Introducing Copilot Agents

Many of our readers work at an organization that can be counted as a Microsoft shop, i.e., the organization primarily uses Microsoft software for its operations. If you’re in that bucket, there’s a decent chance you have access to Microsoft 365 Copilot, the paid version of Microsoft’s Copilot integration into the Microsoft suite of productivity tools — incorporating the power of large language models into tools such as Word, Excel, Powerpoint, Outlook, Teams, Sharepoint, and OneNote.

Unlike the base version of Microsoft Copilot, which is called Copilot Chat, Microsoft 365 Copilot is able to draw upon what is referred the Microsoft Graph, which includes information about the relationships and activities over your organization’s data. So instead of merely drawing on web-based content in response to user prompts, the LLMs that power Copilot can also draw upon organizational data as broader context for its responses, which in turn can help provide actionable insights and to build custom tools that integrate with emails, files, and calendars.

How exactly can one build custom tools with Copilot? That leads us to the subject of today’s post, Copilot Agents.

What is a Copilot Agent?

In Microsoft’s own words: “Agents extend what you can do with Microsoft 365 Copilot—you can customize your Copilot experience by connecting to your organization’s knowledge and data sources, as well as by automating and executing business processes. These AI-driven agents can perform a variety of tasks, working alongside you to offer suggestions, automate repetitive tasks, and provide insights to help you make informed decisions.”

If you’re familiar with ChatGPT’s custom GPTs, Copilot agents are just the Copilot analogue of GPTs, except that agents have the additional feature of being able to tap into the Microsoft Graph to inform their responses.

Copilot agents come in several varieties: agents prebuilt by Microsoft, agents built by users (and various outside entities), ready-made Sharepoint site agents, and advanced agents built using developer tools.

Today we’ll focus on prebuilt agents, while next week we’ll look out how to build agents of our own.

Prebuilt agents

Prebuilt agents can be found on Copilot’s Agent Store, which can be accessed in Copilot by clicking “All Agents” (for me, when I access Copilot through my Teams app, I can find “All Agents” on the lefthand side bar). Let’s highlight a few of Microsoft’s prebuilt agents that you might experiment with.

Researcher

Researcher is the Copilot version of the Deep Research tools that can be found on other LLM platforms like ChatGPT and Google Gemini. The idea is that Researcher will draw together a number of sources to build a custom research report on any topic of your choice.

As an example, we asked Researcher to give us a report on the status of three-year undergraduate degree programs. Initially, Researcher found an internal file in one of our OneDrive folders that described work performed by a Drake committee on the status of three-year degree programs and incorporated this material into its report. We followed up with a request for a report that didn’t draw upon the Microsoft Graph, resulting in a report on three-year degrees with only web-based references.

Analyst

Analyst is a Copilot agent that you can use to draw insights from your data. If you attach a CSV file to your prompt, you can ask Analyst to identify key trends within the data, to perform specific calculations with the data, or even to create visualizations of the data. The power of Analyst lies in its dependence on an underlying reasoning model, which in turn writes code in Python to carry out various analyses.

Learning Coach

Learning Coach is the Copilot analogue of ChatGPT’s Study and Learn mode. If you want to have a Socratic dialogue with an LLM that will guide you, quiz you, and playfully engage you to learn about a topic of your choosing, you might try Learning Coach.

Surveys

Surveys is an agent that not only creates surveys, but, in Microsoft’s words, “streamlines the entire survey lifecycle—from creation and distribution to analysis—with minimal manual effort, powered by AI and integrated directly into Microsoft 365 Copilot.” What’s particularly valuable about Surveys is that it allows users to create their survey with a side-by-side preview window, so that changes to the survey can be made in real-time by either conversing with the agent or by directly manipulating the survey.

Other Prebuilt Agents

You can also find prebuilt agents made by other companies in the Agent Store. Options include Dropbox, Asana, Miro, Trello, Mural, and ArcGIS.

Some of these agents require an account with an outside tool. For instance, the Dropbox agent allows Copilot to access a user’s Dropbox folders, which increases the pool of user data that Copilot can use to inform its responses to user prompts. As another example, the Asana agent allows users to embed Asana projects into Teams channels.

What’s Next?

There are tons of Copilot agents to experiment with. If you have a Microsoft 365 Copilot subscription, you might try exploring some of the agents we’ve discussed here or look at some of the other available agents.

What if you have a task that you’d like an agent perform but there is no agent to carry out that task? The answer is that you can build your own custom agent to do the trick. We’ll discuss how to do that in next Friday’s edition of the newsletter.