What You Will Learn

• Build Copilot Agents using Copilot Studio without coding
• Integrate knowledge sources like websites, PDFs, SharePoint, and Dataverse
• Design conversation flow with topics, entities, and variables
• Extend agents using actions and Power Automate for real-time data
• Publish and deploy agents to websites, Microsoft Teams, and other channels

Study Guide

Introduction

Welcome to the comprehensive guide on building Copilot Agents using Copilot Studio. This course is designed to empower you to create intelligent conversational AI agents, known as Copilot Agents, without needing extensive coding knowledge. These agents can transform how businesses interact with customers and manage internal processes, offering tailored, efficient, and engaging user experiences. Throughout this guide, we'll delve into the core components and processes involved in building these agents, providing you with the knowledge and skills to harness the full potential of Copilot Studio.

Empowering Users to Build Intelligent Agents

The primary focus of this course is to demonstrate how users, even those with no prior coding experience, can create sophisticated conversational AI agents. Copilot Studio offers a guided, step-by-step approach, making it accessible to beginners and ensuring a smooth learning curve.

Example 1: Imagine a small business owner who wants to automate customer service inquiries. With Copilot Studio, they can build a Copilot Agent that answers frequently asked questions, freeing up time for other tasks.
Example 2: A teacher could use a Copilot Agent to provide students with instant feedback on assignments, creating a more interactive learning environment.

Integrating Diverse Knowledge Sources

Integrating relevant information from various sources is crucial for building effective Copilot Agents. This includes public websites, uploaded documents like PDFs, and potentially SharePoint and Dataverse. Providing your agent with diverse knowledge sources ensures it can deliver accurate and comprehensive responses.

Example 1: A Copilot Agent for a travel agency might pull data from travel blogs and uploaded brochures to offer destination recommendations.
Example 2: A healthcare Copilot Agent could access medical journals and patient information (with appropriate permissions) to provide health advice.

Controlling Conversation Flow with Topics

Topics are predefined conversational paths triggered by user input, allowing for controlled and specific responses beyond purely generative AI answers. Managing topics effectively is crucial for ensuring your Copilot Agent delivers accurate and relevant responses.

Example 1: A customer service Copilot Agent might have a topic for handling refund requests, guiding the user through the necessary steps.
Example 2: An educational Copilot Agent could use topics to provide step-by-step explanations of complex subjects, like algebra or chemistry.

Leveraging Entities and Variables for Contextual Understanding

Entities and variables enhance a Copilot Agent's ability to understand and respond dynamically. Entities are recognizable pieces of information within unstructured text, while variables store and reuse information within a conversation.

Example 1: A restaurant Copilot Agent might recognize an entity like "vegetarian" and store it as a variable to recommend appropriate menu items.
Example 2: A financial services Copilot Agent could identify a user's age and use it to offer tailored investment advice.

Extending Functionality with Actions

Actions enable your Copilot Agent to interact with external systems and retrieve real-time information. This can include accessing data from sources like Excel spreadsheets using Power Automate flows.

Example 1: A weather Copilot Agent could use actions to fetch the latest weather forecast from an online service.
Example 2: An HR Copilot Agent might retrieve employee data from an Excel sheet to assist with payroll inquiries.

Deployment and Integration

Publishing your Copilot Agent to various channels is the final step in making it accessible to users. This can include demo websites, Microsoft Teams, and external customer service systems. Integration with Microsoft 365 Copilot extends the functionality of your agent.

Example 1: A retail Copilot Agent could be integrated into a company's website to assist customers with product inquiries.
Example 2: A support Copilot Agent might be deployed in Microsoft Teams to help employees with IT issues.

The Evolving Landscape of Copilot Agents

The field of Copilot Agents is rapidly evolving, with new features like enhanced AI-powered topic creation, voice capabilities, and more autonomous agents on the horizon. Staying up-to-date with these developments is crucial for maximizing the potential of your Copilot Agents.

Example 1: A future update might allow for voice-activated Copilot Agents, making interactions more natural and intuitive.
Example 2: Enhanced AI capabilities could enable Copilot Agents to handle more complex queries, improving their usefulness across various industries.

Conclusion

By completing this course, you have gained a comprehensive understanding of building Copilot Agents with Copilot Studio. You have learned to integrate diverse knowledge sources, manage conversation flow with topics, leverage entities and variables, extend functionality with actions, and deploy your agents across multiple channels. These skills empower you to create intelligent, efficient, and engaging Copilot Agents that can transform how businesses interact with customers and manage internal processes. Remember, the thoughtful application of these skills is key to unlocking the full potential of your Copilot Agents.

Podcast

There'll soon be a podcast available for this course.

Frequently Asked Questions

Welcome to the FAQ section for the 'Video Course: Building Copilot Agents with Copilot Studio'. This resource is designed to address common questions and provide insights into the development of co-pilot agents. Whether you're new to this field or an experienced practitioner, you'll find answers to help you navigate the complexities of building and deploying co-pilot agents effectively.

What are the core components involved in building a co-pilot agent?

Building a co-pilot agent involves several key elements working together. Firstly, understanding entities and variables is crucial for the co-pilot to grasp the context and specific details within a conversation. Actions allow the co-pilot to interact with external systems and data sources, such as retrieving information or performing tasks on the user's behalf through flows. You also need to consider publishing options to make your co-pilot accessible to users on various channels. Underlying all of this is the ability to provide the co-pilot with knowledge, which can come from websites, documents, and other data sources, and to define the conversational flow and behaviour through topics.

How can I provide my co-pilot agent with relevant information and knowledge?

You can equip your co-pilot with knowledge in several ways. One method is by connecting it to public websites, allowing it to search and extract information from the site's content. You can also upload documents, such as PDFs and Word documents, which the co-pilot can then process and use to answer questions. More advanced options include connecting to SharePoint document libraries and Dataverse tables, although these require more setup. It's important to describe your knowledge sources clearly to help the AI understand their purpose. Be aware that websites with heavy JavaScript or information presented solely in tables might not be easily readable by the co-pilot.

What are topics and why are they important in co-pilot development?

Topics define the conversational paths and how your co-pilot will respond to specific user inputs. They consist of trigger phrases (in the "classic" approach) or rely on generative AI to understand user intent based on topic descriptions (in the newer preview experience). Topics allow you to have precise control over how the co-pilot interacts, ensuring it provides accurate and appropriate responses for critical or legally sensitive queries. You can create custom topics for specific scenarios or utilise system topics for fundamental functionalities like conversation greetings and error handling. It's essential to manage topics effectively, disabling any that might conflict with your co-pilot's intended purpose.

How can entities and variables enhance my co-pilot's understanding and interaction?

Entities represent specific pieces of information that the co-pilot can recognise within user input, such as ages, dates, locations, or custom defined terms. This allows the co-pilot to understand the meaning behind unstructured text. Variables are used to store information captured during a conversation, such as an identified entity. This stored information can then be used later in the conversation to personalise responses, perform actions, or create conditional logic within topics. Variables can be scoped to a specific topic or be made global for use throughout the entire co-pilot agent.

What are actions and how can they extend the capabilities of my co-pilot?

Actions enable your co-pilot to go beyond simply answering questions by interacting with external systems and data. You can create actions in several ways, including using pre-built connectors (like a weather forecast service) or by creating Power Automate flows. Actions allow the co-pilot to retrieve real-time information, perform tasks, and integrate with your existing business processes. For example, you could use an action to fetch the pollen forecast from an online spreadsheet based on the user's location, demonstrating how to overcome limitations of reading complex website structures.

How can I control the flow of conversation within a topic?

Within a topic, you have several tools to control the conversation flow. You can ask questions to gather specific information from the user, send messages as responses, and use conditional branching to direct the conversation based on the user's input or identified variables. For example, you can ask for a person's age (identifying it as an entity) and then branch the conversation to provide different mental health support service recommendations based on whether they are a child, teenager/young adult, or adult.

What are the options for publishing my co-pilot agent and making it accessible to users?

Once your co-pilot is built, you have various publishing channels to choose from depending on your target audience. For internal use, you can publish to Microsoft Teams. For external access, you can embed it on your website, use it through platforms like Twilio or Facebook, or integrate it into a mobile app. Microsoft also offers a demo website option for testing and practice. The available channels might depend on your co-pilot's configuration, particularly its security settings and authentication requirements.

How does building custom co-pilot agents relate to Microsoft 365 Copilot and the broader Microsoft ecosystem?

Building custom co-pilot agents using tools like Copilot Studio allows you to create tailored AI assistants for specific scenarios and knowledge domains. This complements the broader Microsoft 365 Copilot offering, which is integrated into Microsoft applications and can access your organisation's data (with appropriate permissions). The skills learned in building custom agents can be used to extend the functionality of Microsoft 365 Copilot by connecting it to other systems and data sources relevant to your business. Furthermore, Microsoft is introducing the concept of building co-pilot agents directly within Microsoft 365 Copilot, blurring the lines further and highlighting the growing importance of this skillset within the Microsoft ecosystem.

What are generative answers and when should they be used or disabled?

Generative answers refer to the co-pilot's ability to use its underlying large language model to synthesize responses based on the information it has access to. You might disable this feature to ensure the co-pilot only provides answers based on specific, curated knowledge sources and avoids potentially inaccurate or off-topic responses from its general knowledge. This is particularly important in scenarios where accuracy and relevance are critical.

How can I define the personality or behaviour of my co-pilot agent?

During the initial setup and within the instructions section of the co-pilot configuration, you can define custom instructions specifying the desired tone (e.g., friendly, casual), the level of detail, and any specific guidelines like avoiding jargon or using emojis. This helps ensure that the co-pilot's responses align with your brand's voice and the expectations of your users.

What is the conversation map and how does it help in co-pilot development?

The conversation map is a visual tool that represents the flow of a conversation, showing how the co-pilot makes decisions based on user input. It helps developers understand the interaction pathways, identify potential bottlenecks, and refine the conversational design for improved user experience. By visualising the conversation, you can more easily spot areas for enhancement and ensure logical progression.

How can I enable my co-pilot to retrieve real-time information?

To enable real-time information retrieval, you can use actions that connect to external data sources. For example, if a website's table data is not easily readable by AI, you could create a Power Automate flow that accesses the data (e.g., from an online spreadsheet), takes the user's region as input, retrieves the corresponding forecast, and then configure an action within the co-pilot to call this flow and present the result to the user.

What are some potential channels for publishing a co-pilot agent?

A built co-pilot agent can be published through various channels, including embedding it on a public-facing website using a provided code snippet, integrating it into Microsoft Teams for internal use, connecting it to customer service platforms, and potentially deploying it through mobile apps or social media platforms like Facebook. The choice of channel depends on your audience and the intended use of the co-pilot.

What are the potential benefits and drawbacks of using different types of knowledge sources?

Different knowledge sources offer unique advantages and challenges. Public websites provide a broad range of information but may have readability issues due to JavaScript or complex table structures. Uploaded documents offer controlled and specific content but require regular updates. SharePoint and Dataverse provide structured, enterprise-level data access but involve more complex setup and permissions management. Choosing the right mix of sources impacts the co-pilot's effectiveness in delivering accurate and relevant responses.

How has topic design evolved from trigger phrases to generative AI?

Topic design has shifted from relying on explicit trigger phrases to leveraging generative AI for intent recognition. The classic method uses specific words or phrases to trigger topics, which provides precise control but requires extensive setup. Generative AI allows for more natural language processing, understanding user intent from broader descriptions. This evolution improves flexibility and user experience but may require more nuanced training to ensure accuracy and relevance.

Why are entities and variables significant in creating dynamic co-pilot conversations?

Entities and variables play a crucial role in personalising and enhancing co-pilot interactions. Entities allow the co-pilot to extract specific information from user input, while variables store this data for later use, enabling tailored responses and actions. This dynamic approach leads to more natural conversations and efficient interactions. Best practices include defining clear entity types and managing variable scopes effectively to maintain context throughout the conversation.

What are key considerations when building and deploying a co-pilot agent?

Building and deploying a co-pilot agent involves defining its purpose, selecting and configuring knowledge sources, designing topics, and choosing appropriate publishing channels. Key considerations include ensuring data accuracy, managing user expectations, and maintaining security and privacy standards. Challenges may arise in integrating with existing systems and adapting to user feedback. A systematic approach, including thorough testing and iterative refinement, contributes to a successful deployment and user satisfaction.