Microsoft Releases Copilot SDK 2.0 to Help Enterprises Build Scalable AI Agents

Microsoft has released Copilot SDK 2.0, introducing a redesigned architecture that gives developers greater control over how AI agents operate within enterprise applications. The update focuses on multi-user environments, session management, security permissions and long-running AI workflows.

Announced on November 17, 2026, the release introduces a rebuilt runtime based on Rust, allowing developers to host multiple agent sessions within a single application process. The development is aimed at organisations building AI-powered software, internal platforms and customer-facing services.

A New Runtime for Multiple AI Agent Sessions

One of the major changes in Copilot SDK 2.0 is its ability to support multiple independent agent sessions within a shared runtime.

Previously, developers faced additional complexity when managing separate agent environments for different users. The updated architecture separates the agent's model execution loop from its shell, files and user identity.

This allows applications to manage request routing and authorisation while the SDK handles session-specific state, authentication and tools.

The approach is particularly relevant for SaaS platforms and enterprise applications that need to support multiple customers and users simultaneously.

Greater Control Over Permissions and Security

The new architecture gives developers more control over which tools and resources an AI agent can access.

Shared services can start with an empty environment, without automatically inheriting access to host files, shell commands or default tools. Developers can then explicitly enable approved tools, workspace paths, skills and Model Context Protocol (MCP) servers.

Credentials can also be assigned separately to individual sessions rather than allowing every agent to inherit the identity of the application server.

This is important for enterprise environments where customer information and permissions must remain isolated.

However, developers remain responsible for enforcing authorisation and ensuring that their tools respect tenant boundaries.

Persistent Sessions Support Long-Running Workflows

Another important improvement is durable session management.

With this capability, conversations can continue beyond the lifetime of an individual application connection. Developers can assign session IDs, store conversation information and resume existing sessions through another client instance.

This can help applications maintain context when users disconnect, restart an application or return to an ongoing task.

The SDK also allows developers to control session storage through a custom storage provider, giving enterprises greater flexibility over how conversation data is managed.

Flexible Sandbox Architecture for AI Agents

Copilot SDK 2.0 introduces different ways to manage agent execution environments.

Developers can operate shared sessions, move selected sessions into sandbox environments or use coordinating agents to delegate work to isolated workers.

This flexibility allows applications to balance infrastructure efficiency with execution isolation.

Microsoft continues to recommend container isolation for workloads involving potentially untrusted scripts, code compilation and other operations that require stronger security boundaries.

Performance Improvements and Developer Support

Microsoft reported a performance improvement in a specific warmed workload involving client creation, session creation and completion of an agent turn.

The reported execution time fell from 5.25 seconds using the TypeScript CLI to 55.3 milliseconds with the Rust core running in-process.

These figures reflect Microsoft's particular test scenario and should not be interpreted as a universal performance improvement across all applications.

The SDK also supports .NET developers through APIs such as CreateSessionAsync and ResumeSessionAsync. The wider SDK ecosystem includes support for TypeScript, Python, Go, Java and Rust.

What This Means for Enterprise AI Development

The release reflects a broader shift towards AI agents becoming integrated components of business applications rather than standalone assistants.

Businesses are increasingly exploring agents for customer support, workflow automation, internal operations and software development.

For these applications, managing multiple users, protecting customer data, maintaining conversation history and controlling access to business tools are essential requirements.

Microsoft's updated SDK provides developers with more flexibility to address these challenges while building AI-powered enterprise services.

Conclusion

Microsoft Copilot SDK 2.0 introduces a redesigned foundation for building and deploying enterprise AI agents. Its Rust-based runtime, multi-session architecture, durable conversations and flexible sandboxing capabilities give developers greater control over how agents operate.

As organisations move towards more autonomous and connected AI workflows, infrastructure that supports security, scalability and persistent interactions will become increasingly important.