What are Copilot agents and are they right for my business?

As we move into another crucial phase for business communications, the possibilities for artificial intelligence (AI) assistance is becoming somewhat the norm.

While we’re not talking about totally offloading work to robots, we can confidently lean on AI components for help with menial tasks, catching up on items we missed during busy periods, and kickstarting drafts for new documents, spreadsheets, and presentations, etc.

Copilot agents, one of the latest additions to the Microsoft 365 stack, show that Microsoft isn’t just flexing its muscles when it comes to AI power. Rather, it’s learning where businesses can benefit from machine learning and cross-ecosystem integration and introducing automations and in-app help to streamline processes and ease the burden of “busywork”.

Contents

• What is a Copilot agent?
• Types of Copilot agents
• What is the difference between a Copilot and an agent?
• What's the use case for Copilot agents?
• How do Copilot agents work?
• How to access Copilot agents
• Copilot agent pricing
• Thinking about introducing Copilot agents?
  
  
  

A Copilot agent is an in-app virtual assistant you can pose questions to and get proactive prompts and reminders from. It learns when tasks need automating, completing, or flagging to help speed up your daily tasks.

The major focus is there’s less reliance on humans to prompt the activity. Copilot agents know what needs to be completed using AI, grounded data, and integration with the Microsoft apps you grant permission to.

There are three core competencies of a Copilot agent:

1. Retrieval: At the most basic level, a Copilot agent can retrieve information when prompted. Think about when you need to ask for a template to start a new sales proposal.
2. Task-conducting: As we progress through the spectrum, Copilot agents carry out tasks when asked. Here, they won’t just search for that template, but apply the template to your copy and images.
3. Autonomous: On the far right-hand side of the scale, you don’t even need to ask. Copilot agents conduct operations independently. When a sales proposal for a specific client gets mentioned during a meeting, and you mention you need to create this, your Copilot agent has already checked to retrieve the template, pricing, and custom details you need.

While you may have used Copilot to ask a question when you’ve been stuck on a task or needed a helping hand formatting a document, the latest Copilot agents conduct background activity, meaning there’s less discovery of tasks Copilot could help with. 

Instead, when granted permission at an admin level, Copilot is autonomous.

Instead of having a Copilot that waits there until someone chats with it, what if you could make your Copilot more proactive and for it to be able to work in the background on automated tasks.
- Charles Lamanna, Corporate VP of Business Apps and Platforms at Microsoft.

Types of Copilot agents

Originally launched as part of the Dynamics 365 AI offering, Copilot agents come templated for different use cases:

• Sales Qualification Agent
• Supplier Communications Agent
• Customer Intent Agent
• Customer Knowledge Management Agent

As you can imagine, these all highlight insights into the sales process, customer journey, and general communications between parties in a traditional Contact Center setup. But they also carry out tasks in the background that are traditionally time-consuming for sales and operations teams.

At the time of launching, Microsoft included the following case study:

Using Copilot, one sales team has achieved 9.4% higher revenue per seller and closed 20% more deals. And thanks to Copilot, one team is resolving customer cases nearly 12% faster.

In the no-code platform, Copilot Studio, you can already create your own customizable agents to automate specific tasks. 

The latest Copilot agents, for Teams and SharePoint, are different, in that they are Microsoft-designed agents ready out of the box. 

Once added to Teams or SharePoint, you have the ability to ask questions around chats you may have missed, retrieve documentation, and benefit from the autonomous work Copilot agents carry out in the background. 

We’ll explore the subtle differences between Teams and SharePoint Copilot agents in the sections that follow.

What is the difference between a Copilot and an agent?

While we’re still in the early phases of Copilots and “regular” AI agents, the distinction between a Copilot agent and a regular agent hasn’t been made all that clear.

It’s also important to flag here that these terms are used interchangeably by different people in different businesses.

In his coverage for The Verge, Tom Warren, summed up how Copilot agents have the potential to become virtual employees.

AI-powered Copilots that can work like virtual employees and perform tasks automatically. Instead of Copilot sitting idle waiting for queries, it will be able to do things like monitor email inboxes and automate a series of tasks or data entry that employees normally have to do manually.

Microsoft defines a Copilot as:

“A supportive tool that works with a user to complete a task by providing suggestions and recommendations. Copilots are not fully automated and rely on user interaction. For example, Microsoft 365 Copilot can help with creating, understanding, and editing documents in Word.”

Whereas a Copilot agent is defined as:

“An artificial intelligence that can autonomously execute tasks and improve processes based on predefined parameters and learning from its environment. Agents can take on delegated tasks and make decisions that align with the user's objectives. For example, an agent can send an email on command.”

What's the use case for Copilot agents?

What can Copilot agents actually do? That’s the burning question.

While Copilots are now Microsoft-wide, it’s helpful to picture them in specific apps. We’re going to focus on the latest (and potentially most-used) scenarios.

1. Copilot agents in SharePoint

Designed to extract answers and insights from your company’s content repository, SharePoint agents provide immediate answers after scanning your uploaded documents.

Each SharePoint site comes with a built-in agent specific to that site. So there’s no concern over who has access to data and information.

Furthermore, you can create your own agent for specific files and folders within that SharePoint site. Now, you can personalize and share the agent or the output created by that agent to specific users.

As of January 2025, you can chat with agents in SharePoint about files in the following file types:

• Office documents: DOC, DOCX, PPT, PPTX, XLSX
• New Microsoft 365 formats: FLUID, LOOP
• Universal formats: PDF, TXT, RTF
• Web files: ASPX, HTM, HTML
• OpenDocument formats: ODT, ODP

Once this agent has been created, the typical next step is to share that agent with your colleagues in Teams. Upon doing so, the agent becomes embedded in your chat for other users to ask questions.

From here, you can use commands like, Create a pitch about our latest product, [product name] for me to present to [client name].

For larger projects, you can even add external sources (those outside your SharePoint sites and your business). Then, you can customize permissions and access in Copilot Studio. Here’s where the autonomous aspect of a Copilot agent, rather than a simple agent, comes into play.

By creating custom workflows and automations in Copilot Studio, you open the door to back-end activities happening on auto-pilot.

2. Copilot agents in Teams

From February 2025, you’ll be able to add Copilot agents to Teams group chats.

The three primary functions of Copilot agents in Teams are:

• Facilitator: Takes real-time notes in Teams meetings and chats
• Translator: Translates real-time conversations in meetings and chats.
• Project manager: Automates plan creation and completes tasks in Microsoft Planner.
  
  

Microsoft has also stated that, “Copilot in Teams can understand, recap, and answer questions based on visual content shared on screen—from PowerPoint to the web—in addition to the transcript and chat. You can also ask Copilot for a quick summary of a file shared in a chat to get all its main points without having to open the file.”

3. Copilot agents for Contact Centers

With all the chatter about agents, it’s important to flag the most obvious Contact Center agent use case.

With Copilot, you can use generative answers in your voice-enabled agents. When you create a voice-enabled agent, it automatically supports the addition of knowledge for generative answers.

This means you’re not just implementing a self-service IVR, but, using natural language processing, customers can have entire conversations with a voice-enabled AI agent.

It’s important to flag that this AI and machine learning is still somewhat immature technology. We advise caution when implementing voice-enable agents when it comes to the extent of self-service you allow.

For example, basic queries like opening times, checking account balances, etc. are absolutely fine. These are data dips and system referencing. Even more complex inquiries like working through troubleshooting are possible. However, when conversations get heated or when technical escalation is needed, it’s best to program your AI agent to seek help from a human.

How do Copilot agents work?

On the back end, Microsoft’s own AI, large language models (LLMs), and machine learning technology is powering three core components that make up the DNA of a Copilot agent:

• Personas: Determine the manner in which agents respond to users or perform actions.
• Plugins: Let agents retrieve information from the user or other systems. You can use pre-built plugins and your own custom plugins.
• Planners: Allow agents to plan how to use available plugins.

Behind the scenes, what’s happening here is fivefold:

• Input processing: When you ask a question or make a statement, input text is processed and analyzed to understand your intent and extract relevant information.
• Model invocation: This processed input is fed into a language model that has been trained on vast amounts of diverse text data. The model uses this training to predict and generate a coherent, contextually appropriate response.
• Response generation: The language model generates a response by evaluating various factors such as context, relevance, and appropriateness. This involves complex algorithms and neural networks that work together to produce a natural-sounding reply.
• Output delivery: The generated response is formatted and delivered back to you. This process includes ensuring that the response adheres to conversational norms and is engaging and informative.
• Continuous learning: Copilot agents continuously learn and improve from interactions. Feedback and new data are incorporated to enhance the model’s accuracy and usability.

For autonomous activities, where human input isn’t required to trigger a task, the model looks more like this:

• Autonomous functionality: Copilot agents have predefined objectives and are capable of making decisions and initiating actions on their own. They can monitor events, processes, or data streams and act based on predefined rules or learned behavior.
• Data collection and monitoring: Autonomous agents continuously collect data from their environment or specific sources. They monitor this data to identify patterns, anomalies, or trigger events that require action.
• Decision-making algorithms: The collected data is processed using advanced decision-making algorithms. These algorithms can include machine learning models, rule-based systems, or both. The Copilot agent evaluates various factors and potential outcomes to decide on the best course of action.
• Action execution: Once a decision is made, the autonomous agent executes the necessary actions. This could involve sending notifications, updating databases, controlling devices, or interacting with other systems.
• Feedback loop: Autonomous agents often have a feedback loop to assess the outcomes of their actions. They learn from these outcomes to improve their decision-making process over time, adapting to new situations and optimizing their performance.

How to access Copilot agents

There are two prerequisites to access Copilot agents:

• You must be a Microsoft 365 Copilot license holder; or
• Non-Microsoft 365 Copilot license holders must have Copilot Studio capacity packs or Pay-as-you-go meters enabled.

In the Copilot Studio, you can create your own agents. Alternatively, you can use the pre-built agents in SharePoint, Teams, etc. All you need to do is click the Create a Copilot agent in the relevant app.

Copilot agent pricing

You can access Copilot agents with a Microsoft 365 Copilot license.

As of January 2025, Microsoft 365 Copilot for businesses and enterprises costs $30 per user per month with an annual subscription. There is also a personal plan costing $20 per user per month with no annual commitment.

Thinking about introducing Copilot agents?

Like introducing any technology, there must be cautions and considerable planning. With AI introductions, we —as an industry—advise a more white-gloved, step-by-step approach.

When planning your Copilot introduction, we advocate a four-pronged approach:

• Recognize the limitations: Understand the limitations of what you can and can’t expect Copilot agents to do. Communicate this across all users to avoid unrealistic expectations.
• Gradual scaling: Introduce Copilot agents to power users and technology enthusiasts first. Create an environment of encouraged feedback and testing before full rollout.
• Training and development: Use online material and hold Q&A sessions to empower self-service and continuous development. Going it alone is never the best option.
• Data security: Review plugins for Microsoft Copilot, check privacy policies and terms to protect organizational data effectively.

Need help with your Copilot implementation?

Book a free call with our Microsoft experts here.