel, then launch an agent on the remote machine; the session will persist across device switches.
What it means if you are choosing GitHub Copilot
The new agent capabilities make Copilot a stronger partner for teams that need to automate routine coding, review, and testing tasks. However, the experimental nature of some features and the need for careful security controls may steer developers toward alternatives that offer more mature agent support, such as JetBrains Code With Me or IntelliJ’s AI plugins. If you prioritize a stable, single‑model suggestion engine, the classic Copilot experience may still be preferable.
FAQ
What are the new AI agent features in 2026?
The update includes cloud agents, multi‑agent sessions, Copilot Fusion, Computer Use, Agent Host, plugins, and enhanced testing and security controls.
How does Copilot Fusion work?
Copilot Fusion automatically selects the most suitable AI model for each task, coordinating multiple models if needed, to balance speed, quality, and cost.
Can Copilot run several agents at once?
Yes. The Agents window lets you manage multiple agent sessions simultaneously, assigning distinct tasks to each.
Bottom line
GitHub Copilot’s 2026 update transforms it from a code‑completion helper into a versatile AI agent platform. With cloud agents, multi‑session management, smarter model selection, and experimental desktop interaction, developers can delegate larger tasks, work across devices, and automate routine workflows. The changes are powerful but still evolving; users should evaluate the experimental features and security implications before fully adopting the new agent model. For teams that need robust, multi‑agent support and remote flexibility, Copilot 2026 offers compelling options, while those preferring a more traditional, single‑model assistant might consider other tools.
Sources
Full guide: GitHub Copilot Review: Features, Pricing, Setup, Pros, Cons, and Best Practices for Smarter Coding — plans, prices and alternatives.