Create an AI chatbot in CoPilot - Teachers Tech channel
https://www.youtube.com/watch?v=Iz3ul6hSeTs Here is a Markdown summary of the video tutorial on building an AI Agent in Microsoft 365 Copilot.
Building a Custom AI Agent with Microsoft 365 Copilot
In this video, Jamie from Teachers Tech demonstrates how to build a custom AI customer service agent from scratch using Microsoft 365 Copilot. The tutorial uses a fictional business, “Pooch Haven Spa,” to create an internal bot named “PoochPal” without writing any code.
🔑 Key Concepts
- Target Audience: Requires a Work or School account (not available for personal accounts).
- Platform Distinction:
- M365 Copilot (This video): Used for internal knowledge base bots within an organization.
- Copilot Studio (Next video): The advanced “workshop” used for creating public-facing bots and complex automations.
🛠️ Step-by-Step Guide
1. Getting Started
- Log in to Microsoft 365 Copilot.
- Select Create Agent from the dashboard.
- Initial Prompt: Use natural language to describe the agent.
- Example: “A friendly AI assistant for Pooch Haven Spa customer inquiries.”
2. Configuration & Branding
- Naming: The agent was renamed from “New Agent” to “PoochPal”.
- Icon: A custom image (created via AI) was uploaded to replace the default icon.
- Description: Edited to include specific details, such as a support email address.
3. Defining Instructions (The “Employee Handbook”)
This section defines how the AI behaves. Copilot generates a draft based on the initial prompt, but custom rules should be added.
- Tone: Set to be “cheerful and professional.”
- The “No Hallucination” Rule: Jamie added a critical instruction to prevent false information: “If you don’t know the answer to a question from the provided documents, do not make one up. Instead, politely direct the user to email our support team…“
4. Building the Knowledge Base (The “Brain”)
You can connect various sources for the agent to reference when answering questions:
- Websites: Enter a URL (e.g., the company website). The agent will scrape this for info.
- Files: Upload documents like PDFs, Word docs, etc.
- Example: Uploaded “Frequently Asked Questions” and “Services & Pricing” documents.
- Note: If an uploaded file changes, it must be re-uploaded. Linked websites update automatically.
5. Capabilities & Prompts
- Capabilities: Options to enable Code Interpreter (for math/data) or Image Generator. (Left off for this specific bot).
- Suggested Prompts: These are the clickable buttons users see to start a conversation. You can add, delete, or modify these to guide users on what to ask.
🧪 Testing the Agent
Before publishing, the agent can be tested within the configuration window:
- Success Test: Jamie asked “When are you open?” The agent correctly pulled the hours from the uploaded pricing document.
- Constraint Test: Jamie asked “Who are the owners?” (Information not in the documents). The agent followed instructions and provided the support email instead of making up names.
🚀 Publishing & Sharing
- Click Create to finalize the build.
- Sharing Settings: You can determine who has access to the agent:
- Only you.
- Specific users.
- Anyone in your organization.
- Once saved, the bot appears in the All Agents list and is ready for interaction.
Next Steps: This tutorial covered the basics of an internal knowledge bot. Part 2 will cover Microsoft Copilot Studio, focusing on advanced features like public access and automated tasks (e.g., sending emails). ====
Microsoft 365 (M365) Copilot is an AI-powered productivity assistant that integrates with your daily Microsoft apps. It combines the power of Large Language Models (LLMs) with your own organizational data—like your emails, documents, and calendar—to help you work more efficiently.
What can M365 Copilot do?
Copilot works across the entire Microsoft 365 suite, performing different roles depending on the app you are in:
- In Word: It can draft entire documents from a short prompt, summarize long reports, or rewrite text to change the tone (e.g., from casual to professional).
- In Outlook: It can summarize long email threads, draft replies based on your existing files, and help you “catch up” on your inbox by highlighting key action items.
- In Teams: It can summarize meetings in real-time, even if you join late. It can also identify who said what and list the next steps or action items decided during the call.
- In Excel: It helps you analyze data, identify trends, and generate formulas or visualizations (charts) simply by asking questions in plain English.
- In PowerPoint: It can transform a Word document into a full presentation deck, complete with speaker notes and images.
How is it used?
There are two primary ways to interact with it:
- Within Individual Apps: You’ll see a small Copilot icon in your toolbar. Clicking it opens a side panel where you can give it specific instructions related to the file you are working on.
- Copilot Chat (shown in your image): This is a dedicated “work” version of an AI chatbot. You can use it to ask complex questions that require pulling information from multiple sources (e.g., “Summarize all the emails and documents I have regarding Project X from the last week”).
Key Features shown in your image
The screenshot you provided highlights some newer, advanced features of the M365 Copilot app:
- Create Agent: This allows you to build a custom “mini-AI assistant” (like the “PoochPal” example in the video) that is trained on a specific set of your files or websites. This is useful for creating a specialized bot that only answers questions about one specific project or company policy.
- Work vs. Web Toggle: You can choose between Work (where the AI strictly uses your private company data to answer) or Web (where it acts more like a search engine to find public information).
- Enterprise Data Protection: Unlike free AI tools, M365 Copilot includes a “shield” (security) that ensures your prompts and business data are never used to train the underlying AI models, keeping your sensitive information private.