Microsoft introduces Agent Framework to simplify multi-agent systems

Microsoft has announced Microsoft Agent Framework, an open-source SDK and runtime in public preview that unifies AutoGen, a project originating from Microsoft Research, with Semantic Kernel into an enterprise-grade framework for building multi-agent systems. The tool addresses the fragmentation of existing frameworks and the difficulty of taking agent prototypes into production environments.

The framework allows local experimentation and subsequent deployment to Azure AI Foundry with integrated observability, durability, and regulatory compliance. Its technical capabilities include integration of any API through OpenAPI, collaboration across different environments via the Agent2Agent protocol, and dynamic tool connection using Model Context Protocol. The tool incorporates advanced patterns like Magentic One and enables orchestrating agents into structured workflows.

Microsoft has also implemented multi-agent workflows in Foundry Agent Service, currently in private preview. This functionality extends the framework's capabilities in the cloud through a layer with persistent state that coordinates multiple agents in long-running tasks. Workflows can be created visually through VS Code or Azure AI Foundry and include error handling and automatic recovery.

The framework is open source and Microsoft invites developers to contribute. Among the use cases organizations are exploring are customer support through avatars, IT operations automation, digital content management, and the development of specialized agents to improve productivity in enterprise environments.

Key points

• Microsoft Agent Framework unifies AutoGen and Semantic Kernel into an open-source SDK for multi-agent systems.
• The framework allows local experimentation and deployment to Azure AI Foundry with integrated observability and regulatory compliance.
• Integrates any API through OpenAPI, collaborates across environments with Agent2Agent, and connects tools with Model Context Protocol.
• Incorporates advanced patterns like Magentic One and enables orchestrating agents into structured workflows.
• Multi-agent workflows in Foundry Agent Service coordinate multiple agents with persistent state.
• Workflows can be created visually through VS Code or Azure AI Foundry with integrated error handling.
• Use cases include customer support, IT operations automation, and digital content management.