Skip to content

Latest commit

 

History

History
114 lines (100 loc) · 5.25 KB

index.md

File metadata and controls

114 lines (100 loc) · 5.25 KB

Lab 6: Create a Custom Agent Using Natural Language

Lab Title

Create a Custom Agent with Copilot

Lab Objectives

By the end of this lab, you will be able to:

  1. Understand what custom agents are vs. pre-built agents
  2. Use natural language prompts to define agent behavior
  3. Connect multiple knowledge sources (websites, SharePoint, documents)
  4. Ground your agent with enterprise data
  5. Test responses with citations from knowledge sources

Prerequisites

  • Microsoft 365 account with Copilot Studio access
  • Copilot Studio environment with maker permissions
  • SharePoint site with sales documentation (from Course Setup)
  • Sample sales FAQ document or knowledge file
  • OneDrive or SharePoint for document storage

Step-by-Step Guide

Step 1: Understanding Custom Agents (~5 minutes)

  1. Review what custom agents are:
    • You define the purpose and behavior
    • You ground with your own data and knowledge sources
    • Full control over personality, instructions, and actions
  2. Understand the difference from pre-built agents:
    • Pre-built: Templates with predefined structure
    • Custom: Built from scratch with your specific requirements
  3. Explore use cases for custom agents (IT support, sales, HR, etc.)

Step 2: Create Your Sales Agent (~10 minutes)

  1. Navigate to copilotstudio.microsoft.com

  2. Select Agent in the top-left corner

  3. Select Create blank agent

  4. Enter the following agent details by clicking on the Edit button and click on the save button after entering each field:

    • Name: Sales Agent
    • Description: This agent helps summarize client meetings and prepare next steps in the sales process. alt text

  5. In the Instructions section, click on the edit button and enter:

    • Instructions: "You are an agent who helps the user with all parts of the sales process. Creating meeting recaps, pricing, and making offers for potential clients."

    Click on the Save button after entering the instructions.

Step 3: Add Multiple Knowledge Sources (~20 minutes)

  1. Navigate to the Knowledge section
  2. Add Knowledge Source 1 - Upload a File:
    • Select Add knowledge
    • Select Upload a file
    • Choose your Sales FAQ document and add to the agent alt text
    • Select Add to agent
    • Wait for status to reach "Ready"
    • Add description by clicking on the ... button > Edit button: This knowledge source can answer questions about the sales process, discount policies, and pricing guidelines alt text
    alt text
    • Click on the Save button after entering the description.
  3. Add Knowledge Source 2 - Public Website:
    • Select Add knowledge
    • Select Public website
    • Enter URL: https://support.microsoft.com (or your company sales documentation site) and select Add
    • Add description: Contains support articles and best practices for client solutions alt text
    • Select Add to Agent
    • Wait for the knowledge source to reach "Ready"

Step 4: Test Your Agent with Multiple Knowledge Sources (~15 minutes)

  1. Open the Test pane on the right side
  2. Refresh the test window to load the updated agent with all knowledge sources
  3. Test with different types of questions to trigger each knowledge source:
    • For uploaded file: What is our discount policy?
    • For website: How do I configure a Microsoft product?
  4. Observe the Activity Map to see:
    • Which knowledge sources are being referenced for each query
    • How the agent prioritizes and searches across multiple sources
    • The confidence level of responses
  5. Verify responses include:
    • Proper citations from the knowledge sources
    • References to which document/source was used
    • Accurate and relevant information alt text

Step 5: Validate Enterprise Data Grounding (~5 minutes)

  1. Verify your agent is properly grounded with enterprise data:
    • Check that the agent references internal documentation
    • Confirm responses include citations to your knowledge sources
    • Ensure the agent doesn't hallucinate or make up information
  2. Review the grounding process:
    • Knowledge sources provide the "ground truth" for the agent
    • The agent retrieves information from these sources when responding
    • Citations show users where information came from
  3. This enterprise grounding is critical for:
    • Maintaining accuracy and compliance
    • Building trust with users
    • Ensuring consistent company information

Step 6: Understand the Agentic Pattern (~5 minutes)

  1. Review the difference between a chatbot and an agent:
    • Chatbot: Responds to questions with knowledge
    • Agent: Performs actions, creates documents, sends emails, executes workflows
  2. In the next lab, you'll make your agent truly agentic by adding:
    • Topics: Structured conversations for specific tasks
    • Tools: Power Automate flows that create documents and send emails
  3. These components transform your agent from a knowledge assistant into an automation platform

Duration

~60 minutes

Next Steps

Proceed to Lab 7: Add New Topic with Trigger and Nodes