Copilot Studio is enhancing its platform with knowledge improvements, Azure AI integration, and more

Microsoft Copilot Studio is the place to build agents. Here, organizations can build the agents they need to augment their workforce and execute business processes. Microsoft is investing in agent capabilities that help customers boost the quality of their agents to unlock business value.

Here is a summary of the key announcements for Microsoft Ignite 2024: 

• Copilot Studio expands knowledge with advanced tuning capabilities and third-party sources. Makers will be able to boost the quality of their agents with the latest generative models, new knowledge sources, and Retrieval Augmented Generation (RAG) enhancements. They will also be able to curate and fine-tune their knowledge—seeing what works well and what doesn’t—so that agents can provide relevant answers to its users. Analytics on knowledge sources will also be available. 
• Use Azure AI capabilities directly in Copilot Studio. Makers will now be able to utilize custom Azure AI Search indexes as a knowledge source for custom RAG scenarios and call any of the more than 1,800 prebuilt or custom-built Azure AI models from within Copilot Studio. 
• Autonomous agents are now in preview. Makers can now build agents that work on their behalf, without having to prompt the agent, saving human hours and increasing efficiency. They can create these agents from scratch or configure agents that are prebuilt in Copilot Studio. 
• New multi-modal capabilities allow agents to interact beyond text. Organizations can now add generative AI to their voice solutions, including to their interactive voice (IVR) system, or deploy agents in applications and other experiences for users to interact with using their voice. Not only will users be able to speak to these agents, they will also be able to upload images and prompt their agent to analyze and answer questions about the image.
• The Microsoft 365 Agents SDK is in preview. Developers can now build enterprise-grade, scalable, multi-channel agents with code using the SDK. These agents can operate across a variety of channels including Microsoft 365 Copilot, Microsoft Teams, web, and more. They can also access Copilot Studio agents, to extend its capabilities, enabling continuity from Copilot Studio to code-first scenarios. The SDK is available in C# and, developers can call AI services of their choice, including Azure AI Foundry, Semantic Kernel, or others. 
• New analytics are available. In addition to analytics on knowledge sources, makers will now also be able to see session data and trends sooner. Key KPIs, such as engagement rate and satisfaction scores will show trends from the previous period.
• Information workers can build agents directly in Microsoft 365. With Copilot Studio’s simplified agent builder experience, accessible directly within Microsoft 365 Copilot, information workers can easily build and share their own agents. 
• Copilot Control System enables IT to confidently adopt and deploy Copilot and agents. The Copilot Control System is a collection of IT capabilities that include data access and security policies, management controls for Copilot and agents, and measurement reports and tools to track adoption and business value. 

Below, we will explore some of the categories in more detail. 

Copilot Studio adds knowledge curation and third-party sources 

Copilot Studio is deepening knowledge and RAG capabilities. We’re lighting up connectors to bring in new knowledge sources. Makers can now enable agents with real-time knowledge from multiple third-party sources, such as Salesforce, ServiceNow, and Zendesk. Copilot Studio only semantically indexes the metadata from these tables, and without data movement, helps answer complex questions from these third-party sources. This is now in preview. 

Beyond access to new knowledge, Copilot Studio is utilizing the latest GPT models and optimizing RAG strategies, to get high-quality answers. Finally, with advanced knowledge tuning, makers can have more control over data sources. Makers will now have visibility over what sources are being used, what types of user questions remain unanswered, and what the accuracy of responses are. They can also see analytics, like usage rates and error reporting for their top five sources, to understand if they should tweak certain sources. Knowledge tuning also suggests additional knowledge that should be added from your organization and provides a guided way to filter and prioritize sources. 

Use custom Azure AI capabilities directly in Copilot Studio 

Copilot Studio and Azure AI Foundry are becoming more integrated to enable more complex, custom scenarios. This lets Copilot Studio customers take advantage of their investments in Microsoft’s broad Azure AI Foundry offerings.

For example, makers often have large corpuses of data in Azure AI that they now can use as a knowledge source directly in Copilot Studio. Makers can already perform search operations on vectorized indexes built in Azure AI Search to access unstructured data. Now, they can bring this data and custom RAG to Copilot Studio for accurate results. By connecting all their data sources, organizations can see that agents are more grounded in their business data and provide specific, high-quality responses. 

That’s not all. More advanced Azure AI capabilities are being surfaced in Copilot Studio. Makers will soon be able to access models from the Azure AI model catalog, which has more than 1,800 models available. They’ll also be able to access and call their own custom fine-tuned models that they have built from inside Copilot Studio. 

Finally, as accessing Azure AI capabilities in Copilot Studio becomes more common, there will be an update to purchasing. Microsoft will introduce an option to buy Copilot Studio messages with an Azure meter for pay-as-you-go consumptive billing to provide customers with flexibility. 

Autonomous agents

Organizations have been creating agents that provide users with information and that take actions when asked. Now, they can take it a step further with autonomous agents that work independently on behalf of a user, team, or organization to automate complex business tasks.

Users can design an agent to detect select events, like an email arriving, respond, and then use generative AI to trigger a chain of actions to orchestrate a series of complex, long-running business processes. For instance, an agent can assess the intent of an email, look up the sender’s details and account, see prior communications, check inventory, respond to the sender, ask for their preferences, and then take the appropriate actions to close a ticket—saving users time as they focus on high-priority tasks. Learn more about the features in-depth.

Autonomous agents enable AI to work for its users. They can run all the time and take on the burden of functional business processes, only notifying users when necessary. Makers can build autonomous agents from scratch, or configure agents that are prebuilt in Copilot Studio, like select functional agents. 

As a reminder, agents come with comprehensive enterprise data protection features for security, along with other guardrails and controls including encryption, data loss prevention, and a responsible AI strategy to address risks like prompt injection. Administrators can also implement strict security measures for data sources, manage access controls, and establish policies to safeguard data and monitor usage, while overseeing the entire lifecycle of agents to ensure compliance and continuous improvement. Learn more about enterprise protection, guardrails, controls, and lifecycle management.

New voice and image features further enhance the Copilot Studio platform 

Copilot Studio has now been enhanced to include voice capabilities. Organizations can embed their agent into a modern interactive voice (IVR) system, transforming a rigid experience into a flexible, generative AI system using all of Copilot Studio’s knowledge and action capabilities. Simply point to your knowledge sources and your IVR can handle a long array of sophisticated questions over the phone, allowing organizations to reduce escalations. These intelligent voice-enabled agents can recognize speech, handle interruptions, detect silence, and offer re-prompts to improve user satisfaction. 

Beyond just IVRs, speech enabled agents can be embedded into applications, standalone kiosks, concierge systems, and more. A hotel app, for example, can use a Copilot Studio agent that has been grounded in the hotel’s menus and the hotel website to act as an in-room concierge to guests. 

In addition to voice features, users can now upload images and ask Copilot questions about their image such as, “Why is my bill higher this month?” Uploading images can also add rich context to a dataset, whether it’s in customer service, sales, or elsewhere. For example, a seller can say, “Add these hand-written notes to the CRM,” saving time, while easily adding data to their system. 

Code-first developers now have access to the Microsoft 365 Agents SDK 

Copilot Studio makers and developers can now expand their agent’s capabilities with code, enabling them to bring together the best of low-code and code-first worlds. The Microsoft 365 Agents SDK enables developers to bring capabilities hosted in Copilot Studio to code-first agents, facilitating seamless low-to-pro code agent development. 

With the Agents SDK, developers can build enterprise-grade, scalable, and multichannel agents. Developers have full flexibility to connect to the AI services of their choice including Azure AI Foundry, Semantic Kernel, Copilot Studio or others. Agents built with the SDK can be deployed across a variety of channels such as Microsoft 365 Copilot, Microsoft Teams, web, and over 10 other third-party messaging platforms. The SDK is now available in preview. 

Improved analytics and insights 

With our new analytics investments, makers will be able to get deep insights into how their generative system is working. In addition to the knowledge analytics metrics mentioned earlier, like source usage and accuracy of responses, there will also be improvements to top level metrics.

These metrics will be improved, more performant, and expose longer term trends. Makers will be able to choose between graphs filtered by specific outcomes, so they can understand KPIs and customer satisfaction. They will also be able to drill down and get more detailed insights. All of this will make it even easier to iterate and improve their agents.  

Users can add and build agents directly within Microsoft 365 Copilot 

Information workers already use Microsoft 365 Copilot to enhance their day-to-day productivity. Now, they can make Copilot more personalized by extending it with specialized agents. Agents reference knowledge given to them via graph connectors or file uploads, like enterprise-specific data in Dynamics 365, SharePoint sites, and line of business systems. Agents can also be given actions like sending emails, updating records, or creating support tickets. Users can either add these agents from a store within Microsoft 365 or build their own in a new, streamlined Copilot Studio experience embedded directly inside Business Chat (BizChat) and SharePoint. Individuals can customize or deploy agents that are prebuilt. Examples of these agents include Employee Self-service Agent, Store Operations Agent, and others. 

The new Copilot Studio agent builder experience enables users to easily create an agent in a few minutes, simply by describing what it should do with natural language and pointing it to the knowledge it needs—all within the Microsoft 365 Copilot experience. Agents can be shared, @ mentioned in Microsoft Teams, or interacted directly with through a chat window in Microsoft 365 Copilot. This functionality is now generally available and is already being used by users to extend Microsoft 365 Copilot to transform their business processes—read the full announcement.

Join us in exploring Microsoft Copilot Studio and build your own agent! 

Get started

Join us at Microsoft Ignite, in person or online. You can: 

• Watch Microsoft Ignite session BRK165: What’s new with Copilot Studio and Agents. 

• Check out features as they come to preview, including autonomous agents, advanced knowledge tuning, image upload, analytics, and the Azure AI Search integration. Agent builder and generative AI in IVR systems are both generally available.