GitHub Unveils AI Agent That Autonomously Writes and Fixes Code

At Microsoft’s Build developer conference on Monday, GitHub introduced a new Copilot coding agent designed to autonomously handle specific programming tasks, notify developers when complete, and integrate its work directly into existing codebases.

The AI-powered agent can add features, fix bugs, refactor code, enhance documentation, and extend tests—particularly in well-tested codebases.

"Using state-of-the-art models, the agent excels at low-to-medium complexity tasks in well-tested codebases, from adding features and fixing bugs to extending tests, refactoring code, and improving documentation," GitHub said in a blog post.

Developers can assign tasks through GitHub, receive progress updates, review the agent’s output, request changes, and merge the code—all from within the platform.

"When you assign one or more issues to Copilot, it responds with a 👀 reaction and spins up a secure sandbox using GitHub Actions. It clones the repo, configures the environment, and analyzes the codebase with RAG powered by GitHub code search. The coding agent can use local tools and MCP servers, configured by you," Thomas Dohmke, CEO at GitHub, said.

CNBC reports that the AI agent is powered by Anthropic’s Claude 3.7 Sonnet model.

Available in preview, the coding agent will be accessible via Copilot Pro+ and Copilot Enterprise plans—not the free tier.