Rubber Duck Thursday!

Key Concepts

• Rubber Ducking: A debugging technique where developers explain their code problem to an inanimate object (like a rubber duck) to clarify their thinking.
• Spec-Driven Development with AI: Using AI tools, particularly the GitHub SpecKit, to build projects based on specifications rather than traditional methods.
• GitHub Copilot Coding Agent: An AI tool that assists developers with coding tasks, including generating code, fixing bugs, and creating issues.
• Auto Model Selection: A feature in GitHub Copilot that automatically selects the most appropriate AI model (e.g., GPT-5, Claude Sonnet 4.5) for a given task.
• GitHub Copilot CLI: A command-line interface that allows developers to use Copilot's AI capabilities within their terminal.
• Model Selection: The process of choosing the most suitable AI model for a specific task, considering factors like cost, accuracy, and token limits.

GitHub Updates and New Features

Spec-Driven Development with AI

• GitHub has open-sourced SpecKit to facilitate spec-driven development using AI.
• This approach aims to improve the engineering process when building with AI tools, moving beyond simple "vibe coding" for initial MVPs.
• The presenter encourages viewers to explore the GitHub blog post on this topic.

One-Click Merge Conflict Resolution

• GitHub now offers one-click merge conflict resolution directly in the web interface.
• Users can accept current changes, incoming changes, or both, without needing to clone the PR locally or use a codespace.

Auto Model Selection in VS Code

• GitHub Copilot now features auto model selection in VS Code for Business and Enterprise users.
• Based on the prompt, Copilot intelligently chooses the most suitable model from options like GPT-5 mini, GPT-4.1, Claude Sonnet 4, and Sonnet 3.5.
• The presenter demonstrates this feature, showing how Copilot selects GPT-5 to implement a task from a GitHub issue.
• The selected model is indicated by hovering over the model's response.

New Repository with Copilot Coding Agent

• When creating a new repository on GitHub, users can now provide a prompt to Copilot Coding Agent to jumpstart the project.
• Copilot creates a draft pull request with the requested code.
• The presenter demonstrates this by creating a "Pomodoro Timer" repository and prompting Copilot to create a simple timer using HTML, CSS, and JS.
• Users who use this feature have a chance to win a gift card.

Anthropic Claude Sonnet 4.5 Public Preview

• Anthropic's Claude Sonnet 4.5 is now available in public preview for Copilot Coding Agent and in agent mode in VS Code.
• GitHub is swiftly integrating the latest models into Copilot.

Copilot Can Create Issues with Code Snippets

• Copilot can now create issues with code snippets in public preview.
• Users can instruct Copilot to create a task (e.g., removing a feature flag) and include relevant code snippets directly from the Copilot UI on github.com.

Deprecation of GH Copilot CLI

• The original GH Copilot CLI, which assisted with terminal commands, is being deprecated on October 25th.
• Users are encouraged to switch to the newer GitHub Copilot CLI, which offers more comprehensive coding agent capabilities in the terminal.

Demonstrations and Examples

Auto Model Selection Demo

• The presenter demonstrates auto model selection in VS Code by asking Copilot to implement a task from a GitHub issue.
• Copilot uses the GitHub MCP server to fetch the issue details and selects GPT-5 to perform the implementation.
• The presenter notes that hovering over the response reveals which model was used.

Copilot Coding Agent Demo

• The presenter showcases a personal project: an app that generates chapters for YouTube videos using AI.
• The presenter uses Copilot Coding Agent to debug an issue with the app, where it incorrectly identifies valid YouTube URLs as invalid.
• Copilot CLI uses Playwright MCP server to visually test the feature.
• The presenter demonstrates the Copilot CLI's ability to generate chapters, which can then be copied and pasted into YouTube video descriptions.
• The presenter credits a viewer for identifying and resolving the issue by removing "ant" from the URL.

Key Arguments and Perspectives

AI in Development

• The presenter addresses skepticism about using AI for idea generation, arguing that AI can be valuable for research and validation.
• AI can help developers quickly assess the viability of an idea by researching market demand and existing solutions.

Model Selection Considerations

• The presenter discusses the factors to consider when choosing an AI model, emphasizing that the latest model isn't always the best choice.
• For coding tasks, the latest models are generally preferred, but for applications using LLMs, cost-efficiency and accuracy are also important factors.
• The presenter notes that access to different models may depend on the user's Copilot plan.

Technical Terms and Concepts

• GitHub SpecKit: A toolkit for spec-driven development with AI.
• MCP (Model Collaboration Platform) Server: A server that provides access to various AI models and tools.
• LLM (Large Language Model): A type of AI model used for natural language processing tasks.
• Token Limits: Restrictions on the amount of text that can be processed by an AI model.
• API Key: A code used to authenticate and authorize access to an API.

Logical Connections

• The discussion of spec-driven development with AI leads into the demonstration of Copilot Coding Agent, highlighting how AI tools can be used to improve the development process.
• The demonstration of auto model selection connects to the broader discussion of model selection considerations, illustrating how Copilot can help developers choose the right model for their task.
• The presenter's personal project serves as a practical example of how Copilot can be used to build useful applications.

Notable Quotes

• "So developers often refer to talking to like an inanimate object as rubber ducking and it really helps us to just parse and just think about the problem logically step by step and really explain what we're experiencing to someone else that doesn't know what we're talking about." - Cadesha, explaining the concept of rubber ducking.
• "Vibe coding is really for building that initial MVP of a product. There's so much more that goes into an application before you actually ship it to users. And so, SpecKit aims to provide a solution for you to be more specific and following a like a more engineering process when you're building with AI tools." - Cadesha, explaining the purpose of SpecKit.
• "I think that's something that's very important for you to know is like, you know, we're always adding the newest, latest, and greatest models with the swiftness into GitHub copilot so that you have them available in your editors." - Cadesha, highlighting GitHub's commitment to providing access to the latest AI models.
• "It depends on what you're doing. So like if you're using if you're using a model for coding like in this case, of course, use the latest and greatest models. But if you're building an application that uses an LLM... I would say you would choose the most costefficient model and you would you choose the one that's gives you, you know, the most accurate results." - Cadesha, explaining the factors to consider when choosing an AI model.

Synthesis/Conclusion

The stream covers several new features and updates to GitHub Copilot, including spec-driven development with AI, one-click merge conflict resolution, auto model selection, and the ability to create new repositories with Copilot Coding Agent. The presenter demonstrates these features and discusses the considerations for choosing the right AI model for different tasks. The stream emphasizes GitHub's commitment to providing developers with access to the latest AI tools and encourages viewers to explore these features and provide feedback. The presenter also shares a personal project built with Copilot, showcasing its potential for automating tasks and improving developer productivity.