Rubber Duck Thursdays - Time to build!

Key Concepts

• GitHub Actions: Automation workflows directly within GitHub repositories.
• MCP (Model Context Protocol): A standardized approach for AI agents to interact with tools and resources.
• Elicitation: The process of gathering additional information needed for a tool call in MCP.
• Copilot: An AI pair programmer that offers code suggestions and assists with development tasks.
• 3JS: A JavaScript library for creating and displaying 3D graphics in a web browser.
• Playwright: A framework for end-to-end testing of web applications.
• Custom Instructions: Directives provided to Copilot to guide its behavior and coding style.
• Coding Agent: An automated agent within Copilot that can perform more complex coding tasks.

GitHub Actions and MCP Server Development

The speaker discusses ongoing projects involving GitHub Actions and an MCP (Model Context Protocol) server. The MCP server is designed for turn-based games like tic-tac-toe and rock-paper-scissors. The speaker has been experimenting with visualizing the tic-tac-toe board in 3D using 3JS and Copilot.

3D Tic-Tac-Toe Visualization

The speaker demonstrates a 3D tic-tac-toe game built using 3JS, highlighting the use of Copilot to accelerate development. The 3D visualization allows players to interact with the game in a more immersive way. The speaker acknowledges some issues, such as the X and O pieces standing upright instead of lying flat, and usability problems with camera controls.

Example: The speaker shows how the 3D game integrates with the existing MCP server, allowing seamless switching between 2D and 3D views while maintaining the same game state.

Process: The speaker describes using Copilot to generate the initial 3D game UI, then iterating on the code to refine the layout, navigation, and camera controls.

Copilot and Development Workflow

The speaker emphasizes the role of Copilot in reducing the "activation energy" for starting new projects. Copilot allows developers to focus on architectural decisions and high-level logic, rather than getting bogged down in the details of code implementation.

Example: A viewer shares a project where they used LLMs to build a Go agent that monitors a camera feed and detects when their baby wakes up, sending notifications via WhatsApp and Home Assistant.

Argument: The speaker argues that Copilot does not replace the need for domain expertise. Developers still need to provide context and guidance to Copilot to ensure that the generated code meets their specific requirements.

GitHub Announcements

The speaker reviews recent updates to GitHub, including:

• GitHub Copilot and Visual Studio August Update: GT5 model support, MCP support generally available, smarter Copilot chat, bring your own model, sign up with Google, and greater control over Copilot suggestions.
• GPT-4 Mini Public Preview: Available in Visual Studio, JetBrains IDEs, Xcode, and Eclipse.
• Improvements to the Home Dashboard: A customizable view of pull requests and issues.
• Copilot Coding Agent Tasks from Raycast: Integration with the Raycast launcher.
• Agents.mmd Custom Instructions: Support for a standardized format for providing instructions to agents.
• WEBP Image Support: Support for the WEBP image format in issues, discussions, pull requests, and gists.
• Nextedit Suggestions in JetBrains IDEs: Public preview of a feature that provides multiple code suggestions.
• GraphQL API Resource Limits: Introduction of limits to safeguard resource consumption.
• Improved Notifications in Security Campaigns: Email notifications for developers with right access to repositories.
• License-Based Budgets: Budgeting based on the number of licenses instead of dollar amounts.
• CodeQL Updates: Support for Go 1.25, Rust enhancements, and Java/Kotlin updates.
• Copilot Code Review Path Scope to Custom Instruction: The ability to target custom instructions to specific file paths in a repository.

Significance: The speaker highlights the importance of the Copilot Code Review Path Scope to Custom Instruction feature, as it allows for more granular control over Copilot's behavior based on the location of the code being reviewed.

Model Context Protocol (MCP)

The speaker explains the purpose of MCP as a standardized way for AI agents to interact with tools and resources. MCP consists of a host, client, and server. The MCP server provides tools and resources that the AI agent can use.

Example: The speaker's tic-tac-toe game uses an MCP server to handle game logic and state management. Copilot interacts with the MCP server to create new games, make moves, and retrieve game state.

Elicitation: The speaker mentions a blog post coming out later that day about elicitation, which is the process of gathering additional information needed for a tool call in MCP.

Playwright and UI Testing

The speaker attempts to use Playwright to automate UI testing of the 3D tic-tac-toe game. The goal is to have Playwright inspect the position of the X and O pieces on the board and automatically adjust their placement. However, the attempt is unsuccessful due to issues with Playwright interacting with the browser used for the live stream.

Conclusion

The speaker summarizes the topics covered in the stream, including GitHub Actions, MCP server development, 3D tic-tac-toe visualization, Copilot, GitHub announcements, and Playwright UI testing. The speaker expresses enthusiasm for the progress made on the 3D tic-tac-toe game and the potential for Copilot to accelerate development. The speaker announces an upcoming stream later that day and another stream the following week.