Overview
When GitHub Copilot was introduced in 2021, the goal was to assist developers with an AI-powered pair programmer. Over time, Copilot has evolved beyond simple code suggestions, offering features like automated pull request feedback, security vulnerability fixes, and brainstorming solutions for coding problems.
Now, GitHub Copilot is entering a new phase with Agent Mode, General Availability (GA) of Copilot Edits, and a first look at Project Padawan, an autonomous Software Engineering (SWE) agent.
🚀 GitHub Copilot Agent Mode: Available in Preview
What is Agent Mode?
Agent mode enables Copilot to:
- Iterate on its own code
- Recognize errors and fix them automatically
- Suggest terminal commands for execution
- Analyze run-time errors with self-healing capabilities
How Does It Work?
Instead of merely following a user’s request, Agent Mode infers additional necessary tasks, catches its own errors, and iterates until the task is fully completed. This minimizes the need for developers to manually copy and paste commands between the terminal and chat.
How to Enable Agent Mode
To use Agent Mode, follow these steps:
- Download VS Code Insiders
- Enable the Agent Mode setting in GitHub Copilot Chat
- Switch from ‘Edit’ to ‘Agent’ in the Copilot Edits panel
Future Plans
Agent Mode will expand to all IDEs that Copilot supports, and user feedback is welcome to refine its functionality.
🎉 Copilot Edits: Now Generally Available (GA) in VS Code
What is Copilot Edits?
Copilot Edits allows developers to make inline changes across multiple files with a conversational flow. It combines:
- Chat and Inline Chat capabilities
- Multi-file editing support
- Fast iteration on code changes
How It Works
- Select a set of files for editing
- Use natural language to request modifications
- Copilot suggests edits inline in your workspace
- Accept, reject, or iterate on the suggestions
- Verify changes with unit tests and debugging tools
Enhanced Editing with AI
Copilot Edits leverages a dual-model architecture:
- A foundation model (GPT-4o, o1, o3-mini, Claude 3.5 Sonnet, or Gemini 2.0 Flash) generates initial suggestions.
- A speculative decoding endpoint ensures fast application of changes.
Key Features
- Supports transitions between Copilot Chat and Copilot Edits
- Suggests files to include in edits
- Undo specific suggested changes
- Voice interaction for a natural experience
Future Roadmap
- Improve speculative decoding performance
- Enhance integration between Copilot Chat and Copilot Edits
- Expand support for Visual Studio 2022
🤖 Project Padawan: The Future of SWE Agents on GitHub
What are SWE Agents?
Software Engineering (SWE) agents assist developers by:
- Generating and reviewing code
- Refactoring and optimizing projects
- Automating workflows like testing and CI/CD pipelines
- Providing guidance on architecture and best practices
SWE agent performance is measured using SWE-bench, a dataset of 2,294 Issue-Pull Request pairs from 12 popular Python repositories.
How Project Padawan Works
- Developers can assign GitHub issues directly to Copilot.
- Copilot clones the repository and sets up a secure cloud sandbox.
- It analyzes code, edits files, builds, tests, and lints code.
- Upon completion, it assigns human reviewers and iterates based on feedback.
Future Vision
Project Padawan will transform software development by automating critical yet routine tasks like bug fixes and test automation, allowing developers to focus on more valuable work.
🔥 Conclusion
GitHub Copilot continues to evolve, making software development faster and more efficient. With Agent Mode, Copilot Edits, and Project Padawan, AI will play an even greater role in helping developers create better code while maintaining full control over their projects.