An introduction to Microsoft 365 Copilot Agents

Creating, extending and customising Copilot agents is a fundamental process for adapting functionality to specific business or personal needs.

In this article, I will provide an overview of Copilot Agents, explaining what they are, which out-of-the-box agents Microsoft has released to date, what types of agents we can customize, and I will also explain what autonomous agents and SharePoint Agents are.

What are Copilot Agents?

Agents are scoped versions of Microsoft 365 Copilot that act as AI assistants to automate and run business processes. They enable customers to bring custom knowledge, skills, and process automation into Microsoft 365 Copilot for their specific needs. 

Pre-built Copilot agents

Microsoft has launched a wide range of pre-built agents that span multiple business functions. These agents can be deployed immediately or further configured by incorporating your organisation’s knowledge and skills.

At the time I'm writing they are all in preview. A brief description of the available pre-built agents is provided here below:

• Website Q&A: Answers common questions from users using the content on your website.
• Team Navigator: Assists employees in finding colleagues and their hierarchy within the organisation.
• IT Helpdesk: Empowers employees to resolve issues and create/view support tickets.
• Store Operations: Improves the efficiency of retail frontline workers by enabling easy access to store procedures and policies.
• Case Management: Provides around-the-clock automated support to customers by understanding their issues and creating cases.
• Safe Travels: Provides answers to common travel questions and related health and safety guidelines.
• Inclusivity: Helps employees to have a safe place to ask questions and to learn how to activate inclusivity in a modern and diverse workforce.
• Sustainability Insights: Enables users to easily get insights and data about a company’s sustainability goals and progress.
• Weather: Gets the current weather conditions and forecast.
• Benefits: Provides personalized information to your employees on benefits offered to them.
• Citizen Services: Enables Public Sector Organizations to assist their citizens with information about services available to them.
• Financial Insights: Helps financial services professionals get information from their organization’s financial documents.
• Self-Help: Enables customer service agents to resolve issues faster.
• Awards and Recognition: Streamlines the process of nominating and recognizing your employees for their contributions and achievements.
• Leave Management: Streamlines the leave request and time-off process for your employees.
• Wellness Check: Conducts automated wellness checks to gauge employee morale.
• Sales Qualification: Enables sellers to focus their time on the highest priority sales opportunities.
• Sales Order: Automates the order intake process from entry to confirmation by interacting with customers and capturing their preferences.
• Supplier Communications: Autonomously manages collaboration with suppliers to confirm order delivery, while helping to prevent potential delays.
• Finance Reconciliation: Helps teams prepare and cleanse data sets to simplify and reduce time spent on the financial period close process.
• Account Reconciliation: Automates the matching and clearing of transactions between subledgers and the general ledger, helping accountants and controllers speed up the financial close process.
• Time and Expense: Autonomously manages time entry, expense tracking, and approval workflows.
• Customer Intent: Enables evergreen self-service by continuously discovering new intents from past and current customer conversations across all channels, mapping issues and corresponding resolutions maintained by the agent in a library.
• Customer Knowledge Management: Helps ensure knowledge articles are kept perpetually up to date by analysing case notes, transcripts, summaries, and other artifacts from human-assisted cases to uncover insights.
• Scheduling Operations: Enables dispatchers to provide optimized schedules for technicians, even as conditions change throughout the workday.

How do I create a Copilot Agent?

Before diving into how we can extend Copilot, it is essential to understand two concepts:

• the anatomy of Microsoft 365 Copilot,
• the two main types of agents that we can create: Declarative Agents and Custom Engine Agents.

Anatomy of Microsoft 365 Copilot

Foundation Models

Foundation models are large language models (LLMs) that form the core of Copilot's capabilities. These models, such as GPT-4, are trained on vast amounts of data and use deep learning techniques to understand, summarize, predict, and generate content. They provide the underlying AI intelligence that powers Copilot's functionality.

User Experience

The user experience component focuses on how users interact with Copilot within Microsoft 365 apps like Word, Excel, PowerPoint, Outlook, and Teams. It ensures that the integration is seamless and intuitive, allowing users to leverage Copilot's capabilities without disrupting their workflow. This includes features like drafting, summarizing, and answering questions in the context of the user's work.

Orchestrator

The orchestrator is responsible for coordinating the various components of Copilot and ensuring that they work together harmoniously. It manages the flow of information between the foundation models, user data, and the Microsoft 365 apps, ensuring that responses are accurate and relevant to the user's context.

Knowledge

The knowledge component involves the integration of Microsoft Graph, which includes information about users, activities, and organizational data. This allows Copilot to access and utilize relevant data from emails, chats, documents, and meetings to provide contextually appropriate responses and insights.

Skills

Skills refer to the specific capabilities and functionalities that Copilot offers within different Microsoft 365 apps. For example, in Word, Copilot can help users create, understand, and edit documents; in Excel, it can assist with data analysis and visualization; and in Teams, it can facilitate communication and collaboration.

Declarative Agents

Declarative agents are designed to be configured through predefined rules and well-defined scenarios. These agents function on the basis of declarations of intent and specific conditions that guide their behaviour.

• Ease of Use: They do not require in-depth programming knowledge.
• Speed of Implementation: They can be set up quickly using intuitive graphical interfaces.
• Limitations: They might be less flexible when it comes to handling complex scenarios outside the preset rules.

Custom Engine Agents

Custom engine agents offer the highest level of control and customisation.

• Flexibility: They allow detailed customisation by writing code and integrating advanced algorithms.
• Power: They can handle complex and dynamic scenarios requiring sophisticated and adaptive logic.
• Development Effort: They require more technical expertise and more time for development and maintenance.

Blue pill or pink pill...?

When it comes to extending Copilot, you can decide to swallow the blue pill or swallow the pink pill…

Staying on the Blue side of things, you're going to reuse the foundations of Copilot: you're going to reuse the UI, you're going to reuse the templates and also the Microsoft Orchestrator. You're going to add Knowledge and Skills instead.

On the other hand, if you swallow the pink pill, you are going to completely replace the Copilot engine with your own custom.

Swallowing the blue pill: creating a Declarative Agent

Declarative agents are a collection of custom knowledge and custom skills hosted on the Microsoft 365 Copilot orchestrator and foundation models.

You can add knowledge to Declarative Agents via connectors, and skills via plugins.

Adding Knowledge

Graph Connectors

Microsoft Graph Connectors provide a platform to ingest your unstructured data into Microsoft Graph, so that Copilot can reason over the entirety of your enterprise content. Microsoft Graph Connectors existed before Copilot, in fact it powers other Microsoft 365 experiences, like Microsoft Search.

Power Platform Connectors

Power Platform Connectors are essentially API wrappers that allow Copilot Agents to interact with various external services and applications. These connectors enable Copilot to perform a wide range of tasks by connecting to services both within the Microsoft ecosystem (like Office 365, SharePoint, Dynamics 365) and outside it (such as Twitter, Google services, Salesforce). There are three main types of connectors:

1. Standard Connectors: These are included with all Copilot Studio plans and cover common services in Copilot Studio.
2. Premium Connectors: Available in select Copilot Studio plans, these offer more advanced functionalities in Copilot Studio.
3. Custom Connectors: These allow you to connect to any publicly available API for services not covered by existing connectors in Copilot Studio.

By leveraging these connectors, Copilot Agents can access and utilise data from various sources, enhancing their capabilities and making them more dynamic and responsive to specific business needs.

Adding Skills

Plugins

Plugins don't ingest data, they look directly in real-time at the external systems, they can also interact with the external systems, that is not only read data but also write data.

There are 3 types of plugin:

• Copilot Studio Actions: plugins that extend the functionality of Microsoft 365 Copilot, allowing users to perform specific tasks, retrieve data, and interact with external systems. By leveraging a low-code interface, Copilot Studio makes it accessible for users without extensive technical knowledge to create and manage these actions.
• Message Extensions: they are the well known search and action capability for Microsoft Teams, that now work also as plugins for Copilot agents.
• API Plugins: they enable declarative agents in Microsoft 365 Copilot to interact with REST APIs that have an OpenAPI description.

The pink pill: Custom Engine Agents

Custom engine agents are developed using custom foundation models and orchestrators and can be tailored to specific enterprise needs with your own stack. Custom engine agents currently work as standalone Teams apps.

These are an evolution of Microsoft Teams bots and, just like before, you can use Teams AI Library and Teams Toolkit to create them. I intend to come back to Custom Engine Agents with a dedicated article.

Declarative? or Custom Engine?

When to create a declarative agent:

• You want to take advantage of Copilot's model and orchestrator.
• You have external data that you want to make available to Copilot to reason over via a Microsoft Graph connector.
• You have an existing API that could be used as an API plugin for read and write access to real-time data.
• You have an existing Teams message extension that you can use as a plugin.

When to build a custom agent:

• You want to use specific models for your service.
• You need agentic AI support.
• You want your service to be independent from Microsoft 365 Copilot, accessible to all Microsoft 365 users regardless of their Copilot licensing status.

Autonomous Agents

An autonomous agent is an AI system that can perform complex tasks independently, that is without direct human intervention.

Let's see a typical example of an autonomous agent created with Copilot: a prospective client sends you an email requesting an engagement. As soon as the email is received, the agent gets to work, extracting the relevant details.

It then follows a series of steps, including verifying any previous engagement with this customer, validating the industry sector, summarising the client needs, it then writes and sends an email to the relevant expert in your organisation, with all the client details.

The particular feature that makes a Copilot Agent an autonomous agent is the fact that it can activate by itself: this is possible thanks to the event triggers, defined in Copilot Studio, which may kind of remind you the Power Automate triggers.

SharePoint Agents

SharePoint Agents are a specialised type of agents, tailored on SharePoint Online and that can be created outside of Copilot Studio. Every SharePoint site includes an agent scoped for the site’s content, ready to assist you instantly. These ready-made agents are an easy starting point to get answers without combing through the site or digging around with search—they can be used immediately without any customization.

For specific projects or tasks, any SharePoint user can create a customized agent based on the relevant files, folders, or sites, with just one click. 

The SharePoint Agents can easily be shared via email or within Teams chats. Not only are coworkers able to use the agent that you shared, but @mentioning the agent in a group chat setting gives the team a subject matter expert ready to assist and facilitate collaboration.  

They adhere to existing SharePoint user permissions, they don’t broadly share the files you selected whenever you share the agent with others in your organization.

Agents created using SharePoint data are file-based. They are stored within the same site where they were created. Since they are files, you can manage them just like you manage other files. You can copy, move, delete, or archive them.