At Microsoft’s Build developer conference, GitHub announced the rollout of a new AI coding agent built directly into GitHub Copilot. This upgraded assistant can now handle development tasks like fixing bugs, writing features, refactoring code, and improving documentation.
Developers can assign issues to Copilot through GitHub.com, GitHub Mobile, or the GitHub command-line interface, just like assigning them to a human. The agent reacts with a 👀 emoji and kicks off its work.
Behind the scenes, it boots up a virtual machine, clones your repo, configures its environment, and starts reading your codebase. It uses retrieval augmented generation (RAG) and GitHub code search to understand what’s going on and what it needs to do.
Built-in security, and human oversight required
Despite all its power, the Copilot agent doesn’t act without checks. Every change it makes goes into a separate branch and draft pull request; it won’t push code live without human review and can’t approve its own work.
“The agent can only push to branches it created, keeping your default branch and the ones your team created safe and secure,” Thomas Dohmke, chief executive officer of GitHub, wrote in a blog post.
You’ll see its reasoning through session logs, which trace every decision it makes. Developers can also leave comments on pull requests, and Copilot will respond by editing the code accordingly.
GitHub’s agent is tightly integrated with GitHub Actions, the company’s CI/CD platform that runs more than 40 million daily jobs. That means the agent works within your current workflow, not outside of it.
And with the Model Context Protocol (MCP), it can use external data sources and follow custom repository instructions, helping it stay aligned with your team’s coding style and intent.
Availability and cost
The agent is currently in preview and only available to users of Copilot Pro+ and Copilot Enterprise.
To activate it, you’ll need to enable it in your repository settings. For Enterprise users, admins must turn on the policy. Starting June 4, 2025, using the coding agent will cost one premium request per model call.
GitHub says the agent is available on GitHub’s website, mobile app, CLI, and more IDEs, including Xcode, Eclipse, JetBrains, and Visual Studio.
GitHub levels up in the AI coding race
With the debut of its AI agent, GitHub is stepping into a more competitive space. Other tech giants have also entered the AI coding agent race: Google unveiled Jules last year, and OpenAI recently showcased its Codex agent.
Still, GitHub’s advantage lies in its seamless integration into the everyday developer workflow. And with over 15 million Copilot users and counting, the company hopes to redefine what it means to work with AI in software development.
