Build AI Agents to Automate Your Workflows
Published by O'Reilly Media, Inc.
Using MS Copilot Studio to create AI agents
What you’ll learn and how you can apply it
- Plan a Copilot Studio agent from persona, job-to-be-done, instructions, topic map, guardrails, and success metrics
- Build a grounded AZ-900 study assistant using knowledge sources, topics, generative answers, and test prompts
- Extend the agent with actions, agent flows, MCP and subagent concepts, and human handoff patterns
- Evaluate, publish, observe, and govern the agent using analytics, agent evaluations, Power Platform Well-Architected guidance, and security principles
Course description
This four-hour live course is for cloud beginners preparing for Microsoft Azure Fundamentals certification. Tim Warner teaches the current Copilot Studio agent model through a single, progressive build instead of a grab bag of disconnected demonstrations.
Your agent starts with a planned scenario with clear instructions, topics, guardrails, and success metrics. Then you’ll develop it into a grounded knowledge agent, a tool-using agent, and finally a governed agent that can be tested, observed, and prepared for real users. You’ll work with current Copilot Studio concepts including generative orchestration, knowledge sources, topics, actions and tools, agent flows, MCP and connected-agent patterns, model selection, evaluations, analytics, VS Code/YAML awareness, and Power Platform Well-Architected guidance.
This live event is for you because...
- You’re a Microsoft 365 or Power Platform maker, technical trainer, business analyst, IT professional, or solution architect who needs a current mental model for Copilot Studio agents.
- You support training, enablement, help desk, cloud adoption, internal knowledge work, or governed automation where accurate answers and clear handoff paths matter.
- You want to understand current Copilot Studio capabilities without chasing every feature rename in the portal.
Prerequisites
- A Microsoft 365 work or trial account with Copilot Studio access
- Microsoft Teams available in the browser or desktop app
- A browser, plus optional VS Code if you want to inspect exported Copilot Studio YAML files
- Comfort using Microsoft 365 and web-based admin or authoring tools
- Basic familiarity with Microsoft Azure concepts is helpful but not required
- No coding required (some Power Automate or workflow design experience is helpful)
- No prior Copilot Studio experience needed
Recommended preparation:
- Access the course repository materials before the event (link to come)
Recommended follow-up:
- Use the course repo to review the AZ-900 assistant, extend the topic map, inspect exported Copilot Studio YAML, and compare current Microsoft Learn feature names with what appears in your tenant
Schedule
The time frames are only estimates and may vary according to how the class is progressing.
Inception: Designing the AZ-900 agent (60 minutes)
- Presentation: The current Copilot Studio agent model (instructions, knowledge, topics, tools, triggers, channels, and orchestration); the course arc; plan the AZ-900 topic map and knowledge strategy
- Demonstration: Creating the agent shell, drafting instructions, enabling generative orchestration, and sketching the topic map; defining guardrails for unsupported claims and human mentor handoff
- Hands-on exercise: Write one instruction, one success metric, and one topic for the AZ-900 assistant
- Q&A
- Break
Build: Topics, triggers, knowledge, and grounded answers (60 minutes)
- Presentation: When to use agent-level knowledge versus topic-level grounding; exam domain overview, practice question, study plan, fallback, and search topics
- Demonstration: Testing grounded responses, improving fallback behavior, and tuning topic inputs; using AutomaticTaskInput and explicit questions where each makes sense; knowledge limits, source ordering, and citation expectations
- Hands-on exercise: Write a topic that teaches, checks understanding, and hands off cleanly
- Q&A
- Break
Extend: Actions, MCP, and subagent patterns (60 minutes)
- Presentation: When the assistant should answer, ask a follow-up question, call a tool, or hand off to a human mentor; how actions, agent flows, HTTP tools, MCP, and add-other-agent patterns fit into the current Copilot Studio model; current multi-agent taxonomy; DLP, authentication, tenant approvals, external models, and least-privilege tool use
- Demonstration: Extending the AZ-900 assistant with a study-plan or lab-generator action pattern
- Hands-on exercise: Choose one tool the agent should call and one situation it must escalate
- Q&A
- Break
Operate: Testing, observing, governing, and publishing (50 minutes)
- Presentation: Preparing the agent for publishing to supported channels such as Teams, SharePoint, and Microsoft 365 Copilot; using native agent evaluations, test sets, activity maps, and version comparison to find failure patterns; analytics, conversation transcripts, containment signals, and user experience metrics; applying the Power Platform Well-Architected lens; environment strategy, DLP, data boundaries, tenant approvals, and rollout planning; the extension path from AZ-900 study assistant to other internal knowledge and enablement agents
- Hands-on exercise: Name the first agent you would safely pilot and the governance question it must answer first
Wrap-up and Q&A (10 minutes)
Your Instructor
Tim Warner
Tim Warner has been a Microsoft MVP in Azure AI and Cloud/Datacenter Management for 6 years and a Microsoft Certified Trainer for more than 25 years. His O'Reilly Live Training classes on generative AI, GitHub, DevOps, data engineering, cloud computing, and Microsoft certification reach hundreds of thousands of students around the world. He's written for Microsoft Press, presented at Microsoft Ignite, and contributed to several Microsoft open-source projects. You can connect with Tim on LinkedIn: timw.info/li.