Microsoft 365 Copilot extensibility is the next frontier of enterprise AI at Microsoft. By managing it effectively, we’re giving our employees the power to revolutionize how they access data and accomplish tasks.
But how are we implementing this new framework?
Thanks to our team at Microsoft Digital, the company’s IT organization, we’re extending Copilot’s value by taking better advantage of our Microsoft 365 Graph in more connected and efficient ways. As a result, Copilot extensions can connect with widely dispersed organizational information to help our employees find answers more effectively, work more efficiently, and think more creatively.
Taking Copilot to this next level is enabling us to extend its reach into more specific business scenarios. Those include quickly spinning up specialized agents that range from large, companywide experiences—like an agent that’s streamlining how our employees interact with HR and IT—to small, citizen developer-built solutions that solve specific tasks for individual employees or small teams. Using Copilot Studio, organizations from across the company can build agents that provide tailored AI assistance to unlock new experiences for our employees, optimize our core processes, and provide deeper business insights.
What is Microsoft 365 Copilot extensibility?
Microsoft 365 Copilot extensibility enables users to enhance and customize Copilot’s capabilities by integrating external data sources, creating plugins, and developing agents. This framework helps organizations tailor Copilot to their specific needs, providing a more personalized and efficient user experience.
Agents add specialized skills and knowledge to the Copilot experience while providing the option to automate specific tasks. They work alongside or on behalf of users, teams, or organizations to handle both simple, mundane tasks and more complex, multi-step business processes.
Types of agents
- Retrieval agents surface information from grounding data, summarize and reason over information, and answer user questions.
- Task agents take action when asked, automate workflows, and tackle repetitive tasks for users.
- Autonomous agents, currently in private preview, operate independently, plan dynamically, orchestrate other agents, learn, and escalate tasks to humans when necessary.
Agent builder
The Copilot Studio Agent Builder provides a simple interface that users can access to quickly and easily build retrieval agents, either manually or by using natural language prompts.
Copilot Studio
Copilot Studio is a graphical, low-code tool designed to help users create and customize task and autonomous agents within the Microsoft 365 ecosystem. This platform allows users to build automation workflows, integrate enterprise data, and extend the capabilities of Microsoft 365 Copilot to meet specific business needs through plugins, graph connectors, and other components.
Opening new horizons for intelligent assistance
Out of the box, Microsoft 365 Copilot provides powerful access to company data across our Microsoft tenant, discoverable through the Microsoft Graph. But not all information lives in the Microsoft Graph, and our employees do their work using more than just one set of tools.
Many companies have data spread across an expansive digital landscape. Within Microsoft, our employees access a vast quantity of content within Microsoft 365 apps and in other data and systems. That breadth can create issues with discoverability and block our employees from taking action effectively. On top of that, some processes benefit from narrowing search parameters instead of widening them.
Microsoft 365 Copilot extensibility is the framework that makes closing those gaps possible. It enables users and developers to customize Copilot’s behavior with additional instructions, grounding, and actions while providing a clear and discoverable entry point in the user interface.
“Organizations have mountains and mountains of data available, with an ever-increasing number of tools and experiences,” says Matthew Marzynski, product manager with Microsoft Digital. “The problem is that our needs as employees don’t conform neatly to the different tools that we use to fulfill them. Copilot extensions are a way to collapse that complexity.”
Through Copilot extensibility, employees no longer need encyclopedic knowledge of each and every app, tool, or repository that pertains to their work, enabling them to avoid time-consuming manual exploration. Instead, users and organizations can configure Copilot extensions to intelligently surface the information they need within a single pane. And at the center of this shift is a new kind of interface: the agent.
Agents enable users and developers to customize Copilot’s behavior with additional instructions, grounding, and actions, and they provide a clear and discoverable entry point through a single, accessible user interface. As a result, it’s now possible to extend the power of Copilot beyond our first-party product portfolio to deliver highly transformative and personalized employee experiences.
Users have enormous flexibility in terms of the kinds of agents they can create, their level of complexity, and how they create them. Creation approaches include Microsoft Copilot Studio Agent Builder, Copilot Studio itself, or working with both Copilot Studio and Azure AI.
For the simplest agents, creators can access Microsoft Copilot Studio Agent Builder right in Teams through Microsoft 365 Copilot Chat. They can use it to walk through a streamlined process driven by natural language. Simple prompting and configuration panes make this interface accessible to anyone who wants to create agents.
Accessing Copilot Studio directly provides more power and flexibility. Based on Microsoft Power Platform, this tool provides a low-code or pro-code environment for creating and enhancing agents with custom experiences.
By focusing Copilot intelligence on specific repositories and apps, agents created in Copilot Studio provide greater discoverability outside the Microsoft 365 data estate while unlocking custom workflows. Most importantly, they make it possible to explore different information silos and take different kinds of actions from one interface, accessed through Copilot.
“We’re looking at every workflow, process, and interaction to find ways of applying a Copilot-first perspective,” says Nitul Pancholi, a lead on in the Microsoft Digital AI Center of Excellence. “It’s an opportunity to redefine how we do work and drive greater impact than ever before.”
Whether an employee creates a personal agent tailored to their role or a line of business builds an enterprise tool to support their team’s work, Copilot Studio brings their vision to life. It lets users create agents using two building blocks, connectors and plugins, plus the ability to customize those elements.
Kinds of Copilot agents
“Through extensibility, we’re enabling anyone to create a Copilot-powered solution tailored to their team, suited for their organization, and focused on the right content,” says Dodd Willingham, program manager for internal deployment of search and chat in Microsoft Digital. “Graph connectors and plugins make the right content easily identifiable for their agents so they can target the data they need appropriately.”
Our Copilot extensibility journey
Microsoft is actively building and deploying agents. The Microsoft Digital team has been instrumental in creating and implementing these early scenarios using Copilot Studio through Microsoft Elite Builders, our internal program that encourages Microsoft employees to build and share their custom agents with fellow employees.
Simple retrieval agents are enabling much of this early work. This kind of extension powers scoped, task-specific experiences to accomplish specialized tasks in Copilot by accessing Microsoft Graph data alongside additional semantic ground.
The IDEAS Copilot is one example. It democratizes access to our IDEAS knowledge base’s insights on app usage to empower informed decision-making. Through natural language queries, it enables users to take action on crucial usage data without technical expertise.
“The copilot extensibility concept is really about creating agents that are low-maintenance and high-value for users,” says Aisha Hasan, Power Platform and Copilot Studio product manager for Microsoft Digital. “Anyone on the tenant can create their own agents, while engineering teams can build enterprise-level solutions using Copilot Studio to solve business problems and boost productivity on a wider scale.”
Our most ambitious project so far has been an agent that unlocks scoped, role-specific Copilot experiences we can embed in public and private apps. The Employee Self-Service Agent in Microsoft 365 Copilot grants employees access to HR and IT information and tools through their choice of interface—Copilot or our company sites. It’s now available to customers in private preview.
The agent connects to SharePoint, Microsoft Teams, Microsoft Outlook, and other resources that comprise the Microsoft Graph to retrieve the right information. Depending on whether the employee’s query is related to HR or IT, the agent integrates their personalized data from a variety of knowledge bases and third-party apps. From there, it provides the employee with an answer that creates a single, reliable starting point for resolving their query.
Employees are finding HR and IT answers more accurately and faster, without the need for extensive searches across wide-ranging toolsets. In our initial pilot, people who use the Employee Self-Service Agent for HR receive 42% greater accuracy in answering their questions. On the IT side, the overall self-help success rate increased by 36%. But the greatest benefit is the way this agent keeps employee productivity flowing by allowing users to seek out crucial HR and IT information in a single pane within their flow of work.
Learning from early extensibility projects
As one of the first enterprises to explore Microsoft 365 Copilot extensibility, developing and implementing these solutions comes with its own challenges and learnings.
“As Customer Zero, we need to balance product innovation with security and operational needs,” says Mykhailo Sydorchuk, Customer Zero lead for Microsoft 365 integrated apps at Microsoft Digital. “We’re balancing the priorities of driving productivity for the business, supporting the product teams in deploying new features, and also maintaining the guardrails to protect employees and our organization.”
Within Microsoft Digital, we have two chief advantages as we address these IT challenges:
- A trustworthy technology ecosystem operating at an incredible scope and scale across a huge product suite, ripe with the kinds of data that empower AI tools.
- A mature IT organization that has decades of experience in adopting and operationalizing new technologies and mitigating risks.
One of the most important lessons we’ve learned has been about adapting our existing governance structures to the framework of Copilot extensibility.
Agents rely on connections with existing tools that have well-established parameters for reasoning over data and governing information. Because Copilot respects Microsoft 365 governance and data loss prevention protocols, it honors all the access controls, security policies, and personally identifiable information (PII) data-handling structures that an organization puts in place across its tenant. As a result, businesses can rely on Microsoft 365’s robust foundation of security even as they forge ahead into new AI capabilities.
But this is a new approach to technology, so we’ve adapted our review process for new agents alongside our implementation to ensure governance and security keep pace with innovation. Those reviews largely revolve around key questions we ask about all our technology, with an added layer around Responsible AI, where we ask ourselves questions like these:
- Security: How does data move from one app to another across the Microsoft 365 tenant boundary, and how does an agent interact with that data?
- Privacy: Where does the agent handle and store PII?
- Accessibility: Does the UI make this technology equally available to all users?
- Responsible AI: Does the agent meet our standards for fairness, reliability and safety, inclusiveness, transparency, and accountability?
From a security and governance perspective, it can be daunting to implement internally built extensions that have access to your organizational data and incorporate them into your business workflows. By asking the same questions we’ve embedded into our review framework and relying on the robust security and governance features of Microsoft 365, you can ensure you maintain control over your organizational data.
It’s also important to realize that adopting this new framework will take time. There’s a natural progression from simpler extensions to more complex tooling. It’s all part of accelerating along an adoption maturity curve with the next iteration of AI tools.
The technical aspects of enabling Copilot extensibility require forethought, but they don’t exist in a vacuum. If your team maintains robust governance policies through your Microsoft 365 tenant and they’re experienced with Power Platform, much of the legwork is already done.
But if your organization is new to this space, our internal experience governing Copilot can act as a guide for keeping your data secure. We’ve also created resources tailored to helping first-time users get up to speed with Copilot Studio. We’ve also made the Employee Self-Service Agent in Microsoft 365 Copilot available to customers in private preview to provide a straightforward on-ramp to extensibility.
It starts with considering where Copilot extensions might fit into your workflows and what data it needs to access, starting with simple implementations, then deliberately building from there.
“A cornerstone of Copilot extensibility is understanding your data and the scenarios you think will be most impactful for optimizing processes,” says Amy Rosenkranz, principal product manager for Customer Zero Extensibility in Microsoft Digital. “Then ensure your endpoints are ready to hook into natural language processing and build your instructions.”
A principled approach to employee usage and adoption is also important. Eagerness among IT professionals and early adopters won’t necessarily drive business-wide transformation. As a result, we’re actively working to inform employees about the value of these tools and provide skilling opportunities.
“One of the main things we observed is that there’s a certain level of change management involved,” Pancholi says. “In order to build these habits, we’ve focused on creating simple workflows with tangible impacts so employees can see the value and start building Copilot-first habits.”
Finally, we’ve discovered that keeping agents lean in scope helps them function more easily. It’s important to think about back-end processes as you create extensions. For example, including too many data sources can become a serious tax on processing power. From both a performance and scenario standpoint, it’s a better strategy to keep agents narrowly focused.
Next steps into the era of extensibility
As we continue our Microsoft 365 Copilot extensibility journey, we’re looking to our internal experience to guide the product for our customers, especially in IT. Our team is currently exploring ways to further unify the ecosystem until administration and management of extensions can all occur through a single layer. That will reduce operational costs and enable successful adoption on a greater scale.
We’re also exploring ways to shift more aspects of governance to Copilot itself. Since many agents access data and third-party apps outside our tenant, those sources might not benefit from Microsoft 365 data protection policies. Allocating more of the burden of governance to agents themselves may help fill that gap.
For now, we’re still exploring what’s possible. And employee uptake tells a strong story about extensibility’s impact. Since the release of our initial Copilot extensions a few months ago, usage of retrieval agents has shot up by 10x. Initial results from our Employee Self-Service Agent in Microsoft 365 Copilot rollout are equally promising.
It’s all coming together to demonstrate the profound value Copilot can provide for businesses.
“The vision around extensibility is that Copilot can be the single place where all tools coalesce into one single pane,” Marzynski says. “It’s a way to make your time much more effective and reduce the cognitive tax of changing channels or swiveling seats, and it’s ultimately a way to make your employee experience more rewarding.”
Here are some tips for getting started with Microsoft 365 Copilot extensibility at your company:
- Be curious: Don’t limit your imagination, because it’s a chance to redefine workflows.
- Be patient but continue driving forward. This is the beginning of a more long-term journey, and it’s important to get more comfortable with the technology.
- Hackathons produce amazing results, and it’s important to give people time to experiment together.
- Have a strong change management program. No matter how wonderful the tech is, people need to see helpful use cases and real value.
- Use your existing investment, processes, governance, and management practices for Microsoft 365 and scale it out.
- Build a well-documented review process.
- Start with only allowing approved apps or enable guardrails for self-service to ensure IT maintains control and security, then grow from there. When it comes time for citizen developers, open the door to retrieval agents.
- Think in terms of data—it’s about what data is you have, if it’s properly governed, where it’s being accessed, and what tenant boundaries it crosses.