What's new with the Copilot coding agent | GitHub Checkout

GitHub Copilot Coding Agent: New Features and Future Directions

Key Concepts:

• Copilot Coding Agent: An autonomous developer agent that performs coding tasks in the background.
• Agents Panel: A new overlay on GitHub for managing and assigning tasks to Copilot.
• MCP (Managed Compute Platform) Servers: Servers that provide Copilot with specific capabilities, like web browsing.
• Playwright MCP Server: Enables Copilot to browse websites, take screenshots, and test web applications.
• GitHub MCP Server: Allows Copilot to access issues and pull requests within a repository.
• Tech Debt: Refactoring, improving test coverage, and upgrading dependencies.
• Premium Requests: Units used to measure Copilot Coding Agent usage.

1. Introduction to Copilot Coding Agent

• Tim introduces the GitHub Copilot Coding Agent as an autonomous developer agent that can handle coding tasks independently.
• The key differentiator is its ability to run in the background, allowing developers to delegate tasks and review the results later.
• Copilot validates its work, creates pull requests, and tags the developer for review without requiring constant supervision.

2. Agents Panel: A New Interface for Task Management

• The Agents Panel is a new overlay on GitHub that simplifies task assignment and tracking for Copilot.
• It allows users to easily send new tasks to Copilot and monitor the progress of ongoing tasks.
• Users can access the panel with a button, which automatically selects the current repository.
• Example: A user can instruct Copilot to "Upgrade Node.js to the latest version" directly from the panel.
• The panel also allows users to select a specific branch for the task.
• Mobile app integration is planned, with the "tasks" experience coming to GitHub Mobile.

3. Web Browsing Capabilities with Playwright MCP Server

• Copilot now has its own web browser, enabling it to work more effectively on web applications.
• The Playwright MCP server allows Copilot to test its work and understand how web applications function.
• Example: Copilot added internationalization support to a website and provided a screenshot of the changes in the pull request.
• This feature allows developers to quickly validate Copilot's work without diving into the code or checking out the branch.
• The Playwright MCP server enables Copilot to browse websites, including those running on a local server, and take screenshots.

4. MCP Servers and Their Applications

• Two MCP servers are enabled by default: Playwright (for web browsing) and GitHub (for accessing repository information).
• The GitHub MCP server allows Copilot to examine other issues and past pull requests for inspiration.
• Users can add their own MCP servers, such as Notion, to allow Copilot to access notes and scoping documents.
• The ability to connect to various MCP servers enhances Copilot's capabilities and allows it to integrate with different tools and platforms.

5. Use Cases: Tech Debt and Feature Development

• The primary use case for Copilot Coding Agent is addressing tech debt, such as refactoring, improving test coverage, and upgrading dependencies.
• Copilot excels at these tasks because they often don't require extensive intelligence or deep thinking.
• Examples: Small refactors to improve code style or removing feature flags.
• Copilot can also be used to kickstart new feature development, completing the initial 70-80% of the work.
• Developers can then pick up the remaining tasks and finalize the feature.

6. Cost and Pricing

• Copilot Coding Agent costs one premium request each time a user asks it to create or modify a pull request.
• 300 premium requests allow for the creation of up to 300 pull requests.
• Copilot uses GitHub Actions under the hood, so GitHub Actions minutes are also consumed.
• If GitHub Actions minutes are available for free (e.g., in open-source repositories), Copilot can be used without additional cost.
• Copilot Coding Agent is now available on all paid plans.

7. Future Developments

• Future plans include the ability to mention Copilot in pull requests to request fixes for tests or build failures.
• A public API is being developed to allow users to send tasks to Copilot programmatically, enabling new automation possibilities.
• Improvements are being made to allow Copilot to more easily fix GitHub Actions build failures.

8. Feedback and Community Engagement

• Users can provide feedback through the GitHub community, X, Reddit, and LinkedIn.
• Direct email communication with the development team is also encouraged.
• The goal is to make the product as useful as possible for users based on their feedback.

9. Conclusion

• The GitHub Copilot Coding Agent is a powerful tool that can significantly enhance developer productivity.
• The Agents Panel, web browsing capabilities, and integration with MCP servers provide a comprehensive set of features for automating coding tasks.
• By addressing tech debt and assisting with feature development, Copilot can help developers focus on more complex and creative work.
• The future of Copilot Coding Agent is bright, with ongoing development focused on improving its capabilities and expanding its integration with other tools and platforms.