GitHub's new AI agent writes and repairs code autonomously
19 May 2025

GitHub has unveiled a new AI coding agent that can perform tasks like fixing bugs, adding features, and improving documentation on behalf of developers.

The announcement came at the Microsoft Build conference.

The futuristic tool is a part of GitHub Copilot and gets to work as soon as a user gives it a particular task.

How this AI coding agent works
Operational process

To perform its tasks, the AI coding agent automatically launches a virtual machine, clones the repository, and analyzes the codebase.

As it works, it saves its changes and writes its reasoning in session logs.

Once the task is done, GitHub notifies users for review. Developers can then give feedback that the agent will handle automatically.

GitHub's AI agent understands project intent, coding standards
Contextual understanding

GitHub's AI agent is also trained to understand what you want to do and how you want to do it.

It achieves this by bringing in context from relevant issue or pull request conversations and following any custom repository guidelines.

This way, the AI agent can align its actions with the requirements of every project.

Availability and open-source development
Access and collaboration

The new AI coding agent is available to Copilot Enterprise and Copilot Plus users through GitHub's website, mobile app, and the GitHub Command Line Interface tool.

In a related development, Microsoft has also announced that it will be open-sourcing GitHub Copilot in Visual Studio Code.

This will enable developers to build on the tool's AI capabilities, encouraging further innovation in this space.