​​AI Summary

​

Generated by AI. Be sure to check for accuracy.

Meeting notes:

  • Introduction to Copilot Studio Full and AI Agent Building: Sara and Kevin introduced the session's focus on Copilot Studio Full, outlining the agenda, the interactive format, and the expertise of the presenters, with Sara emphasizing the session's goal to help participants learn about building AI agents using Copilot Studio.
    • Session Overview and Polls: Sara launched interactive polls to gauge participants' experience with AI and Copilot agents, aiming to tailor the session content to the audience's familiarity and needs.
    • Presenter Introductions: Sara introduced herself as a Principal Technical Specialist at Microsoft and Kevin as a Principal Solution Engineer, highlighting their roles and experience in Dynamics applications and Copilot Studio.
    • Session Structure and Engagement: Sara explained the session's structure, including demos, Q&A, and interactive elements, and encouraged participants to engage through polls and questions to maximize learning.
  • Copilot Studio Full Capabilities and Customer Success Stories: Kevin provided an overview of Copilot Studio Full, describing its evolution, integration capabilities, and governance features, and shared real-world customer success stories from Dow Chemical, Cineplex, and the University of Hong Kong to illustrate its impact.
    • AI Evolution and Copilot Studio Positioning: Kevin traced the evolution of AI from data scientist-led efforts to the democratization enabled by Copilot Studio, emphasizing its ability to connect with legacy and modern systems and support both simple and sophisticated agents.
    • Governance and Security: Kevin highlighted the importance of governance in Copilot Studio, noting Microsoft's longstanding investment in secure, scalable low-code solutions and the platform's ability to manage agent sprawl, compliance, and risk.
    • Customer Success Stories: Kevin shared examples of Copilot Studio's impact: Dow Chemical's automation of invoice processing, Cineplex's reduction in refund handling time, and the University of Hong Kong's use of AI agents to support students and faculty.
  • Demonstration of Agent Building and Advanced Features: Kevin demonstrated the process of building and enhancing AI agents in Copilot Studio, covering the transition from simple to advanced agents, the use of orchestrators, adaptive cards, and integration with various data sources and tools.
    • Agent Creation Workflow: Kevin walked through creating a study buddy agent, starting with uploading manuals and providing instructions, then converting it from a light agent to a full Copilot Studio agent for expanded capabilities.
    • Adaptive Cards and UI Enhancement: Kevin explained how adaptive cards can be used to structure user interactions, such as course catalogs and quizzes, and described how M365 Copilot can assist in generating the necessary code for these cards.
    • Orchestrator and Multi-Agent Capabilities: Kevin detailed the orchestrator's role in routing intents, supporting multi-agent interactions, and enabling both generative and traditional automation within agents.
    • Integration with Data Sources and Tools: Kevin described how agents can connect to various data sources like Dataverse, SharePoint, and external systems using connectors, agent flows, and the Model Context Protocol (MCP), allowing for advanced automation and knowledge retrieval.
    • Testing, Evaluation, and Feedback: Kevin demonstrated the built-in testing and evaluation features in Copilot Studio, including test sessions, evaluation sets, content moderation, and user feedback mechanisms to ensure agent quality before deployment.
  • Q&A: Technical Deep Dives and Troubleshooting: Sara and Kevin addressed participant questions on topics such as developing adaptive cards, integrating Copilot Studio with Azure AI Foundry, using sandbox environments, agent accuracy, voice capabilities, data storage, and SharePoint integration.
    • Developing Adaptive Cards: Kevin advised using M365 Copilot chat for guidance on adaptive cards, demonstrated where adaptive cards can be embedded in agent tools or topics, and explained how to bind data to card properties for interactive experiences.
    • Copilot Studio and Azure AI Foundry Integration: Kevin explained that Copilot Studio can integrate with Azure AI Foundry by selecting Foundry as a model or tool within prompts, allowing for the deployment of pro-code agents under a unified governance framework.
    • Sandbox and Developer Environments: Kevin recommended using developer environments in Power Platform for experimentation, described how IT can set up these environments, and outlined the roles and permissions needed to build agents safely without accessing enterprise data.
    • Agent Accuracy and Knowledge Grounding: Kevin discussed differences in agent accuracy between M365 Copilot and Copilot Studio, attributing variations to the underlying knowledge grounding (Microsoft Graph vs. Dataverse RAG) and suggested methods to improve results.
    • Voice Channel and Escalation Features: Kevin described how Copilot Studio supports voice channels, including Direct Line Speech and escalation to live representatives, and how responses can be optimized for voice interactions.
    • Storing Agent Responses in SharePoint: Kevin outlined how to extract structured data from agent prompts and use connectors or agent flows to store responses automatically in SharePoint or Dataverse.
    • Troubleshooting SharePoint Integration: Kevin and Attendee63 discussed troubleshooting steps for SharePoint integration, including checking agent authentication, IRM policies, and permissions, and suggested testing with basic connections or folder links.