Rubber Duck Thursdays - Copilot coding agent and Copilot agent mode

Key Concepts

• GitHub Enterprise Cloud with data residency
• GitHub Copilot Premium request reports
• GitHub Copilot app modernization (Java, .NET)
• Agent mode in JetBrains, Eclipse, and Xcode
• Model Context Protocol (MCP)
• Grock 3, Claude Sonet 4, Claude Opus 4 (GitHub Models)
• Copilot Coding Agent
• Secret protection
• CodeQL
• Dependency auto submission
• Copilot spaces
• GitHub Actions
• Copilot custom instructions
• Issue scoping

GitHub Updates and Announcements

GitHub Enterprise Cloud with Data Residency

• Now available for self-service trials, particularly useful for enterprises with data residency requirements.

GitHub Copilot Premium Request Reports

• Available since May 16th.

GitHub Copilot App Modernization

• Java and .NET modernization capabilities are now in public preview.
• Focuses on agentic AI, targeting specific tasks and scenarios.

Agent Mode Expansion

• Agent mode is now available in JetBrains (IntelliJ), Eclipse, and Xcode.
• Includes terminal execution, error fixing, task completion, and MCP support.
• MCP allows bringing in external context from tools like Notion.

GitHub Models Updates

• Grock 3 is now available in public preview.
• Users can experiment with different models.
• Models can be built into repositories in public preview.
• Prompts can be stored in YAML files within the repository.
• Allows comparison of different models and inputs.
• Evaluators can be added to assess model appropriateness, similarity, relevance, and groundedness.
• JSON schema support in GitHub Models AI develop tooling.

Copilot Coding Agent

• Asynchronous task assignment to Copilot via GitHub issues.
• Copilot works on the issue in the background and creates a pull request.
• Builds on existing GitHub workflows (issue assignment, pull request review).
• Requires Copilot Pro Plus or Copilot Enterprise subscription.
• Requires enabling Copilot coding agent in organization settings.
• Available on the mobile app.

Secret Protection

• Expanded default pattern support for various services (Block Protocol, Data Dog, Docker, Grock, Heroku, MongoDB, Salesforce, XAI).
• Secret Scanning Alerts API now supports hiding secret literals.

CodeQL

• Added support for Coplin 220 in version 2213.
• No longer detects hardcoded secrets (use secret protection instead).
• Incremental security analysis makes CodeQL up to 20% faster in pull requests.
• Can now be enabled at scale on C or C++ repositories in public preview using build free scanning.
• Enhanced metrics for CodeQL Quest alerts.

GitHub Copilot Models

• Claude Sonet 4 and Claude Opus 4 are now in public preview.
• Sonet 4 is available to all paid Copilot plans.
• Opus 4 is available to Copilot Enterprise and Pro Plus plans.
• GPT-4 Turbo model update for code completions with a more recent knowledge cut-off of March 2025.

Dependency Management

• Dependency auto submission now supports Gradle.
• Dependabot support for virtual network and actions runner controller is generally available.
• Easier to grant Dependabot access to repositories from the organization level.

Copilot Code Review

• Generates 80% more comments per pull request.
• Focuses on code quality, correctness, clarity, style, robustness, and accessibility.
• Supports over 900 languages.

Copilot Spaces

• Centralizes instructions and attachments for specific topics.
• Allows creating subject matter expert spaces for documentation, onboarding, etc.
• Can be shared across an organization.

Other Updates

• Sunset of automatic watching of repositories and teams.
• Consistent billing of Copilot seats.
• More control with personal settings for Copilot Pro users.
• Custom roles can now review secret scanning alert dismissal requests.
• Increasing GitHub enterprise importers repository size limits.
• Enterprise managed user repository collaborators available.
• GitHub Enterprise Server 3.17 is now generally available.
• Releases now expose digests for release assets (SHA-256 checksum).
• Copilot chat now supports attaching references using the @ symbol.
• Deepseek R10528 is now generally available in GitHub models.
• Copilot autofixes.

Copilot Coding Agent Deep Dive

Workflow Demonstration

• The speaker demonstrates using Copilot Coding Agent on a "Copilot Airways" application.
• Assigns issues to Copilot, which then creates pull requests.
• The process is asynchronous, with Copilot working in the background.
• The speaker reviews the changes made by Copilot in the pull request.
• The speaker merges a pull request that updates the GitHub Actions workflow for CI/CD.
• The speaker shows how to create a GitHub issue from a Notion document using MCP.

Key Factors for Successful Copilot Coding Agent Usage

1. Well-Scoped Issues:
   • Issues should be solvable in a single pull request.
   • If not, decompose the issue into smaller, more manageable tasks.
2. Issue Structure:
   • Include an overview or problem statement.
   • Provide a detailed description of the problem.
   • Define clear acceptance criteria.
   • Give pointers on which files need to be changed.
3. Copilot Custom Instructions:
   • Provide context about the codebase (e.g., framework, coding standards).
   • Specify commands for building, testing, and linting.
   • Define code styling conventions.
4. Copilot Setup Steps:
   • Use a GitHub Actions workflow to set up Copilot's environment.
   • Install necessary tools and dependencies (e.g., Node.js, npm).
   • Provides a predictable baseline for Copilot to work from.
5. Model Context Protocol (MCP):
   • Use MCP to access external tools and services (e.g., Notion, Playwright).
   • Configure MCP servers and grant access to Copilot.

Examples

• Refactoring a large form: Copilot refactors a 900-line form into smaller components while maintaining 100% test coverage.
• Updating GitHub Actions workflow: Copilot updates the workflow to include deployment to GitHub Pages.
• Implementing user reviews from Notion: Copilot uses MCP to access requirements in a Notion document and create a pull request.

Notable Quotes

• "Don't play koi with co-pilot" - Christopher (colleague), emphasizing the need to be explicit and clear in instructions.
• "Is this solvable in a single pull request?" - Product Manager, highlighting the importance of well-scoped issues.

Technical Terms

• Agentic AI: AI focused on specific tasks and scenarios.
• Model Context Protocol (MCP): A protocol for bringing in external context from other tools.
• Code Spaces: Cloud-based development environments.
• GitHub Actions: A CI/CD platform.
• Copilot Custom Instructions: Instructions that provide context about the codebase and coding standards.

Conclusion

The session provided a comprehensive overview of recent GitHub updates, with a particular focus on Copilot Coding Agent. The speaker demonstrated the capabilities of Copilot Coding Agent and emphasized the importance of well-scoped issues, clear instructions, and proper environment setup for successful usage. The session also highlighted the role of MCP in integrating Copilot with external tools and services. The main takeaways are that Copilot Coding Agent can significantly improve developer productivity, but it requires careful planning and configuration to achieve optimal results.