Multi-agent orchestration and more: Copilot Studio announcements

Explore powerful features in Microsoft Copilot Studio announced at Microsoft Build 2025, including multi-agent orchestration and more developer tools.

Microsoft Build 2025 is here—our annual showcase of the most exciting innovations shaping the future of development and AI. For engineers, makers, and subject matter experts, it’s the moment to see what’s next across the Microsoft ecosystem. This year, Microsoft Copilot Studio has a number of powerful new agent-related features to show you. From multi-agent orchestration to more maker controls, computer use in agents to code interpreter, read on for a first look at Copilot Studio’s exciting announcements.

Recap of major Microsoft Build 2025 announcements

Jared Spataro, Corporate Vice President, Microsoft, announced some of the biggest news from the Copilot Studio and Microsoft 365 Copilot agents teams. In case you missed it, here’s an overview of a few features we’re particularly excited about.

Multi-agent orchestration

Rather than relying on a single agent to do everything—or managing disconnected agents in silos—organizations can now build multi-agent systems in Copilot Studio (preview), where agents delegate tasks to one another. This includes those built with the Microsoft 365 agent builder, Microsoft Azure AI Agents Service, and Microsoft Fabric. These agents can now all work together to achieve a shared goal: completing complex, business-critical tasks that span systems, teams, and workflows.

This evolution reflects a broader shift in how organizations are scaling their use of agents across Microsoft. Imagine a Copilot Studio agent pulling sales data from a customer relationship management (CRM) system, handing it off to a Microsoft 365 agent to draft a proposal in Word, and then triggering another to schedule follow-ups in Outlook. Or, envision agents coordinating across IT, communications, and vendor systems to manage an incident from detection to resolution. Whether it’s executive briefings, customer onboarding, or product launches, agents can now operate in sync—bringing greater connectedness, intelligence, and scale to every step. This feature is currently in private preview with a public preview coming soon.

See how different organizations are employing their agent ecosystems and get inspired for how you could connect yours in Microsoft Corporate Vice President Srini Raghavan’s blog post.

Computer use in Copilot Studio agents

Computer use moves us closer to a more connected, intelligent world where agents collaborate seamlessly with people and systems. Agents can now interact with desktop apps and websites like a person would—clicking buttons, navigating menus, typing in fields, and adapting automatically as the interface changes. This opens the door to automating complex, user interface (UI)-based tasks like data entry, invoice processing, and market research, with built-in reasoning and full visibility into every step. Computer use is currently available through the Microsoft 365 Copilot Frontier program for eligible customers with at least 500,000 Copilot Studio messages and an environment in the United States.

Bring your own model and model fine-tuning

Copilot Studio continues to integrate deeply with Azure AI Foundry, and now you can bring your own model for prompts and generative answers. Makers can access more than 11,000 models in Azure AI Foundry, including the latest models available in OpenAI GPT-4.1, Llama, DeepSeek, and custom models, and fine-tune them using enterprise data. This fine-tuning helps agents generate even more domain-specific, high-value responses.

Model Context Protocol

Now generally available, Model Context Protocol (MCP) makes it easier to connect Copilot Studio to your enterprise knowledge systems. With growing connector support, better tool rendering, evolving scalability, and faster troubleshooting, it’s never been simpler to bring external knowledge into agent conversations.

Developer tools to build agents your way

Microsoft empowers developers to build agents with the tools they prefer—Copilot Studio, GitHub, Visual Studio, and more. With Microsoft 365 Copilot APIs, developers can securely access Microsoft 365 data and capabilities to create custom agents or embed Microsoft 365 Copilot Chat into apps, all while respecting organization-wide permissions.

The Microsoft 365 Agents Toolkit and Software Development Kit (SDK) make it easier to build, test, and evolve agents over time. Developers can swap models or orchestrators without starting from scratch, use SDK templates to jumpstart projects, and deploy to Azure with smart defaults—all now generally available.

Copilot Studio enhancements

New agent publishing channels: SharePoint and WhatsApp

Copilot Studio has exciting updates to available channels, including that publishing agents to Copilot is now generally available. In addition to this highly anticipated update, we’re also adding two additional channels: SharePoint and WhatsApp. These key channels make it easier than ever to bring custom agents to the places where your users already work and communicate. This helps you extend the reach and value of your agents, from serving your teams inside Copilot and SharePoint to engaging customers around the world.

The SharePoint channel, now generally available, lets makers deploy custom agents directly to a SharePoint site with a single click. With authentication and permissions handled automatically, anyone with access to the site can immediately start using the agent. This extends the full capabilities of custom agents into one of the most widely used collaboration hubs in the world.

Starting in early July 2025, makers will also be able to publish Copilot Studio to WhatsApp. This will allow organizations to provide conversational support and engage global users directly within the familiar, mobile-first platform—no separate website or app required.

Additional maker controls for knowledge

Now in public preview, new controls in the Generative AI agent settings give makers more ways to shape how agents respond, reason, and interact with users. In addition to toggles for generative orchestration and deep reasoning, you’ll see multiple categories to further ground and tune your agents.

First, in response to maker feedback, we’re pleased to announce that you can now upload multiple related files into a file collection and use the collection as a single knowledge source for an agent. You can also include natural language instructions to help your agent find the most relevant document in the collection to ground each response.

In the Responses section of the Generative AI tab, you can now choose your agent’s primary response model, provide response instructions, adjust response length, and turn on advanced options like code interpreter (see below) and Tenant graph grounding with semantic search.

Moderation settings control how flagged responses—that is, generated responses detected to possibly have harmful content—are handled. These controls allow you to set a custom response to send when potential responses get flagged. A User feedback section, meanwhile, gives you the option to allow users to provide feedback on the agent, along with a custom disclaimer. This provides you a qualitative assessment of how users perceive your feature—enormously valuable in planning and honing your agent roadmap.

In the Knowledge and User input sections, you decide what the agent draws from: your specific knowledge sources only, foundational model knowledge, or even Bing web search. You can also choose to let users add images during agent interactions.

These new controls, along with improved analytics and testing tools, empower makers to deliver more relevant, responsible, and high-quality agents—leading to greater impact and better user experiences across the board.

Additional knowledge sources for agents

Copilot Studio now supports several new knowledge sources, providing an even broader range of enterprise content for custom agents to access and reference. OneDrive files and folders, SharePoint lists, and Microsoft Teams chats and channels are all supported. Outside this internal data, agents can now employ unstructured data from platforms like Salesforce, ServiceNow, and Zendesk, as well as structured data from platforms like Snowflake, Databricks, and SAP.

Knowledge from Microsoft Azure AI Search is also generally available. Agents can now reference this data more effectively, whether the goal is analyzing sales pipeline trends, identifying churn risks, or supporting role-specific insights across business systems.

As you connect these new sources, note that Microsoft Graph connectors are now called Copilot connectors, a name that better reflects their role in powering the Copilot experience. More than 65 Copilot connectors are already generally available or in public preview in the Microsoft 365 admin center, including Gong, PagerDuty, and Unily. Read more about building Copilot connectors on Microsoft Learn.

Code interpreter

Bring Python-powered logic to your Copilot Studio agents using natural language. Now in preview, code interpreter lets developers extend Copilot Studio agents with Python—enabling advanced logic, dynamic visualizations, and structured data operations directly within runtime or prompt workflows.

There are two ways to use code interpreter. The first, with agents, is dynamic: Python code is generated and executed live at runtime. Agents can analyze uploaded CSV or Excel files, generate pie, line, or bar charts with downloadable outputs, and solve complex math problems in context.

The second, with Prompt Builder, is more static—makers define and edit Python code at design time. When the prompt runs, it executes the same preconfigured logic—ideal for repeatable tasks like Create, Read, Update, and Delete operations on Dataverse tables.

This feature unlocks new abilities in data analysis, reporting, and visualization within agents. Users can get richer and more insightful agent responses, automate calculations, and streamline repetitive processes. By bridging natural language prompts with real code execution, code interpreter helps teams improve quality, reduce turnaround time, and support innovation—while staying inside Copilot Studio’s managed environment.

Pro-developer enhancements

Visual Studio Code extension for Copilot Studio

The Copilot Studio Visual Studio Code extension is designed with professional developers in mind, bringing familiar tooling and workflows to the world of agent development. Now available in the Visual Studio Marketplace, this extension lets you connect directly to Copilot Studio and edit agents from within Visual Studio Code, enabling a modern experience outside the web UI.

For developers and teams, this means you now have new get IntelliSense color-formatting, “find all references,” and the structured clarity of working over a file system—all while building on a fully managed software as a service (SaaS) platform. Copilot Studio handles the infrastructure, compliance, and runtime complexity, so your team can stay focused on delivering value.

The Copilot Studio Visual Studio Code extension is about meeting developers where they are. Whether you’re integrating agents into your broader developer ecosystem or scaling collaboration with GitHub, the Copilot Studio Visual Studio Code extension brings the control and confidence developers expect. No new tools to learn, and no trade-offs.

Embedded agent builder enhancements

Share agents instantly across your organization

The new agent sharing experience in Copilot Chat makes it easier than ever to share agents across your organization. Makers and developers worldwide can now generate and distribute shareable links directly from the right sidebar in Copilot Chat, which streamlines collaboration and increases the discovery of your agents. Once you’ve shared your agent, others can then start using it instantly.

This feature eliminates the extra steps previously required to find and share an agent—no switching tools or hunting for links. Agent sharing reduces friction, speeds up internal adoption, and helps teams get more value from the agents already in use. Easier sharing allows for consistent, cross-platform experiences, whether you’re building with Copilot Studio, the embedded agent builder in Microsoft 365, or Teams Toolkit.

In-conversation agent recommendations

As makers and developers build more specialized agents, it becomes harder for users to know which one to use and when. With in-conversation agent recommendations, Copilot bridges that gap by dynamically suggesting the most relevant agent based on what the user is trying to do.

As the user describes their need in natural language, Copilot determines if an installed agent is better suited to handle a request. It recommends a handoff and carries over the full conversation context so the agent can respond intelligently, access enterprise data, or take action in downstream systems. From the user’s perspective, this process is seamless, effortless, and improves the quality of their experience using AI tools to assist with or accomplish their work.

For developers and makers, this means your agents no longer sit idle, waiting to be discovered. They’re surfaced at the right time, in the right context, driving impact without the user needing to go search for them. It’s a lightweight way to boost discoverability, increase usage, and demonstrate value without changing how agents are built.

Security and governance enhancements

Copilot Studio continues to enhance and improve governance and security capabilities, giving IT teams and security admins control and clarity over every stage of agent creation and operation. Here are some recent updates in agent and platform-level controls as well as data and infrastructure protections.

Agent and platform-level controls

• Privacy controls to disable transcript recording and session downloads, apply real-time sensitive data masking and audio suppression in agent chats, and present a custom privacy notice.
• Automatically assigned identities for agents created through Copilot Studio, automatically listed in the centralized AI agents directory in the Microsoft Entra admin center.
• A separate shared environment for agents built in the Microsoft 365 agent builder, simplifying tracking and paving the way for more granular policies.
• Additional safeguards to prevent unauthorized use of Copilot Maker authentication, enforce Microsoft Entra ID (in public preview) for all interactions, and require explicit consent before agents are shared.
• Blocked role escalation in personal development environments.

Data and infrastructure protections

• Ability for admins to apply Data Loss Prevention to agents in Copilot (to prevent misuse of labeled content) and enforce auto-labeling with downstream inheritance in Dataverse.
• New Data Security Posture Management (DSPM) for AI-focused views, allowing security teams to track how sensitive data moves through apps and agents (including interactions from employees, partners, and anonymous users).
• Automatic connector management enforcement that aligns Copilot Studio with the rest of Microsoft Power Platform—no PowerShell scripts required.
• Network isolation (public preview) to help lock down connectors and App Insights through firewalls and virtual networks (VNETs).
• Federated Identity Credentials to remove stored secrets or certificates from registration.

Together, these enhancements help keep Copilot Studio secure, compliant, and enterprise-ready without slowing down innovation. Learn more about these announcements in Microsoft Corporate Vice President Vasu Jakkal’s blog post.

To dive deeper into these new capabilities, join us for some of our relevant sessions at Microsoft Build 2025: