At Microsoft's Build 2025 developer conference in Seattle, GitHub unveiled a groundbreaking AI coding agent that functions as an autonomous developer on software teams.
The new GitHub Copilot coding agent represents a significant evolution in AI-assisted development. Embedded directly into GitHub, the agent activates when developers assign GitHub issues to Copilot or prompt it through VS Code. It then spins up a secure development environment powered by GitHub Actions, pushing commits to a draft pull request that can be tracked through session logs.
The agent's workflow mirrors that of a human developer, automating repetitive tasks. After launching a virtual machine through GitHub Actions, it pushes progress updates as draft pull requests and records its decision-making in session logs. When finished, it tags the original user for review. Developers can then leave comments, which the agent automatically addresses by making additional changes.
According to GitHub CEO Thomas Dohmke, the agent is designed to perform efficiently in "well-tested codebases" and is particularly adept at managing routine maintenance such as bug fixes, feature extensions, and code refactoring. Using state-of-the-art models, the agent excels at low-to-medium complexity tasks, from adding features and fixing bugs to extending tests, refactoring code, and improving documentation. This allows developers to hand off time-consuming but mundane tasks while focusing on more interesting work.
The Copilot coding agent is available now for Copilot Pro+ and Copilot Enterprise subscribers. Enterprise users will need administrator approval to enable the new Copilot coding agent policy. Using the agent consumes GitHub Actions minutes and Copilot premium requests from entitlements included with subscription plans.
The technology validates its work with tests and linters before pushing changes. Once complete, it tags the user for review. Developers can ask for modifications through comments in the pull request or check out the branch locally to continue work in their IDE with Copilot's assistance. The agent particularly excels at low-to-medium complexity tasks in well-tested codebases.