Rubber Duck Thursday - live from GitHub HQ!

Key Concepts

• GitHub Universe: An annual event hosted by GitHub to announce new features and updates.
• Copilot Coding Agents: AI-powered tools that assist developers in writing code, managing tasks, and creating pull requests.
• Custom Agents: User-defined agents that can be configured with specific tools and capabilities, allowing for specialized automation.
• Agents Tab: A new unified dashboard on GitHub.com for managing and tracking Copilot coding agent tasks.
• MCP Servers: Microservice-based servers that provide specific functionalities to agents.
• MCP Registry: A repository of MCP servers that can be integrated with custom agents.
• Copilot CLI: A command-line interface for interacting with GitHub Copilot.
• Copilot Usage Metrics Dashboard: A tool for organizations to track and analyze their team's Copilot usage.
• Plan Mode (VS Code): A feature in VS Code that allows Copilot to help plan out tasks and requests by asking clarifying questions.
• Agent HQ: A platform to bring various AI agents (from Claude, CodeX, Gemini, etc.) onto the GitHub platform for unified management and use.
• Aski Motion: An open-source project created by a brand designer using GitHub Copilot to generate ASCII art animations.
• Hacker Badges: Customizable badges given out at GitHub Universe, now featuring color displays and programmable capabilities.

GitHub Universe 2023: Key Announcements and Features

This summary details the significant updates and announcements made during GitHub Universe, focusing on advancements in AI integration, developer tooling, and community engagement. The presenter, Kadesha, shares her excitement and provides practical demonstrations of the new features.

1. Enhanced Copilot Coding Agents and New Dashboard

• Unified Agents Tab: A new "Agents" tab has been introduced on GitHub.com, serving as a central hub for managing Copilot coding agent tasks. This provides a unified view to start, share, assign, and track agent activities, eliminating the need to search across different PRs and tasks.
• Improved Task Management: Users can now switch between PRs that the coding agent is working on directly from this unified view. The interface allows for viewing the full PR, collapsing side panels, and navigating through changed files.
• Interactive Feedback: The system is not read-only; users can add comments directly to PRs, which are then sent as feedback to the Copilot coding agent. This allows for iterative refinement of agent tasks.
• New Homepage "Ask Anything" Box: The GitHub homepage now features an "Ask anything" box powered by Copilot, allowing users to query Copilot with context from repositories, spaces, and local files, and select preferred models and extensions.

2. Introduction of Custom Agents

• Empowering Specialization: Custom agents bring the functionality of "sub-agents" from tools like Cloud Code to GitHub Copilot. These agents can be created at the repository or organization level.
• Configuration: Custom agents are configured via markdown files. Key mandatory properties include name and description. Optional properties like tools and MCP servers enhance agent capabilities.
• MCP Servers and Tools: Agents can leverage MCP servers for specialized functionalities. The presentation highlights the Chrome Dev Tools MCP server and mentions the availability of an MCP registry for discovering and integrating these servers.
• Documentation-Driven Configuration: The process of configuring agents involves referencing the documentation of the specific MCP server or tool to correctly map commands, arguments, and types into the agent's front matter.
• Example: Front-end Developer Agent: A detailed example of a "front-end developer" agent is showcased, demonstrating its configuration with tools like read, edit, shell, search, and to-do, and access to the Chrome Dev Tools MCP server.
• Example: LaunchDarkly Agent: Another example, the "LaunchDarkly flag cleanup" agent, illustrates how custom agents can automate specific workflows like feature flag cleanup by interacting with external services.
• Open-Source Resources: The "GitHub Awesome Copilot" repository is recommended as a resource for finding examples of prompts, instructions, and pre-configured custom agents.

3. Copilot Usage Metrics Dashboard

• Public Preview Launch: A new Copilot usage metrics dashboard has been released in public preview.
• Organizational Insights: This dashboard provides a unified view for organizations to track Copilot adoption, model usage, active users, and other key insights into how teams are leveraging Copilot.

4. The Importance of Developers in the AI Era

• Keynote Themes: The keynote presentations emphasized the continued and growing importance of developers, especially with the integration of AI agents into software systems.
• Human Oversight: The presenter stresses that AI agents are tools to enhance, not replace, human developers. Developers are crucial for orchestrating these tools, troubleshooting, and ensuring the responsible deployment of AI.
• Curiosity as a Driver: The importance of leaning into curiosity to build and innovate is highlighted, drawing parallels to the origins of many essential software tools.
• Inspiring Examples: Students showcasing how they use GitHub Copilot to solve real-world problems and the "hacking Furby" demo by Martin Woodward are presented as inspiring examples of developer ingenuity.

5. "Hacking Furby" and Creative AI Applications

• Modernizing Legacy Code: Martin Woodward demonstrated how GitHub Copilot was used to modernize an old codebase for a Furby toy, creating a tool called "Pipluff."
• Resurrecting Old Tech: Pipluff allows users to reprogram and interact with older Furby Connect devices, showcasing the power of AI in breathing new life into legacy technology.
• Interactive Demos: The demo included making the Furby say "hello," change colors, giggle, and even sing, highlighting the creative potential of AI-assisted development.
• Furby Choir: A notable segment featured an engineering manager playing a "Furby choir," further emphasizing the fun and unexpected applications of AI.

6. Aski Motion: A Designer's AI-Powered Creation

• Bridging Design and Development: Cameron Foxley, a brand designer at GitHub, used GitHub Copilot to build an open-source project called "Aski Motion."
• ASCII Art Animation Tool: This tool allows users to create ASCII art animations by drawing directly on a canvas, applying gradients, and controlling playback with an animation timeline.
• Supercharging Creativity: The project exemplifies how Copilot can empower individuals with specialized skills (like design) to acquire technical skills and bring their creative ideas to life, even without a traditional engineering background.
• Practical Application: Aski Motion was developed to implement ASCII art in the Copilot CLI, demonstrating a direct application of AI-assisted creation for product features.

7. Agent HQ: Unifying AI Agents on GitHub

• Platform Integration: Agent HQ is a new initiative to bring various AI agents (from Claude, CodeX, Gemini, etc.) onto the GitHub platform.
• Unified Management: This aims to provide a unified space for users to manage, use, and interact with different AI agents.
• Native Dev Flow: The goal is to make agents native to the GitHub development workflow, accessible across CLI, mobile, web, and editors.
• Future Availability: Agent HQ is slated for release later this year, with more announcements to follow.

8. Plan Mode in VS Code

• Enhanced Planning Capabilities: Plan mode, currently available in VS Code Insiders, allows GitHub Copilot to assist in planning tasks and requests.
• Interactive Planning: Copilot asks clarifying questions based on the task and repository context to help developers think through implementation details.
• Detailed Plans: The feature can generate detailed plans, including endpoints, swagger docs, and phased implementation steps, which can be exported to product requirement documents.
• Overcoming Planning Challenges: This feature addresses the common challenge of project planning by providing AI-driven assistance.

9. Technical Demonstrations and Q&A

• MermaidJS Diagram Generation: A demonstration showed Copilot (using Claude Sonnet 4) generating an architectural diagram using MermaidJS. While initial rendering issues were encountered, the capability to generate such diagrams was confirmed.
• Agent HQ Functionality: In response to a question about Agent HQ, it was clarified that while the exact commit behavior is not yet detailed, it aims to make agents native to the platform and accessible across various interfaces.
• AI Literacy: The presenter emphasized the importance of developers becoming "AI literate" to effectively collaborate with AI agents in building software.

10. Conclusion and Call to Action

• Developer Empowerment: The overarching theme is that developers are more crucial than ever, and AI tools like GitHub Copilot are designed to augment their capabilities and foster innovation.
• Embrace AI: Developers are encouraged not to fear AI but to study and learn how to work with it effectively.
• Community and Curiosity: The importance of community and leaning into curiosity for building new ideas remains a core tenet of GitHub.
• Resources: Viewers are encouraged to check out the day one and day two keynotes from GitHub Universe, explore the "GitHub Awesome Copilot" repository for custom agent examples, and visit the provided link for building custom agents.
• Personal Connection: Kadesha invites further questions and discussions via LinkedIn.