GitHub has just introduced a powerful new coding agent for Copilot that takes on real development work right inside your GitHub workflow. Whether you're using GitHub.com, VS Code, the GitHub CLI, or GitHub Mobile, assigning an issue to Copilot kicks everything off.

Whether you're working on GitHub.com, using GitHub Mobile, the CLI, or directly in VS Code, you can now assign an issue to Copilot and watch the magic happen.

A Developer Agent That Builds Alongside You

The process is simple. Assign a GitHub issue to Copilot just like you would to a teammate. The agent reacts with an emoji and quietly gets to work. It spins up a secure development environment using GitHub Actions, clones your repo, configures the tools, and starts analyzing your code.

As it works, it creates a draft pull request and begins pushing commits with updates. Every step is visible through session logs, showing you exactly what it’s doing and why. This means you’re always in control and never left guessing.

Built for Real Coding Tasks

This isn’t just a glorified autocomplete. The agent excels at low-to-medium complexity work in clean, well-tested codebases. It can,

• Add features
• Fix bugs
• Write or extend tests
• Refactor legacy code
• Improve documentation

You can focus on bigger challenges, while Copilot handles the routine work that often slows teams down.

Just Talk to Copilot

You can also guide the agent through natural language prompts. For example, in Copilot Chat (either in GitHub or in VS Code), you can type.

@github Open a pull request to refactor this query generator into its own class.

The copilot will take it from there starting the work, opening a PR, and reporting back as it makes progress.

Smart Enough to Understand Context

The new agent uses advanced retrieval-augmented generation (RAG) with GitHub’s powerful code search, giving it deep awareness of your project. It doesn’t just look at the file it understands how everything fits together.

With Model Context Protocol (MCP) support, the agent can also pull in data from outside GitHub. It even understands images, so if your GitHub issue includes a screenshot of a bug or a mockup of a new feature, it can “see” and interpret that as part of its work.

Review and Refine, Just Like with Any Teammate

Once the agent completes its task, it notifies you for a manual review. You can leave comments asking for changes just like with a human contributor. The agent reads your feedback and updates the code accordingly.

It also respects any existing repository rules and coding standards, and even considers prior issues or PR discussions for context. The result is clean, relevant, and reviewable code never something random or out-of-scope.

Designed with Security in Mind

Security hasn’t taken a backseat. In fact, GitHub has made sure the agent works within your existing policies.

• It only pushes to branches it creates your default branches to stay untouched.
• Required reviewers still apply the person who triggered the agent can’t approve the PR.
• Workflows don’t run until you approve them.
• The agent’s internet access is restricted to trusted domains, which you can customize.

This ensures that even though you’re getting AI support, your security and compliance posture remains intact.

Powered by GitHub Actions

The agent runs entirely on GitHub Actions, the same CI/CD platform trusted by open-source projects, startups, and large enterprises alike. With over 25,000 actions and more than 40 million daily jobs, Actions offers the scale and flexibility needed for real-world coding workloads.

A New Era of AI-Augmented Development

The Copilot coding agent isn’t replacing developers it’s supporting them. Taking on repetitive and time-consuming tasks, allows developers to focus on deeper, more creative problems.

With this launch, GitHub is paving the way for a future where AI not only writes code but contributes meaningfully, securely, and transparently within the team.