Introduction
Microsoft 365 Copilot is an AI-powered productivity tool that uses large language models (LLMs) and integrates your data with the Microsoft Graph and Microsoft 365 Apps.
When you ask questions, Copilot reasons over your company data and uses the information that it gathered in its response. However, depending on the user intent, this approach isn't always optimal. What if you want Copilot to only focus on specific data? What if you want Copilot to focus on a specific task? How do you optimize Microsoft 365 Copilot for these scenarios?
Declarative agents give you the components to extend Microsoft 365 Copilot for specific scenarios.
Example scenario
Suppose you work in an IT support team. You and your team write knowledge base articles and store them in a SharePoint Online document library in PDF format. You also manage support tickets in a specialized system. You want to make a self-service IT support assistant that employees can use to get answers to common IT support queries using information in the knowledge base articles reducing the number of tickets raised. The knowledge base articles are technical in nature. You want the assistant to be able to create responses that are in plain English and avoid the use of jargon. You also want employees to be able to create and update support tickets using the assistant if they can't fix the issue themselves.
Here, you see how a declarative agent can improve the relevance and accuracy of responses to answer IT support questions. It does this task by using knowledge-base articles and interacting with the ticketing system.
What will we be doing?
We analyze the key features of declarative agents to help you decide when to use them:
- Custom knowledge: what data do we want our agent to reason over and how should the responses be shaped?
- Custom actions: what interactions do we need to perform with external systems?
What is the main goal?
By the end of this module, you're able to decide whether a declarative agent is a good choice for extending Microsoft 365 Copilot.