Skip to content

Lab 011: Copilot Studio — First Agent

Level: L100 Path: Agent Builder — Teams Time: ~30 min 💰 Cost: Free Trial — Microsoft Copilot Studio free trial (no credit card for first 30 days)

What You'll Learn

  • Navigate the Copilot Studio canvas (no-code/low-code agent builder)
  • Create a Q&A agent from a knowledge source (FAQ document)
  • Test your agent in the built-in chat test panel
  • Publish the agent to Microsoft Teams
  • Understand topics, triggers, and fallback behavior

Introduction

Microsoft Copilot Studio is a graphical, low-code platform for building conversational AI agents without writing code. You define topics (conversation flows), connect knowledge sources, and publish to Teams, websites, or other channels in minutes.

This lab builds a customer service agent for the fictional OutdoorGear Inc. company, grounded in a product FAQ.

Prerequisites

  • Microsoft account (free at account.microsoft.com)
  • Copilot Studio trial: copilotstudio.microsoft.com → Start free trial
  • Microsoft Teams (free personal edition works)

No credit card needed

The Copilot Studio free trial lasts 30 days and does not require payment details.

Lab Exercise

Step 1: Create a new Copilot

  1. Go to copilotstudio.microsoft.com
  2. Sign in with your Microsoft account
  3. Click Create → New agent
  4. Fill in:
  5. Name: OutdoorGear Assistant
  6. Description: Customer service agent for OutdoorGear Inc. — answers product and policy questions
  7. Instructions: You are a friendly customer service agent for OutdoorGear Inc. Answer questions about products, return policies, shipping, and warranties. Be concise and helpful.
  8. Click Create

Step 2: Add a knowledge source

  1. In the left panel, click Knowledge
  2. Click Add knowledge → Public website or file
  3. Enter this URL (our sample FAQ): 
    https://raw.githubusercontent.com/lcarli/AI-LearningHub/main/data/knowledge-base.json
    Or click Upload file and paste this content into a .txt file first.

Using the knowledge-base.json

The data/knowledge-base.json file contains 42 documents including product guides, return policies, FAQs, and shipping info — all pre-formatted for RAG.

Step 3: Test the built-in knowledge

  1. Click Test in the top-right corner
  2. In the chat panel, try these questions:
  3. What is your return policy?
  4. Do you have waterproof boots?
  5. How long does shipping take?
  6. The agent should answer from the knowledge source and cite where it found the answer

Step 4: Create a custom topic

Custom topics let you override AI responses with deterministic flows for specific intents.

  1. Click Topics in the left panel
  2. Click Add a topic → From blank
  3. Name it: Order Status
  4. Under Trigger phrases, add:
  5. Where is my order
  6. Track my order
  7. Order status
  8. What happened to my order
  9. Add a Message node: 
    To check your order status, please visit our order portal at outdoorgear.com/orders or call 1-800-OUTDOOR. Have your order number ready!
  10. Add an End conversation node
  11. Click Save

Step 5: Test the custom topic

In the test panel, type: Where is my order?

The agent should use your custom topic flow, not the AI fallback. Notice how deterministic topics take priority over AI generative answers.

Step 6: Publish to Teams

  1. Click Publish in the left panel
  2. Click Publish to make the agent live
  3. Click Channels → Microsoft Teams
  4. Click Turn on Teams
  5. Click Open agent — this opens a deep link
  6. In Teams, click Add to install the agent as an app
  7. Start chatting with your OutdoorGear Assistant in Teams!

Copilot Studio Architecture

┌─────────────────────────────────────────┐
│          Copilot Studio                 │
│                                         │
│  ┌─────────────┐   ┌─────────────────┐  │
│  │   Topics    │   │  Generative AI  │  │
│  │ (no-code    │   │  (knowledge +   │  │
│  │  flows)     │   │   LLM fallback) │  │
│  └──────┬──────┘   └────────┬────────┘  │
│         │   Topic match?    │           │
│         │ ◄─────────────────┘           │
│         ▼                               │
│     User message                        │
└─────────────────────────────────────────┘
         │
         ▼
  Channels: Teams, Web, Slack, ...

Priority order: 1. Custom topics (exact trigger match) → deterministic 2. Built-in system topics (escalate, fallback) 3. Generative AI answers from knowledge sources

When to use Copilot Studio vs Pro Code

Copilot Studio Pro Code (SK/MCP)
Business users, no code Developers
Fast prototyping Complex logic
Teams/SharePoint integration Custom integrations
GUI-based flows Programmatic control
Limited customization Full flexibility

Next Steps