Rubber Duck Thursday!

Key Concepts

• GitHub Copilot Cloud Agent: A feature allowing users to run tasks, manage agents, and utilize various AI models directly on GitHub.com.
• Auto Mode: A dynamic feature in the GitHub Copilot CLI that automatically selects the most efficient AI model for a given task to optimize performance and manage rate limits.
• Agentic Engineering: The practice of using AI agents to automate development workflows, such as PR reviews, testing, and feature implementation.
• Plan Mode: A Copilot CLI feature that allows the AI to outline a step-by-step strategy before executing code changes.
• MCP (Model Context Protocol): A standard for connecting AI assistants to systems and data sources.
• Vibe Coding: A colloquial term for rapid, iterative development using AI agents to build tools quickly, often prioritizing speed and experimentation over rigid, spec-driven processes.

---

1. Recent GitHub Updates & Features

The stream highlighted several new releases at GitHub:

• Agent Session Management: Users can now view and manage agent sessions directly within GitHub Issues and Projects via a new "Running" tab and side panel.
• Dependabot Integration: Pull requests generated by Dependabot can now be assigned to GitHub Copilot for automated analysis and resolution.
• GitHub CLI (gh) Skills: Version 2.9.0 introduces the gh skill command, allowing developers to discover, install, and manage agent skills directly from repositories.
• Auto Model Selection: Now generally available in the Copilot CLI, this feature dynamically routes tasks to models (e.g., Claude Haiku, Sonnet, Opus, or GPT-4o) based on the specific requirements of the task and the user's plan.

2. Methodology: The "Auto Mode" Experiment

The host conducted a live test to evaluate the efficacy of Auto Mode in the Copilot CLI using a Next.js project called "Chapter Smith."

• The Process:
  1. Planning: The user invoked "Plan Mode" to generate a list of project improvements. The AI (Claude Haiku 4.5) successfully identified high-impact areas like UI/UX, testing, and performance.
  2. Implementation: The user requested the implementation of an "inline editing" feature for YouTube chapters.
  3. Observation: The AI remained on the Claude Haiku 4.5 model throughout the entire process. While it successfully created the necessary components and hooks, it struggled with scope creep (attempting to fix unrelated bugs in the export route) and ultimately failed to fully implement the save functionality.

3. Key Arguments & Perspectives

• Model Specialization: The host argues that developers should not rely on a single model for all tasks. For example, Opus is recommended for high-level reasoning and complex debugging, while Sonnet is preferred for specific, instruction-following tasks.
• The Role of Human Oversight: Despite the power of AI agents, the host emphasized that developers must remain "in the loop." The live demo served as a cautionary tale: when the AI attempted to modify files outside the requested scope, the human developer had to intervene to revert changes.
• Actionable Advice for Beginners: The host suggests that developers build a portfolio of projects that solve real-world problems. She recommends using GitHub Pages for free hosting and utilizing the official GitHub YouTube tutorials to learn CLI workflows.

4. Notable Quotes

• "I prove everything with math, not vibe." — A viewer comment highlighting the tension between "vibe coding" and rigorous, spec-driven engineering.
• "Speed isn't always the best... sometimes you have to slow down and keep your practices." — The host’s reflection on the importance of code reviews and testing, even when using AI to move fast.
• "The lesson here is Haiku for planning, not for implementing." — The host’s conclusion after the live demo revealed that the model chosen by Auto Mode was better suited for architectural suggestions than complex code generation.

5. Synthesis & Takeaways

The session demonstrated that while GitHub’s new agentic tools (specifically Auto Mode and CLI skills) significantly lower the barrier to entry for feature development, they are not yet a "set it and forget it" solution. The primary takeaway is that Auto Mode is a powerful utility for task routing and planning, but developers must maintain strict oversight of the AI's output. The host encouraged viewers to experiment with these tools, contribute to open-source projects, and continue building, while reminding them that "typos and bugs are part of the process."