In this session, we build the foundation for working with Microsoft Power Platform and Microsoft Copilot Studio.
Topics Covered:
Power Platform fundamentals
Introduction to Microsoft Copilot Studio
Understanding environments
Creating your first agent
Testing and publishing agents
Introduction to topics and Generative AI
Hands-On Activity:
Create, configure, test, and publish your first Copilot Studio agent.
In this session, we explore how to build intelligent and interactive conversations in Microsoft Copilot Studio using topics, entities, variables, and conditional logic.
Module 2 – Manage Topics
Create and organize topics
Configure trigger phrases
Design conversation flows
Build branching conversations
Configure fallback behavior
Create reusable conversation experiences
Module 3 – Work with Entities & Variables
Understand built-in and custom entities
Capture structured user information
Create and manage variables
Validate user input
Reuse information throughout conversations
Pass information between topics
Hands-On Activity:
Build an intelligent conversation that captures user information, stores it in variables, and uses conditional branching.
In Day 3, we explore how Microsoft Copilot Studio agents can move beyond conversations and take real business actions.
In this session, you will learn how to:
Connect Copilot Studio with Power Automate
Understand agent actions, inputs, and outputs
Create interactive Adaptive Cards
Collect information using Adaptive Cards
Trigger a Power Automate flow from an agent
Send automated email notifications
Build an end-to-end agent experience
Hands-On Lab: Build the UFA Mentorship Agent to collect mentor application information and automate the submission process.
In this session, we learned how to connect Copilot Studio agents to real business processes using Tools and Agent Flows.
Topics covered:
Tools and Agent Flows
Performing business actions from agent conversations
Connecting to external systems
HTTP requests
Passing data between systems
Processing JSON responses
Error handling and troubleshooting
Hands-On Lab:
Connected an agent conversation to an external business action and implemented error handling for a more reliable workflow.
Watch the class recording below and review the provided resources to practice the concepts covered in Day 4.
In this class, we explore how to build more reliable AI-powered responses in Microsoft Copilot Studio using Generative Answers and enterprise knowledge.
Generative Answers
Enterprise Knowledge
Knowledge Sources
Agent Instructions
Prompts and Grounding
Techniques for Reducing AI Hallucinations
Build a grounded knowledge agent using enterprise information and configure the agent to generate relevant, reliable, and context-aware responses.
By the end of this session, you will understand how knowledge sources, instructions, and grounding work together to improve the quality of an AI agent's responses.
In this class, we explored how Tools extend Microsoft Copilot Studio agents beyond conversation, allowing agents to perform real business operations.
Topics Covered:
Extend agents with Tools
Configure and use Custom Prompts
Create and connect Agent Flows
Configure Tool inputs and outputs
Control how an agent calls a Tool
Test Tool execution
Troubleshoot common Tool and Flow issues
Secure Tools and business actions
Understand when to request user confirmation before executing an action
Hands-On Lab:
Configure and test Tools that allow a Copilot Studio agent to perform business operations using Agent Flows.
In this session, we explore how to make a Copilot Studio agent autonomous so it can respond to events and take actions automatically.
We cover autonomous agents, event triggers, agent instructions, generative orchestration, knowledge grounding, tool selection, monitoring, governance, and safe autonomous operations.
Hands-On Lab: Build the UFA Internship Auto Responder, an agent that works in two ways:
Answers internship questions through chat using approved knowledge.
Detects incoming internship emails, understands the inquiry, searches approved knowledge, and automatically sends a grounded response when appropriate.
Topics Covered: Autonomous Agents • Autonomous Triggers • Agent Instructions • Generative Orchestration • Knowledge Grounding • Tool Selection • Automatic Email Responses • Human Escalation • Safe Autonomous Operations
In this session, we explore how to design and build multi-agent solutions in Microsoft Copilot Studio. We begin by completing the previous autonomous agent exercise, including adding an Outlook email tool, testing automated email responses, adding suggested prompts, and publishing the agent to different channels.
We then move into multi-agent architecture, learning how a Parent/Lead Agent can coordinate specialized agents and delegate specific responsibilities.
Topics covered: Multi-agent architecture, specialized agents, Parent/Lead and Child Agents, Connected Agent vs. Child Agent and when to use each, delegation and orchestration, data exchange between agents, security boundaries and least privilege, and building a Purchase Request multi-agent solution.
Hands-On: Create a Purchase Request Parent Agent and a Purchase Policy Child Agent, configure their descriptions and instructions, and test how the Parent Agent delegates work to the specialist agent.
In this session, we continue our Copilot Studio course with a deep dive into multi-agent architecture. We recap the key concepts of connected agents vs. child agents, data exchange between agents, and security boundaries - then move into a live, hands-on lab where we build a complete multi-agent solution from scratch.
What's covered:
Recap: Suggested prompts, tools in instructions, connected vs. child agents, data exchange principles, and least-access security boundaries
Real-world analogy: How a hotel concierge and restaurant kitchen illustrate parent-child agent coordination, input/output management, and guardrails
The UFA AI Command Center vision: ~45 agents working together as a unified "superapp" experience
Live lab - Purchase Request Agent: Build a parent agent that collects employee purchase requests and delegates to three child agents:
Purchase Policy Agent: checks compliance with company purchasing rules
Budget Check Agent: verifies department budget availability
Purchase Approval Agent: determines the approval routing based on cost
Testing & debugging: Running test cases, reading the Activity Map, troubleshooting repeated agent calls, and refining parent-agent instructions
Copilot Studio features tour: Overview of the Activity tab, Evaluation tab (auto-generated test cases, pass/fail scoring), and Monitor tab for production analytics
In this session, you'll learn how to build multi-agent solutions using connected agents - independent, specialized agents that can be invoked by a parent agent when user requests match their area of expertise.
What you'll learn:
Connected Agents vs. Child Agents - What makes a connected agent different, why you'd connect independent specialists instead of creating child agents, and the key benefits of reusability and separation of concerns
Agent Discovery - How to name and describe agents so the orchestrator knows which one to call, and how to avoid overlapping scope that causes routing conflicts
Defining the Contract - Structuring required inputs, predictable outputs, and human-friendly error responses between agents
Microsoft AI Services - Creating connected agents using Copilot Studio, M365 Agent SDK, Foundry, Fabric, Azure, and Power Platform, and when to use each
Delegation Patterns - How the primary agent interprets user intent, selects the right specialist, passes only the needed context, and combines results into a single response
Testing & Troubleshooting - Verifying expected and unexpected routing, handling no-delegation scenarios, incomplete results, and repeated calls
Hands-On Lab:
Watch a live demonstration in Microsoft Copilot Studio where we build a Travel Assistant agent with a Weather connected agent and a Travel Advisory child agent.
The lab covers:
Creating and publishing connected agents
Writing meaningful descriptions and instructions
Configuring orchestration and security settings
Troubleshooting authentication mismatches
Adding guardrails to keep responses on-domain
Testing via the Copilot Studio test pane and demo website
Preparing to integrate the OpenWeather REST API for real-time data