GitHub has introduced a major update to its AI-powered coding assistant, GitHub Copilot, featuring a preview of a new agent mode capable of self-iteration and autonomous error correction. Additionally, GitHub announced that Copilot Edits is now generally available in Visual Studio Code.

Revealed on February 6, these enhancements mark a transition for Copilot from a simple AI pair programmer to an advanced agentic development partner, GitHub stated. This shift underscores the increasing role of AI agents in modern software engineering.

Copilot Edits enables developers to select specific files and instruct Copilot in natural language to make real-time modifications across multiple files simultaneously. The new agent mode allows Copilot to refine its own output iteratively, recognize mistakes, propose terminal commands, and analyze runtime errors with built-in self-correction mechanisms.

To use agent mode, developers must install VS Code Insiders and enable the agent mode setting in GitHub Copilot Chat.

Additional features currently in preview include:

• Enhanced AI models, such as Google Gemini 2.0 Flash and OpenAI o3-mini, integrated into Copilot Chat. Administrators have access to organization-wide control options.
• Next edit suggestions, which streamline code changes by predicting subsequent modifications based on previous edits. Users can implement these changes across an open file with a single keystroke.
• Prompt files, which enable users to create and share reusable instructions within their VS Code workspace. These markdown-based blueprints combine natural language prompts, file references, and code snippets for improved efficiency.
• Vision for Copilot, a feature that generates UI elements, alternative text, and code by analyzing provided images or screen captures.

GitHub also introduced “Project Padawan,” an initiative aimed at developing autonomous software engineering agents capable of executing entire tasks independently. This vision anticipates a future where developers assign GitHub issues to Copilot, allowing AI to complete the tasks autonomously before developers review the final output.