1:02:38
Make your Own Agents in Copilot | Complete Tutorial
Kevin Stratvert
Overview
This tutorial provides a comprehensive guide to creating and utilizing agents within Microsoft 365 Copilot. It explains what agents are, how they enhance AI capabilities for specific tasks, and the prerequisites for using them (Microsoft 365 business/enterprise accounts). The video covers working with pre-built agents like the Career Coach, Researcher, and Analyst, as well as integrating third-party agents such as Canva. It details how to build custom agents using Copilot Studio and the Agent Builder, including defining instructions, knowledge sources, and capabilities. Finally, it touches upon publishing agents and managing authentication for internal and external use, highlighting the role of administrators and potential costs associated with usage credits.
How was this?
Save this permanently with flashcards, quizzes, and AI chat
Chapters
- Agents are customized AI assistants that perform specific tasks within defined parameters.
- They can operate independently or create workflows across multiple applications.
- Using agents in Copilot requires a Microsoft 365 business or enterprise account, not the free or individual subscription.
- The course aims to help users understand agents, use existing ones, and build their own without coding.
Understanding agents is crucial for leveraging AI to streamline work, automate tasks, and achieve more focused results from Copilot.
The video contrasts a basic Copilot prompt for career change advice with a more effective prompt that assigns Copilot the role of an experienced career advisor, demonstrating how agents can handle more complex, role-based interactions.
- Pre-built agents, like the 'Career Coach,' are available within Copilot and Teams to assist with specific professional journeys.
- Agents can be added to your Copilot interface and invoked using the '@' symbol in chat.
- Third-party agents, such as the 'Canva' agent, integrate external applications into Copilot for enhanced functionality.
- It's important to understand an agent's parameters, capabilities, and constraints by asking it to describe them.
Familiarizing yourself with existing agents allows you to immediately benefit from specialized AI assistance without needing to build anything yourself.
The 'Career Coach' agent is demonstrated to guide users through career changes by asking clarifying questions and providing tailored advice, unlike a standard Copilot query.
- The 'Researcher' agent performs in-depth analysis by comparing multiple sources and can access organizational data.
- The 'Analyst' agent specializes in extracting insights from data, such as spreadsheets, and can identify trends and suggest strategies.
- These agents require the full Microsoft 365 Copilot license and are designed for complex, time-intensive tasks.
- Both agents can access enterprise search and organizational data, providing a deeper level of analysis than standard Copilot.
These powerful agents are pre-configured for specific, demanding tasks, saving significant time and effort compared to crafting complex prompts manually.
The 'Researcher' agent is shown analyzing an employee manual to identify outdated or missing policies, while the 'Analyst' agent is used to derive insights from a product inventory spreadsheet regarding backordered items.
- Third-party agents extend Copilot's capabilities by connecting to external applications like Canva.
- Adding a third-party agent requires granting it permission to access your Microsoft 365 data, which necessitates trust in the third-party provider.
- Agents can reference information from your emails, files, and messages to inform their actions, such as designing a graphic based on an email thread.
- While agents can initiate actions in connected apps, final edits or complex manipulations often require signing into the native application.
Third-party integrations transform Copilot into an AI-powered front-end for various applications, allowing you to leverage AI features within tools you already use.
The Canva agent is used to generate promotional graphics for an event based on details from an email thread, demonstrating how Copilot can interact with design software.
- Excel's agent mode allows Copilot to directly modify spreadsheets, unlike the standard chat interface.
- It can perform actions like sorting, filtering, calculating new columns, and applying conditional formatting based on natural language commands.
- Agent mode can also generate dashboards, tables, and graphs, and provide insights and recommendations based on the data.
- This mode enables independent decision-making by the AI within Excel, going beyond simple query responses.
The Excel agent mode significantly enhances data analysis capabilities by allowing Copilot to actively manipulate and interpret spreadsheet data, leading to deeper insights and automated reporting.
The agent mode is used to sort a spreadsheet by retail price, filter for specific product types, calculate profit margins, apply conditional formatting to highlight high-profit items, and generate a dashboard with recommendations for restarting production on backordered items.
- The Copilot Agent Builder allows users to create custom agents without coding skills.
- Key steps involve defining the agent's name, description, and detailed instructions (role, rules, constraints).
- Knowledge sources, such as uploaded documents or web links, can be added to provide context and data for the agent.
- Agents can be configured to search the public internet or be limited to provided knowledge sources.
- Testing and refining the agent through iterative prompts is crucial for its effectiveness.
Building your own agents empowers you to create tailored AI assistants that address specific organizational needs, saving time and effort for yourself and your colleagues.
An 'HR helper' agent is built by providing instructions like 'You are an HR director with 30 years experience' and uploading HR policy documents as knowledge sources, enabling it to answer employee questions accurately.
- Copilot Studio offers more sophisticated tools for building complex agents, including AI model selection and trigger configurations.
- It allows for manual definition of instructions, knowledge sources (uploaded files, websites), and capabilities (document generation, code snippets).
- 'Topics' are used to define specific conversational flows and responses for particular user queries, enhancing agent precision.
- Publishing an agent requires configuration of authentication, channels (Teams, Copilot), and administrator approval for wider deployment.
- Usage credits may be required for agents used by individuals without the full Copilot subscription or for external users.
Copilot Studio provides the advanced features needed to create highly customized and powerful agents, enabling sophisticated automation and AI-driven workflows within an organization.
An HR agent is built in Copilot Studio, using an AI assistant to generate initial instructions and uploading company manuals as knowledge sources. A specific 'conflict report' topic is then created to ensure users are directed to the HR director for sensitive issues.
- Agents built in Copilot Studio can be published to various channels, including Microsoft Teams and the Copilot interface.
- Authentication settings determine who can access the agent, ranging from requiring Microsoft sign-in to no authentication for public access.
- Sharing an agent with teammates requires administrator approval and depends on their Microsoft 365 subscription level.
- Publishing externally requires careful consideration of authentication and potential usage credit costs.
- Agents can be shared within an organization via a link, allowing colleagues to add them to their Copilot.
Successfully publishing and sharing agents ensures that the custom AI tools you build can be utilized by the intended audience, maximizing their value and impact within your organization.
An agent is published to Teams and Copilot, requiring Microsoft authentication for internal users. The process involves submitting the agent to an administrator for approval before it becomes available in the agent store for colleagues.
Key takeaways
- Agents transform Copilot from a general AI assistant into specialized tools for specific tasks, significantly boosting efficiency.
- A Microsoft 365 business or enterprise account is essential for accessing and building advanced Copilot agents.
- Leveraging pre-built and third-party agents offers immediate benefits and showcases the potential of AI integration.
- Understanding an agent's parameters, capabilities, and limitations is key to using it effectively.
- Building custom agents requires careful planning of instructions and knowledge sources to ensure accurate and relevant responses.
- Copilot Studio provides advanced capabilities for creating sophisticated agents, including custom conversational flows (topics).
- Proper publishing and authentication management are critical for ensuring agents are accessible to the right users and secure.
- The distinction between agents built with Copilot Agent Builder (no-code) and Copilot Studio (more advanced) is important for managing complexity and features.
Key terms
AgentMicrosoft 365 CopilotCopilot StudioCopilot Agent BuilderInstructionsKnowledge SourcesParametersCapabilitiesConstraintsTopicsAuthenticationChannelsUsage Credits
Test your understanding
- What is the primary difference between a standard Copilot prompt and using a Copilot agent?
- Why is a Microsoft 365 business or enterprise account necessary for using advanced Copilot agents?
- How can you discover the specific functions and limitations of a pre-built or third-party agent?
- What are the key components to define when building a custom agent using the Copilot Agent Builder?
- Explain the role of 'Topics' in Copilot Studio for customizing agent behavior.
- What are the main considerations when publishing an agent for internal versus external use?