Rubber Duck Thursdays! | Let's code and cowork!

Key Concepts

• GitHub Copilot App: A standalone application built on the GitHub Copilot CLI and Rust SDK, designed for managing AI-assisted development workflows.
• TUI (Terminal User Interface): A text-based interface within the CLI that allows users to navigate issues, pull requests, and gists using keyboard shortcuts.
• MCP (Model Context Protocol): An open protocol that acts as a standardized interface for AI tools to connect with external data sources, APIs, and services.
• Git Worktrees: A Git feature that allows developers to have multiple branches checked out in separate directories simultaneously, enabling parallel development without constant stashing or switching.
• Agentic Development: A paradigm where AI agents perform tasks, manage workflows, and interact with various tools (like Playwright or Slack) to automate complex software engineering processes.

---

1. GitHub Copilot App and CLI Improvements

The speaker highlights the rapid evolution of the GitHub Copilot ecosystem. The new TUI allows developers to manage sessions, issues, and pull requests directly from the terminal.

• Key Features: Users can add context to chats by hitting the 'C' key on specific issues.
• BYOK (Bring Your Own Key): A highly requested feature allowing users to integrate their own model providers and local models into the Copilot app.
• Customization: The app supports themes, voice dictation, and the ability to add custom MCP servers and plugins (e.g., Slack, Work IQ).

2. GitHub Platform Updates

• Repository Switcher: A new UI element that allows users to jump between repositories without navigating back to the profile page or modifying URLs.
• Maintainer Protection: A feature to limit the number of pull requests from users without write access, helping maintainers manage spam and protect their time.
• GitHub Actions: Now supports asynchronous background actions, allowing steps to run concurrently.

3. Model Context Protocol (MCP)

The speaker explains MCP as a "glorified API surface" for AI, drawing a parallel to the Language Server Protocol (LSP).

• The Problem: Before MCP, developers had to build custom integrations for every individual AI tool (e.g., ChatGPT, Claude, Copilot).
• The Solution: MCP provides a universal standard. Once a server is configured, it works across any AI service that supports the protocol.
• Real-World Application: The speaker uses the Playwright MCP server to automate UI testing. The AI agent takes screenshots at various screen sizes during development, allowing the user to verify responsive design without manual window resizing.

4. Git Worktrees for Parallel Development

The speaker advocates for Git Worktrees as a superior alternative to traditional branching for parallel tasks.

• Methodology: Instead of stashing work, checking out a new branch, and reinstalling dependencies, a worktree creates a separate directory linked to the same repository.
• Benefits: Allows developers to work on a "hotfix" in one folder while keeping a feature branch active in another.
• Technical Gotcha: Because each worktree is a separate folder, it requires its own node_modules or dependency installation, which can lead to disk bloat. The speaker emphasizes "worktree hygiene"—deleting folders once the task is complete.

5. Canvas: The Future of Agentic UI

The "Canvas" is introduced as a non-chat interface for interacting with agents.

• Functionality: It acts as a visual workspace (described as "Notion for a repo") where users can orchestrate multiple agents across different repositories.
• Example: The speaker demonstrated a "Gesture Canvas" that uses a webcam to allow physical gestures (like a thumbs-up) to approve or reject pull requests, showcasing the potential for non-textual agent interaction.

---

Synthesis and Conclusion

The presentation underscores a shift toward agent-native development, where the focus is moving away from simple chat-based interactions toward orchestrated, multi-tool workflows. By leveraging MCP for connectivity, Worktrees for parallel task management, and Canvas for visual orchestration, GitHub is providing developers with a highly flexible, "choose-your-own-adventure" style of software engineering. The core takeaway is that modern tooling is increasingly designed to reduce context-switching fatigue and automate the "dance" of traditional Git and dependency management.