GitHub is leveling up its AI coding assistant with a new autonomous agent mode that allows Copilot to write, debug and deploy code with minimal human oversight. The update marks a significant evolution for the Microsoft-owned platform as it faces growing competition from AI coding startups.

The new agent mode enables Copilot to iterate on its own code output, recognize and fix errors automatically, and even suggest relevant terminal commands - capabilities that push it closer to being a true "peer programmer" rather than just an assistant. The system can also infer additional required tasks that weren't explicitly specified in the original prompt.

"Developer teams will soon be joined by teams of intelligent, increasingly advanced AI agents that act as peer-programmers for everyday tasks"— Thomas Dohmke, GitHub CEO

GitHub is also previewing "Project Padawan," a more ambitious initiative that would allow developers to assign entire coding tasks to an autonomous Copilot agent. The system could potentially handle everything from initial development to setting up repositories and requesting human code review.

The moves come as GitHub faces increased competition from AI coding startups like Cursor, Replit, and Bolt, which have already launched their own agent-based features. But with over 77,000 organizations already using Copilot and its deep integration with Microsoft's developer tools, GitHub maintains significant advantages in the increasingly crowded AI coding space.

The agent mode preview is currently only available to users of VS Code Insiders, with no details yet on pricing or general availability. GitHub says it's also adding support for new AI models from Google and Anthropic to give developers more options beyond OpenAI's GPT-4.

Source: GitHub Blog