GitHub Developer Advocate Demo: Microsoft Build 2025

Key Concepts

• Azure AI Foundry: A platform for building and managing AI agents.
• GitHub Copilot: An AI pair programmer that assists with coding tasks.
• Agent Mode: A new feature in GitHub Copilot that allows it to perform more complex tasks.
• GPT-4o: A large language model used by the VibeTravel AI agent.
• Grounding: Providing an AI agent with external knowledge to improve its accuracy.
• Actions: Capabilities that allow an AI agent to perform tasks on behalf of the user.
• Model Context Protocol (MCP): A protocol that allows tools like GitHub Copilot to access context and capabilities from other services.
• Hallucination: When an AI agent generates incorrect or nonsensical information.

Azure AI Foundry and VibeTravel

• Main Topic: Demonstrating how Azure AI Foundry can be used to improve AI agents and how GitHub Copilot can streamline development workflows.
• VibeTravel App: A chat-based application that uses an AI travel agent to help users plan trips.
• Initial Problem: The VibeTravel agent initially provided incorrect information about skiing in New Zealand, suggesting it was possible in January (summer season).
• Solution with Azure AI Foundry:
  • Accessing the AI agents within Foundry.
  • Identifying the VibeTravel agent, which uses GPT-4o.
  • Adding grounding to the agent by enabling Bing search.
  • Giving the agent access to a flight reservation API (OpenAPI).
• Result: The agent was able to provide accurate information about the skiing season in New Zealand and book flights.
• Quote: "Being able to add data sources or search capabilities makes the agent smarter and far less likely to hallucinate."
• Key Takeaway: Azure AI Foundry simplifies the process of improving AI agent accuracy by adding knowledge sources and enabling actions.

GitHub Copilot and Agent Mode

• Main Topic: Using GitHub Copilot, specifically Agent Mode, to address issues in the VibeTravel application.
• Agent Mode Capabilities:
  • Access to various LLMs from different providers.
  • Ability to go beyond simple code suggestions and perform actual development tasks.
• Model Context Protocol (MCP):
  • Enables GitHub Copilot to access information from other services, such as GitHub issues.
  • Example: Using GitHub MCP Server to fetch details about assigned issues.
• Issue 1: Reset Button Warning:
  • Problem: The reset button in the app clears the chat session without a warning.
  • Solution: Using GitHub Copilot to implement a warning message before clearing the chat.
  • Process:
    1. Providing Copilot with the issue details and a design sketch (screenshot).
    2. Granting Copilot permissions to use the GitHub MCP Server.
    3. Copilot analyzes the issue, identifies relevant files, and implements the changes.
    4. Copilot adheres to coding styles and standards.
• Issue 2: Copilot Assigned Issue:
  • Issue was pre-assigned to Copilot.
  • Copilot automatically opened a pull request with a description, implementation details, and code changes.
  • Full session log available for review.
  • Ability to provide feedback via comments in the pull request, which Copilot addresses.
• Issue 3: Satya's Issue (Group Size Filter):
  • Copilot implemented a group size filter on the Azure Dev Community website.
  • Included logic for Redis caching and testing instructions.
  • Deployed the app to a staging environment.
• Vision Capabilities: Agent Mode's new vision capabilities, means that Copilot can even understand the sketches of what I want.
• Key Takeaway: GitHub Copilot Agent Mode significantly accelerates development by automating tasks, understanding context, and adhering to coding standards.

Synthesis/Conclusion

The video demonstrates how Azure AI Foundry and GitHub Copilot can be used together to improve AI-powered applications and streamline the development process. Azure AI Foundry provides a platform for managing and enhancing AI agents by adding knowledge sources and enabling actions. GitHub Copilot, particularly with Agent Mode, automates development tasks, understands context, and adheres to coding standards, leading to increased efficiency and faster issue resolution. The combination of these tools empowers developers to build more intelligent and functional applications with greater ease.