Rubber Duck Thursday!

Key Concepts

• GitHub Universe: A major GitHub event, with both in-person and free virtual attendance options.
• AI Adoption Pace: The rapid advancement of AI technology is outpacing the availability of education and training for its use in professional settings.
• GitHub Copilot: A suite of AI-powered tools for developers, including IDE extensions, a CLI, and features on github.com.
• Model Agnosticism: GitHub Copilot's ability to integrate with and utilize various AI models from different providers.
• Model Picker: A feature within GitHub Copilot that allows users to select preferred AI models.
• Auto Model Picker: An intelligent feature that automatically selects the most suitable AI model for a given task.
• Claude Haiku 4.5: A new, cheaper AI model now generally available in GitHub Copilot.
• Gemini 2.5 Pro: A powerful AI model with a large context window, now fully available in GitHub Copilot.
• GitHub Copilot CLI: A command-line interface for GitHub Copilot, enabling its use in terminals.
• GitHub Copilot Spaces: A feature for organizing files, pull requests, and issues, allowing for customized AI-powered development environments.
• Knowledge Bases to Spaces Conversion: The ability to convert existing knowledge bases into Copilot Spaces.
• MCP Servers: A component within Copilot CLI for managing and configuring servers.
• Multi-line Input (Copilot CLI): Support for multi-line input in the Copilot CLI via Shift+Enter in compatible terminals.
• Sub-agents/Agents within Terminal: A feature request for creating specialized agents within the terminal environment, similar to cloud code platforms.

GitHub Universe and AI Adoption

The stream begins with a welcome from Kadesha, the host of "Rubber Duck Thursday." She apologizes for missing the previous week due to preparations for GitHub Universe. Kadesha highlights the intensely fast pace of AI development and expresses concern that the pace of AI adoption and the expectation for professionals to use this technology are outpacing the available education and training. She invites viewers to share their thoughts on this topic.

GitHub Universe is announced to be happening the following week. Kadesha strongly encourages viewers to attend virtually for free by signing up at githubuniverse.com. She mentions that virtual attendees may gain access to specific features. For those attending in person, she invites them to say hello.

GitHub Copilot Updates and Features

The main focus of the stream shifts to recent updates and features within GitHub Copilot. Kadesha shares her screen to go through the change log.

New and Available AI Models

• Claude Haiku 4.5: This model is now generally available in all supported IDEs, not just VS Code. It is noted as a cheaper model, though Kadesha expresses a personal preference for Claude Sonnet 3.5 and Claude Sonnet 4 for her own use, primarily for brainstorming and planning.
• Gemini 2.5 Pro: This model is now fully available and no longer in preview. Kadesha emphasizes its large context window (over a million tokens) and its effectiveness for tasks like planning and evaluating different media types (video, audio). She believes it is "underrated" and "not spoken about enough."

Model Management and Selection

• Model Agnosticism: A key feature highlighted is GitHub Copilot's model-agnostic nature. Users can integrate their own models from providers like OpenRouter, Google, OpenAI, Olama, and Azure by adding their API keys.
• Model Picker: Within the IDE (specifically VS Code is demonstrated), users can access a model picker to select which models appear in their dropdown. This helps to manage a potentially long list of available models.
• Auto Model Picker: This intelligent feature automatically selects the most appropriate model for a given task. For example, it might use Gemini 2.5 Pro for planning and Claude Haiku 4.5 or GPT 4.1 for implementing tests. Kadesha finds this feature very useful.

GitHub Copilot CLI

Kadesha inquires about viewers' experiences with the GitHub Copilot CLI, calling it "chef's kiss" and a "really good addition." She clarifies the distinction between the CLI and the VS Code extension:

• VS Code Extension: For use within code editors.
• GitHub Copilot CLI: For use in terminals.
• GitHub.com: For use on the GitHub platform itself (assigning issues, etc.). She states that the goal is to allow users to use Copilot where they work best, rather than one being inherently "better" than the other.

GitHub Copilot Spaces

• Functionality: Copilot Spaces are presented as a way to create customized environments for organizing files, pull requests, and issues. Kadesha uses it for gathering and writing scripts for "The Download" show.
• Creation Process: Users can create a new space, name it, select models, add files, repositories, or upload documents as context. It's compared to "projects" in ChatGPT and Claude, where additional context and custom instructions are provided to define Copilot's role, focus, and goals.
• MCP Server Integration: A significant update is the ability to install MCP servers via the GitHub MCP registry within Copilot Spaces.
• Knowledge Base Conversion: Users can now convert their existing knowledge bases into Copilot Spaces, which is highlighted as a great enhancement.

Copilot CLI Improvements

The change log details recent improvements to the Copilot CLI:

• Multi-line Input: Support for multi-line input via Shift+Enter in terminals that support the Kitty protocol. The integrated terminal in VS Code can be configured for this by running /terminal setup.
• Claude 4.5 Support: Integration of the Claude 4.5 model.
• MCP Server Enhancements: Easier setup for MCP servers. The process of adding an MCP server involves specifying the server name, type, start command, environment variables, arguments, and available tools.

Kadesha demonstrates the Copilot CLI, showing how to add MCP servers and send requests to GitHub Copilot for tasks like adding tests to a component. She also shows the "banner" animation that appears when using certain commands.

Viewer Interaction and Questions

Throughout the stream, Kadesha actively engages with the audience, asking questions and responding to comments:

• Geographic Reach: Viewers join from various locations including Charlotte, Auckland, Brisbane, Silicon Valley, Ukraine, Seattle, Bangladesh, India, Poland, Mexico, Cyprus, France, Kenya, Paris, Brazil, Sri Lanka, and Italy, highlighting the global nature of the audience.
• AI Model Evaluation: Kadesha asks how viewers evaluate new AI models. Hashim shares an approach using a standard prompt to create an employee directory.
• Copilot CLI vs. VS Code Extension: A viewer asks about the benefits of using the Copilot CLI versus the VS Code extension. Kadesha explains it's about user preference and where they prefer to work.
• GitHub Universe Registration: A viewer reports being asked for registration charges for GitHub Universe. Kadesha investigates and clarifies that signing up for updates via email on the GitHub Universe website is the correct way to be added to the list for the free virtual event. She confirms this with a message to the team.
• GPT-5 CodeX: Kadesha admits she has not yet tried GPT-5 CodeX and it's not in her workflow.
• Coding Agents and Supervisors: A viewer describes creating a YOLO supervisor as a GitHub Action to provide guidance and corrective measures when delegating to a coding agent. Kadesha expresses strong interest in seeing this in action.
• Sub-agents in Terminal: A viewer asks if it's possible to create agents within the terminal like on cloud code platforms. Kadesha states that this feature is not currently available but acknowledges it as valuable feedback for the team.
• AI for Social Good: Douglas shares an example of a community pantry application that uses AI to guide users to services, which Kadesha praises as technology used for social good.

Conclusion and Wrap-up

Kadesha reiterates that the stream is a chill, relaxed session focused on chitchatting, answering questions, and going through the change log. She mentions that GitHub is in "full universe prep mode," leading to an intense period of preparation.

She concludes by encouraging viewers to attend GitHub Universe virtually or in person. She provides advice for those attending in person, suggesting to give her space if she appears overwhelmed at conferences. She emphasizes that the virtual attendance is 100% free.

The stream ends with a final confirmation about the GitHub Universe virtual pass registration process: sign up for updates via email on the website to be added to the list for the free virtual event.