Rubber Duck Thursdays! Let's talk about Microsoft Build!

Key Concepts

• GitHub Copilot App: A standalone desktop application for managing AI-driven development workflows, including repository management, PR/issue tracking, and agentic automation.
• Git Worktrees: A Git feature that allows multiple branches of a repository to be checked out into separate directories simultaneously, enabling parallel development without merge conflicts or stashing.
• Agentic Workflows: The use of AI agents to perform tasks (coding, planning, testing) in parallel, managed through the Copilot app.
• Rubber Ducking: A feature within GitHub Copilot that uses a secondary, different AI model to vet plans or code generated by the primary model to identify logic gaps or biases.
• MCP (Model Context Protocol): A standard used to connect AI agents to external tools, data sources, and services (e.g., email, Slack, or web testing tools).
• Canvases: Interactive, UI-based interfaces within the Copilot app that allow for bidirectional communication between the user, the AI agent, and project management tools.

---

1. GitHub Copilot App Overview

The GitHub Copilot app is designed to be a comprehensive development hub. It leverages the Rust SDK to provide the same functionality as the Copilot CLI but with a graphical user interface. Key features include:

• Session Management: Users can run multiple AI sessions in parallel.
• Model Selection: Users can choose different reasoning models (e.g., GPT, Claude, or the new Microsoft MAI model) based on the task requirements.
• Easter Egg: A hidden game featuring "Mona" (the GitHub mascot) can be accessed by dragging the icon in the app.

2. Git Worktrees: Parallel Development

The app defaults to using Git Worktrees for sessions.

• Mechanism: Unlike standard branching, which switches the state of a single directory, a worktree clones the repository structure into a separate folder.
• Benefits: It allows developers to work on multiple features or issues simultaneously without needing to stash changes or deal with merge conflicts.
• Trade-offs: Because each worktree is a separate folder, dependencies (like node_modules or Python virtual environments) must be installed per worktree, which can be resource-intensive.

3. Advanced AI Features

• Rubber Ducking: This feature dispatches a sub-agent using a different model family (e.g., if the primary is Claude, it might use GPT) to review the proposed plan. This cross-model verification is statistically shown to improve code quality and reduce errors.
• Automations: Users can set up recurring tasks (e.g., daily standup summaries, changelog generation, or service health checks) that run in the cloud or locally.
• Canvases: These act as "glorified web pages" with specific APIs that allow users to interact with agents visually. For example, a Kanban board canvas can be created where dragging a card triggers an AI session to work on that specific issue.

4. Microsoft Build Announcements & Updates

The speaker highlighted several recent releases:

• Agentic Workflows: Now in public preview, these no longer require personal access tokens (PATs) and include built-in security guardrails.
• Copilot CLI: New /settings command for centralized configuration.
• Model Updates: Claude 3.5 Sonnet is now available in GitHub Copilot.
• CodeQL: Ongoing security improvements for automated vulnerability scanning.

5. Notable Quotes

• "Rubber ducking is a cool feature... it consults other models for you... we've found that there's an actual statistically significant improvement when you rubber duck in terms of the code that stays in your codebase." — Cassidy Williams
• "Work trees are like tabs in a browser... it's a really nice way of working without having to deal with the stepping on each other and stashing amongst all of your different changes." — Cassidy Williams (referencing a viewer's analogy)

6. Synthesis and Conclusion

The GitHub Copilot app represents a shift toward "agentic development," where the IDE is no longer just a text editor but a command center for AI agents. By integrating Git worktrees, MCP servers, and multi-model verification (Rubber Ducking), GitHub is enabling developers to handle complex, parallel tasks with greater efficiency. While the app is currently optimized for paid users, the focus remains on reducing the friction of context switching and manual environment management. The platform continues to evolve rapidly, with a heavy emphasis on providing multiple "surfaces" (CLI, VS Code, and the standalone App) to suit different developer workflows.