What is Copilot Studio?
By now, all of the tech world and more know that if you’ve ever wanted to build your own AI-powered chatbot or intelligent assistant – without needing to be a coding wizard – Copilot Studio might be the new best friend. It’s a low-code tool with a super intuitive, drag-and-drop interface that makes creating smart agents feel way less intimidating. Even if you’re a business analyst or project manager with limited technical skills, you’ll feel right at home. Plus, it plays really well with the Microsoft ecosystem – like Microsoft 365, Dynamics, and Power Platform.
Key Features of Copilot Studio:
Just to recap the features of the Studio,
- Low-Code Agent Development: Use a drag-and-drop interface and visual workflow designer to map out conversation flows and interaction scenarios.
- Pre-Built Components and Templates: A robust library of pre-built components and industry-specific templates accelerates development.
- Advanced Customization Capabilities: Tailor AI agents to specific business processes, industry requirements, and unique organizational challenges.
- Integration: Supports integration with Microsoft 365, Power Platform, Dynamics 365, custom API connections, and third-party services.
- Scalability: Designed to handle varying levels of complexity and scale, with cloud-native architecture enabling automatic scaling, high availability, and robust security.
Step-by-Step: Creating Your First Declarative Agent
Our goal is to create a “New Hire Agent” that can answer questions based on an internal employee benefits guide.
Prerequisites
- Access to Microsoft Copilot Studio (you can often sign up for a free trial to build and test).
- Your employee benefits guide document (e.g., a Word or PDF file).
1. Access Copilot Studio and Create a New Agent:
- Sign in to Copilot Studio at copilot.microsoft.com or copilotstudio.microsoft.com.
- On the home page, select “Create an agent”.
- You can choose to “Describe” your agent in natural language or “Skip to configure” for a more structured setup. You also have templates for For this guide, let’s use the “Configure” tab to specify details.
2. Name and Describe Your Agent:
- Give your agent a clear and descriptive name. For example, “New Hire Agent”. The name can be up to 30 characters long.
- Provide a description that outlines its overall purpose and why it exists. Description must be in detail to explain the purpose of the agent well. It is IMPORTANT as the description helps the underlying large language model understand its function.
For instance, ” Support new employees in navigating their first 90 days, answering onboarding questions, providing company policy information, connecting them to the right internal resources, and helping them feel welcomed and productive.”
3. Define Agent Instructions:
- Instructions are crucial as they define the agent’s behavior, tone, and limitations. Think of this as instructing a human assistant.
- Purpose: Clearly outline the agent’s role and expertise. Example: “You are an agent dedicated to assisting new employees by providing clear, accurate, and friendly guidance on company policies, benefits, and common HR questions.”.
Guidelines & Tone: Define its communication style, tone, context capabilities. The more descriptive you are, the better the agent will understand its role.
Example: “You are the New Hire Agent. Your goal is to support new employees during their onboarding process. Assume the user is a new team member unless otherwise stated. Be proactive, friendly, and concise.
Capabilities:
– Answer company policy, IT, HR, and tooling-related questions.
– Provide quick access to onboarding resources and documentation.
– Help users complete first-week/month onboarding checklists.
– Route issues to the right team (HR, IT, Legal, etc.) when needed.
– Recognize the user’s department and role (when provided) to personalize the help.
Do not assume information if unsure – guide the user to the right contact or documentation.
Tone: Friendly, encouraging, and professional. Always reassure the user that it’s okay to ask questions as a new hire.
Prompt examples:
– “Welcome aboard! What do you need help with today?”
– “Would you like to see your onboarding checklist?”
– “I can help you book a 1:1 with your team lead – want me to do that?”
Context: You may refer to internal documents, company wiki, benefits portal, or internal directory if available.
You should always aim to reduce first-week friction and create a warm onboarding experience.”
4. Add Knowledge Sources:
This is where your agent learns. For a retrieval agent, providing relevant, high-quality data is key.
- In the “Knowledge” section, select “Add knowledge”.
- You have several options:
- Upload File: Drag and drop your employee benefits document (e.g., Word, PDF). This is suitable for relatively static documents.
- SharePoint Site: Connect to a SharePoint site or specific document libraries where your documents are stored. This is ideal for internal use cases where users are logged in.
- Public Website: You can connect to a public website URL.
- Dataverse: Connect to data stored in Microsoft Dataverse, which can include data from CRM systems or other business applications.
- Important Note: For a retrieval agent that should only use your provided data, disable “Allow the AI to use its own general knowledge.” This prevents the agent from fabricating answers from its large language model knowledge base if it can’t find information in your sources.
- You can add up to 20 knowledge sources. For our “New Hire Agent,” let’s assume we uploaded the employee benefits Word document to a SharePoint library and connected to it.
5. Configure Starter Prompts (Optional but highly recommended)
- Starter prompts suggest ways for users to initiate conversations with your agent.
- On the “Overview” page, find the “Starter prompts” section and select the “Edit” icon.
- Add up to six prompts related to your agent’s purpose, e.g., “What is my 401K benefit?”, “How much time off do I get per year?”, “How do I set up my work laptop?” ,”What’s the dress code?”.
6. Create Your Agent:
- Once you’ve configured the name, description, instructions, and knowledge sources, select “Create”.
- The agent will provision in the background, which might take a few moments.
7. Test Your Agent
Testing is a crucial iterative process to ensure your agent performs as expected.
- After creation, you’ll see a test chat panel. Use this to ask questions related to your knowledge source.
- For our “New Hire Agent,” try asking:
- “What is my 401K benefit?”
- “How much time off do I get per year?”
- Verify the responses against your actual document. The agent should provide accurate information directly from your uploaded source.
- If you change the instructions (e.g., tone of voice), re-test to see how the responses change. You can also change the introductory message for your agent.
8. Publish Your Agent
To make your agent available for others to use, you need to publish it.
- From the agent’s overview page, select “Publish” at the top, then confirm.
- Once published, you can often generate a demo website URL to share.
- To share within your organization (e.g., via Teams chat), select “Share” and configure permissions (e.g., “Anyone within your organization”).