Cloud agents let developers delegate coding tasks and review results across multiple devices.
Copilot Fusion helps select suitable AI models for different development tasks.
Computer Use and Agent Host extend Copilot beyond the code editor and support remote workflows.
GitHub Copilot does a lot more than suggest code inside Visual Studio Code (VS Code) now. New AI agent features give developers more control over big coding tasks, remote work, model choice and desktop apps.
The latest updates, announced around GitHub Universe on October 28, 2026, point to a clear shift toward agents that can handle whole development tasks. Developers can hand work to agents, review the code changes, compare different solutions, and run several tasks at once.
Cloud-based agents are one of the biggest changes. A developer gives an agent a coding task, and it works on it through GitHub-hosted systems. There’s no need to keep a local computer running the whole time. The agent carries on in the cloud, and the developer can check progress and review results later.
That helps with bug fixes, code changes, and routine project work. Follow-up instructions can go through VS Code, GitHub Mobile, the Copilot app, or GitHub.com, so projects can be managed from different devices without staying tied to one development machine.
Copilot also supports several agent sessions at once. Developers can give separate tasks to different agents and keep each one in check through the Agents window. One agent might look into a bug, another might suggest a new feature, and a third might check a failing test.
On large projects, that can save real time. Developers can also compare different solutions before picking one, and the Agents window makes it easy to jump between sessions and follow each task. Complex work gets split into smaller parts without losing control of the codebase.
Copilot Fusion, which reached general availability on October 28, 2026, helps Copilot pick suitable AI models for different tasks. Developers often want a fast model for simple edits and a stronger one for tricky logic or software design. Fusion can send work to the right model and coordinate several models when a task needs more than one.
That cuts down on choosing a model by hand for every request, and it can balance speed, reasoning quality, and cost. GitHub has also announced an experimental preview of local model support in Fusion, which gives developers more say over which models join in.
Also Read - Visual Studio Code (VS Code) | Developer Tools | Software for Code Editing & Software Development
The new Computer Use feature takes Copilot out of the code editor. Supported agents can read desktop screens, click controls, type text and work with certain desktop applications, which helps with jobs that need more than code edits or terminal commands.
An agent might need to open an app, read something on its screen or finish a simple action through its interface. Computer Use is still experimental and supports local Copilot harness sessions on macOS and Windows. Users have to grant the required app permissions. Careful access control matters here, because desktop agents can touch tools and files outside the project folder.
Agent Host keeps agent sessions separate from individual VS Code windows. It also supports remote work through Secure Shell (SSH) and development tunnels. A developer can connect to a remote machine, start an agent task and come back to the same session later.
Teams that rely on remote servers or shared development environments may find that useful, and it gives developers more flexibility when they switch devices. The remote machine does have to stay active if an agent needs to keep working there, so cloud agents and remote Agent Host sessions fit different needs.
GitHub has widened support for agent skills, plugins and Model Context Protocol (MCP) servers. These give Copilot access to extra services and project information, such as Slack, Notion and Microsoft Learn, so agents get more context and more useful tools to work with.
The Customization Marketplace offers a way to discover skills and connectors. GitHub lists several of these features as previews, though, so access may depend on account settings and rollout status.
Other updates target routine software work. Agent automations can run tasks hourly, daily, weekly or on demand, covering things like repository checks, issue reviews and project reports. Agent Merge can help prepare pull requests by dealing with review feedback, failed checks and merge conflicts. Local sandboxing limits what agents can reach in terms of files, networks and credentials, and it reached general availability across supported Copilot tools.
Why this MattersThe new agent features mark a real shift in how software gets built. Developers can hand off complex tasks, manage several agents and spend less time on routine code work. Teams can ship faster, fix issues sooner and put more energy into product quality, system design and key technical decisions.
Copilot now lets developers assign larger tasks, manage several agents and work across local and remote systems. Fusion brings smarter model choice, and Computer Use pushes agent activity beyond the editor. Some features are still experimental or in preview, and access may vary by Copilot plan and platform.
The biggest change is the developer’s role. Code still needs human review, clear instructions and careful security checks, but agents can now take on more of the routine load. GitHub’s latest VS Code updates give developers more ways to steer that work while keeping control of the final code.
1. What are the new GitHub Copilot AI agent features in 2026?
The new features include cloud agents, multiple agent sessions, Copilot Fusion, Computer Use, Agent Host, plugins, and service connectors.
2. What is Copilot Fusion?
Copilot Fusion helps select and coordinate suitable AI models for different coding tasks. It aims to balance speed, reasoning quality, and cost.
3. Can GitHub Copilot run multiple AI agents at once?
Yes. Developers can manage several agent sessions through the VS Code Agents window and assign separate tasks to different agents.
4. What does Computer Use do in GitHub Copilot?
Computer Use allows supported agents to interact with desktop applications through screen access, clicks, and text input. The feature remains experimental.
5. Does GitHub Copilot support cloud-based coding tasks?
Yes. Cloud agents can handle assigned coding tasks through GitHub-hosted systems. Developers can review results and send further instructions through supported Copilot clients.