GitHub Copilot: Your AI Companion for Every Workflow

Key Concepts

• GitHub Copilot: An AI-powered coding assistant that integrates across IDEs, CLI, and web platforms.
• Agentic Workflow: A development paradigm where AI agents perform tasks, plan, and execute code with human oversight.
• TUI (Terminal User Interface): A text-based interface for the Copilot CLI that allows for interactive AI-assisted development.
• MCP (Model Context Protocol): A standard for connecting AI assistants to external systems (e.g., Slack, Jira, GitHub, .NET Aspire).
• Skills: Specialized instructions or code snippets that extend Copilot’s capabilities for specific project tasks.
• Autopilot Mode: A feature where the AI autonomously executes a sequence of tasks until completion.
• Plan Mode: A mode where the AI generates a step-by-step strategy for a task before executing it, requiring user approval.

---

1. Evolution of AI Development

The speaker outlines the progression of AI in software development:

• Wave 1 (2021): Smart autocomplete and "ghost text" (predicting code completion based on context).
• Wave 2 (Current): Agentic workflows where humans act as supervisors, reviewing and refining AI-generated code.
• Wave 3 (Future): Hybrid teams consisting of humans and autonomous agents working collaboratively to build applications.

2. Ecosystem and Integration

GitHub Copilot is designed to be ubiquitous, functioning as a single subscription across multiple environments:

• First-Party: GitHub.com and GitHub Mobile.
• IDEs: VS Code, Visual Studio, JetBrains, Eclipse, and Xcode.
• Productivity Tools: Slack, Teams, Jira, Linear, and Raycast.
• Terminal: The Copilot CLI provides a TUI that supports multi-file context awareness and diff visualization.

3. Practical Demonstration: Agentic Workflow

The presenter demonstrated building a blog feature using the CLI and VS Code:

• Skill Invocation: Used the awesome-copilot repository to suggest project-specific skills (e.g., Chrome DevTools for an Astro-based site).
• Plan Mode: Used to define navigation logic. The user provided a spec file, and Copilot generated a plan before execution.
• Autopilot: Used to implement light/dark mode. The agent autonomously read the specs and applied the changes.
• Delegation: The user "delegated" tasks to the cloud-based coding agent, which created a new branch and implemented the requested features.
• Browser Integration: Within VS Code, the agent was granted access to a live browser view, allowing the user to point at UI elements (e.g., "remove this header") and have the agent modify the code accordingly.

4. Technical Configuration

• Model Selection: Users can toggle between models like GPT-53 Codex, Opus 47, or GPT-54. Local execution is supported via Ollama or Foundry.
• Approval Settings: Users can choose between "Default" (prompting for destructive actions like terminal commands) and "Bypass" (automatic approval of all tool calls).
• MCP Servers: These allow the agent to query external data, such as checking open pull requests or interacting with project management tools.

5. Notable Quotes

• "I didn't think I would love the terminal until I started using it and now honestly I can't get away from it." — Regarding the Copilot CLI.
• "Nothing makes your family happier than when you're starting unloading GitHub Copilot agents on a mobile session." — Highlighting the ability to manage code on the go.

6. Synthesis and Takeaways

GitHub Copilot has transitioned from a simple autocomplete tool to a sophisticated agentic platform. The core value proposition is the unified workflow: a developer can start a task in the CLI, continue it in VS Code, and monitor progress via GitHub Mobile, all while maintaining context across different environments. By leveraging MCP servers and custom skills, developers can tailor the AI to their specific tech stack, significantly reducing the manual burden of repetitive coding tasks.

Resources for further learning: gh.io/all-the-stuff contains the workshop materials and session recordings from the global Dev Days events.