Microsoft 365 Copilot’s AI-powered declarative agents are a great way to quickly and easily create customized agents for different use cases without complex coding or demanding AI skills.
Prerequisites: Microsoft 365 Copilot and Microsoft 365 licenses.
A declarative agent, or AI bot, allows you to customize Microsoft 365 Copilot according to user needs. When building them, they are given instructions, functions, and data sources that will enable them to respond to the user’s needs accordingly.
Examples of declarative agents:
- Translator Agent, which translates texts related to blogs or press releases and utilizes specified instructions or vocabulary.
- Drafting Agent, which utilizes previously created content. Its source of information includes previously created press releases, publications, or other texts, or the necessary information for new text in different forms.
- Project Management Agent, which answers questions related to project management and uses the project management guide as its source of information.
- Competence Agent, which can answer questions about internal responsibilities and expertise within the company. It includes information on who is responsible for each company function and provides contact details for people or teams if needed.
Microsoft’s AI bot is based on the same technology as Microsoft 365 Copilot and is an excellent way to limit and customize Copilot. This article will discuss when a declarative agent is the best solution, how to build one, and how to deploy one.
How does a declarative agent for Microsoft 365 Copilot differ from other agents?
The differences between declarative agents and customizable agents are listed in the table below:
| Feature | Declarative Agents | Customizable Agents |
| Usage | Task-specific application of Microsoft 365 Copilot | Use complex workflows or custom AI systems |
| Customization | Limited Copilot models and functionalities | Customizable AI features |
| Proactive interactions | Not supported | Supported |
| Channels | Integrated into M365 applications | Available in M365 applications and other channels |
| AI Engine | Hosted in Microsoft 365 | Copilot Studio or Azure AI |
Declarative agents are the right solution when you want the agent to operate according to Copilot language and behavior models and to be deployed quickly without complex coding or configuration.
In other situations that require more complex features, logic, or sharing options, you should implement a Microsoft AI bot using Copilot Studio or Azure AI.
Rapid development and sharing
Creating declarative agents in both SharePoint and M365 Chat is very similar. They can be created in a chat window with the help of AI or manually configuring instructions and other agent settings.
In the chat window, agent creation happens by describing its features and answering questions. The creation process in the chat occurs in the language set by the user’s Microsoft account language settings.
The agent can also be configured more manually on the Configure tab. You can set a name, icon, and description for the agent, which will be visible to users. Additionally, you can set text-based instructions for the agent’s behavior. These instructions should be described in detail, as they directly affect the agent’s behavior and have a limit of 8000 characters.
The instructions should include the following:
- General guidelines for the agent’s operation
- The tone in which the agent speaks
- Restrictions (topics the agent should avoid)
- Conversation examples
- Step-by-step instructions for the conversation
A declarative agent can be provided with data sources such as websites or SharePoint data sources. You can add individual files, folders, libraries, or sites from SharePoint as data sources, and they adhere to the user’s permissions. This means the agent’s knowledge depends on the added items and their sharing settings.
Additionally, you can provide the agent with external data sources linked through Microsoft Search, such as the Jira connection shown in the example.
If desired, you can provide the agent with additional capabilities that expand its skills. The code interpreter allows the agent to create and run code to support various calculations and data analyses. The image generator enables the agent to create images using Copilot Designer.
You can also set up introductory messages for the agent, which help the user start the conversation faster and better understand the agent’s capabilities.
You can continuously test the agent’s functionality while creating or modifying its settings. For example, you can directly test the added data source or instructions and adjust them flexibly.
When the agent is ready, you can publish it by selecting “Create.” After this, you can set three different sharing options for the agent:
- Everyone in my organization
- Selected permission groups
- Only you
If you share the agent with others, you do so via a link available from the view. Therefore, the agent cannot be shared, for example, on a Teams channel; instead, the use of declarative agents always occurs through Microsoft 365 Copilot.
An easy-to-use AI bot to solve your organization’s challenges
Declarative agents enable organizations to create consistent and personalized experiences and automate processes ranging from team onboarding to effectively resolving customer issues.
Microsoft’s AI bots offer the ability to connect multiple data sources and add-on capabilities, making them powerful tools for your organization’s needs.
Read more about declarative agents for Microsoft 365 Copilot on Microsoft’s site.
Janne Häkkinen
Business Intelligence Consultant
