​​AI Summary

Generated by AI. Be sure to check for accuracy.

Meeting notes:

  • Overview of Microsoft Copilot and Copilot Studio Lite: Sara and Cory introduced the Master Series session, providing an overview of Microsoft Copilot, its foundational concepts, and the distinction between Copilot Studio Lite and Full, with Cory explaining Copilot’s role in generative AI, data security, and its applicability across various business functions.
    • Session Introduction and Objectives: Sara welcomed attendees, explained the Master Series format, and outlined the session’s focus on Copilot Studio Lite, including upcoming related events and the opportunity for interactive Q&A.
    • Copilot Fundamentals and Data Security: Cory described how Copilot is built on Microsoft 365 foundations, inheriting organizational security and compliance policies, ensuring user data and prompts are protected and not used for model training.
    • Work IQ and Generative AI Vision: Cory detailed the Work IQ concept, which combines data, memory, and inference to improve Copilot’s accuracy and contextual understanding, leveraging Microsoft Graph and organizational data.
    • Copilot Use Cases Across Business Functions: Cory highlighted Copilot’s applicability for knowledge workers and frontline employees in areas such as sales, marketing, operations, IT, HR, and manufacturing, emphasizing its role in streamlining workflows and improving productivity.
    • Copilot Chat vs. Licensed Experience: Cory explained the difference between Copilot Chat (web-grounded, available to all users) and the licensed Copilot experience, which provides enhanced capabilities, priority access, and integration with work data, noting that Copilot Studio Full offers more advanced agent-building options.
  • Prompting Strategies and Effective Agent Instructions: Cory provided guidance on effective prompting for Copilot and agent instructions, emphasizing the importance of clear goals, context, sources, and expectations, and demonstrated how to structure instructions for agents using markdown and workflow steps.
    • Prompting Best Practices: Cory outlined four key elements for effective AI prompting: goal, context, source, and expectations, illustrating with examples how to frame prompts to achieve desired responses from Copilot.
    • Structuring Agent Instructions: Cory explained the importance of actionable language, defining nonstandard terms, building step-by-step workflows, and using markdown formatting in agent instructions to ensure clarity and precision.
    • Error Handling and Output Examples: Cory recommended including error handling procedures and output examples in agent instructions to guide agents in managing missing data or conflicting information, and to provide users with clear expectations.
  • Overview and Use of Out-of-the-Box Copilot Agents: Cory described the available out-of-the-box Copilot agents such as Researcher and Analyst, detailing their advanced reasoning capabilities, use cases, and licensing requirements, with Sara and Paul addressing attendee questions about agent availability and prioritization of data sources.
    • Researcher Agent Capabilities: Cory explained that the Researcher agent uses chain-of-thought processing to provide verifiable information, supports custom data sources, and can handle complex queries such as generating tariff briefs and tracking recent updates.
    • Analyst Agent Capabilities: Cory described the Analyst agent as a data scientist tool for analyzing large Excel files, generating insights, creating visuals, and handling complex data analysis tasks.
    • Licensing Requirements for Agents: Sara relayed an attendee question about Analyst agent availability, and Cory clarified that access to Analyst and Researcher agents requires a full M365 Copilot license, with organizational policies potentially restricting access.
    • Prioritizing Data Sources and Error Handling: Sara and Paul discussed best practices for handling multiple data sources in agent instructions, recommending prioritization of sources and explicit guidance to manage conflicting information and hallucinations.
  • Building a Custom Agent in Copilot Studio Lite: Cory demonstrated the step-by-step process of building a custom tariff analyst agent in Copilot Studio Lite, covering agent configuration, instruction writing, knowledge source selection, capabilities, prompt setup, testing, and sharing options.
    • Agent Configuration and Instructions: Cory showed how to configure a new agent, including naming, description, and writing detailed instructions with objectives, workflow steps, error handling, glossary, and security compliance requirements.
    • Selecting Knowledge Sources: Cory explained how to add up to four public websites as data sources, with options to include SharePoint, Teams, and email for licensed users, and recommended restricting agents to specified sources for accuracy.
    • Capabilities and Suggested Prompts: Cory enabled document creation capabilities for the agent, set up suggested starter prompts, and described how these facilitate common use cases and rapid user adoption.
    • Testing and Output Validation: Cory tested the agent with sample HS codes and cost impact scenarios, validated output formats, and demonstrated the agent’s ability to generate structured responses and create PowerPoint presentations with cost breakdowns.
    • Agent Sharing and Deployment: Cory explained sharing options, noting that agents are private by default but can be shared with specific users or organization-wide depending on policy, and described how users can upload content for analysis.
  • Licensing, Deployment, and Decision Guidance for Copilot Studio Lite vs. Full: Sara and Cory addressed licensing nuances, deployment options, and provided a decision tree for choosing between Copilot Studio Lite and Full, clarifying use cases, organizational scope, and future extensibility.
    • Licensing Requirements and Pay-As-You-Go: Cory clarified that users without a Copilot license can only build agents using web sources, while licensed users can access work data and build more advanced agents, with pay-as-you-go licensing offering additional flexibility.
    • Deployment Scope and Sharing: Cory described how agents can be shared privately, with select users, or organization-wide, and noted that organizational policies govern sharing and store visibility.
    • Decision Tree for Lite vs. Full: Cory presented a decision tree to help users determine whether to use Copilot Studio Lite or Full based on agent scope, complexity, workflow requirements, and governance needs, recommending starting with Lite for personal productivity and evolving to Full as needs grow.
  • Upcoming Events and Session Resources: Sara concluded the session by promoting upcoming Master Series events focused on AI in Microsoft support and Copilot Studio Full, and provided information on accessing session resources and the registration portal.
    • Event Announcements: Sara highlighted two upcoming events: one on December 10th featuring Microsoft support’s use of AI, and another on January 15th focused on Copilot Studio Full, encouraging attendees to register via the provided portal.
    • Accessing Session Resources: Sara informed attendees about the portal for accessing previous and future session materials, and reiterated the opportunity to request follow-up sessions or reach out for more information. 

Download