Design and Build Integrated AI Agent Solutions in Copilot Studio (AB-620)

Course 8779 Advantage Plan Course

  • Duration: 3 days
  • Exam Voucher: Yes
  • Language: English
  • Level: Intermediate

In this hands-on, instructor-led course, you'll learn to design, build, integrate, and manage production-ready AI agents in Microsoft Copilot Studio using the GitHub Copilot harness. You'll create instruction-driven agents, automate business processes with workflows and human approvals, connect agents to enterprise knowledge, tools, Model Context Protocol (MCP) servers, and Microsoft IQ, and build multi-agent solutions with connected agents. You'll also learn how to take agents to production with governance, responsible AI, evaluation, monitoring, and application lifecycle management (ALM). Attend in person or online with Learning Tree AnyWare.

Copilot Studio AI Agent Training Delivery Methods

  • In-Person

  • Online

  • Upskill your whole team by bringing Private Team Training to your facility.

Copilot Studio AI Agent Training Information

  • What You'll Learn

    • Compare Copilot Studio agent harnesses and plan agents with usage and cost in mind.
    • Build instruction-driven agents and package reusable behavior as skills.
    • Automate business processes with workflows, including human-in-the-loop approvals.
    • Integrate agents with enterprise knowledge, connectors, MCP servers, and Microsoft IQ.
    • Design multi-agent solutions that delegate work to connected agents.
    • Govern, evaluate, monitor, and deploy agents to production with Power Platform ALM.

    Prerequisites

    • Hands-on experience creating, testing, and publishing agents in Microsoft Copilot Studio, including instructions, knowledge sources, and tools
    • Intermediate understanding of generative AI concepts, including how AI models produce responses
    • Proficiency working with REST APIs, plus familiarity with identity and authentication concepts such as Microsoft Entra ID
    • Familiarity with Power Platform environments, connectors, and Microsoft Dataverse
    • Basic familiarity with Power Automate cloud flows and Markdown
    • Helpful: familiarity with Application Insights and Kusto Query Language (KQL)

Copilot Studio AI Agent Training Outline

1. Explore Copilot Studio and Agent Harnesses

  • Understand the role of a harness, the runtime architecture that powers an agent
  • Compare the GitHub Copilot, standard, and Copilot Chat harnesses
  • Explore the components and lifecycle of a GitHub Copilot harness agent
  • Select a harness based on business and technical requirements
  • Hands-on: Select a harness for a business scenario

 2. Manage Usage and Cost for Agents

  • Compare billing models across agent harnesses
  • Identify the runtime and build-time decisions that drive Copilot Credit consumption
  • Understand how administrators govern and monitor Copilot Credit usage

3. Create Instruction-Driven Agents

  • Define an agent's purpose, scope, behavior, tone, and boundaries
  • Write effective agent instructions
  • Add file and website knowledge sources
  • Configure the agent's greeting and suggested prompts
  • Test and refine an agent in Preview
  • Hands-on: Create a benefits information agent

4. Design Reusable Behavior with Skills

  • Decide when to use a skill instead of other agent components
  • Author skills from prebuilt templates or from scratch
  • Compose and package a skill for reuse
  • Reference skills from agent instructions and plan reuse across agents and environments

5. Automate Agent Tasks with Workflows

  • Decide when to use a deterministic workflow instead of agent reasoning
  • Design workflow inputs and outputs for reliable agent communication
  • Build and publish a standalone workflow with connector actions
  • Add a workflow to an agent as a tool with end-user confirmation
  • Test and troubleshoot workflow success and failure paths
  • Hands-on: Create a task notification agent and workflow

 6. Add Human-in-the-Loop Steps to Workflows

  • Identify workflow steps that need human review based on risk
  • Pause a workflow to collect missing information from a designated person
  • Branch a workflow on a human decision and require approval before sensitive actions

 7. Design Integration Strategies for Agents

  • Compare knowledge sources, tools, and connected agents for different integration needs
  • Recommend an integration approach for an enterprise requirement
  • Identify the credentials, connections, and policies each approach requires
  • Write tool, skill, and connected-agent descriptions that help the agent choose correctly

 8. Ground Agents in Enterprise Knowledge

  • Compare the knowledge sources available to an agent
  • Add a Copilot connector as a knowledge source
  • Test grounded answers and diagnose source-selection, content, and access issues
  • Plan evaluations for grounded answers

 9. Extend Agents with Tools

  • Understand how enhanced orchestration selects tools
  • Add and configure a connector tool, including its connection, actions, and inputs
  • Choose between a connector, an MCP server, and a workflow for an action
  • Test tool use and diagnose issues in the activity trace

 10. Connect Agents to MCP Servers

  • Decide when to use a Model Context Protocol (MCP) server instead of a connector or workflow
  • Add an MCP server and configure authentication
  • Choose which MCP tools an agent can use
  • Test and troubleshoot MCP tool calls

 11. Connect Agents to Microsoft IQ

  • Choose between Foundry IQ, Fabric IQ, and Work IQ based on where your data lives
  • Connect an agent to a Foundry IQ knowledge base
  • Ground an agent in business data from a Microsoft Fabric semantic model with Fabric IQ
  • Ground an agent in Microsoft 365 work context with Work IQ

 12. Design Multi-Agent Solutions

  • Decide when a solution needs more than one agent
  • Choose how agents coordinate
  • Compare implementation options on the standard and GitHub Copilot harnesses

 13. Build Connected Agent Solutions

  • Prepare an agent so other agents can connect to it
  • Add a connected agent with a clear routing description
  • Test delegation and troubleshoot misrouted requests
  • Call a published agent from a workflow

 14. Design a Governance and Security Strategy

  • Understand governance controls at the tenant, environment, and agent levels
  • Apply a zoned governance strategy based on purpose and risk
  • Select an authentication approach based on audience, channels, and harness
  • Evaluate channel and deployment readiness
  • Monitor and govern agents with Agent 365 and agent inventory

 15. Design a Responsible AI Strategy

  • Apply Microsoft's responsible AI principles to agent design
  • Configure content moderation
  • Evaluate models for safety and production readiness
  • Design guardrails and transparency practices

 16. Evaluate Agents

  • Create test sets that represent real scenarios
  • Configure and run evaluations on the Evaluate tab
  • Interpret results and triage failed test cases
  • Validate an AI model migration against acceptance gates

17. Monitor and Improve Agents

  • Review production health on the Monitor tab
  • Diagnose issues from session evidence and improve agents from transcripts
  • Extend monitoring with Application Insights
  • Communicate agent value with telemetry

 18. Implement Application Lifecycle Management for Agents

  • Plan Power Platform environments for development, test, and production
  • Package an agent and its dependencies in a solution
  • Configure environment variables and connection references
  • Deploy agents with Power Platform pipelines and service principals

Need Help Finding The Right Training Solution?

Our training advisors are here for you.

Copilot Studio AI Agent Training FAQs

This course is designed for advanced app makers, professional developers, Power Platform developers, functional consultants, and agent builders who want to design, extend, and integrate Copilot Studio agents for enterprise use.

A harness is the runtime architecture that powers an agent in Copilot Studio. It shapes how the agent orchestrates its work, which capabilities it can use, and how it's authored, published, and billed. This course focuses on the GitHub Copilot harness, where agents are defined primarily through natural-language instructions, and compares it with the standard and Copilot Chat harnesses.

You should be comfortable working with REST APIs and identity and authentication concepts. Familiarity with Power Platform, Power Automate cloud flows, and Markdown is also helpful.

This is a 3-day, hands-on, instructor-led course. You can attend in person or online with Learning Tree AnyWare, or bring the course to your organization with Private Team Training.

The course focuses on Microsoft Copilot Studio and also covers Copilot Studio workflows, Power Platform connectors, Copilot connectors, Model Context Protocol (MCP), Microsoft IQ (Foundry IQ, Fabric IQ, and Work IQ), Agent 365, Application Insights, and Power Platform pipelines.

Yes. You'll learn when a solution needs more than one agent, how to choose a coordination pattern, and how to build connected agent solutions that delegate work to specialist agents, including calling published agents from workflows.

Yes. This course aligns with the Microsoft Certified: AI Agent Builder Associate certification and helps you prepare for Exam AB-620: Designing and Building Integrated AI Agent Solutions in Copilot Studio. A Microsoft exam voucher is included with your attendance.