Microsoft Agent Framework Reaches Release Candidate
Microsoft’s Agent Framework has reached release candidate status, positioning it as an enterprise solution for orchestrating and governing agentic AI workflows. The company also released official guidance for developers to migrate projects from its earlier Semantic Kernel and AutoGen toolkits to the new, more robust framework.
- The framework unifies two of Microsoft's earlier open-source projects, combining the enterprise-grade stability and connectors of Semantic Kernel with the experimental multi-agent orchestration patterns from the AutoGen research project. - It is designed with two distinct modes of operation: flexible, LLM-driven agent collaboration for creative tasks, and structured, graph-based workflows for predictable, deterministic enterprise processes. - For enterprise-level governance, the framework integrates with Microsoft Entra ID for agent authentication, includes hooks for content safety, and uses OpenTelemetry for built-in observability and debugging. - The framework is a foundational component of Microsoft's larger AI ecosystem, designed for pro-code development of agents that can then be deployed to Azure AI Foundry and managed or customized by business users via the low-code Copilot Studio. - It is model-agnostic, allowing developers to plug in LLMs from various providers, including Azure OpenAI, OpenAI, and local models via Ollama. - Support for open standards is a core feature, using OpenAPI specifications to instantly turn any compliant REST API into a callable tool for an agent without needing custom wrappers. - The architecture is built for complex, long-running tasks by providing robust state management and explicit support for "human-in-the-loop" scenarios where human approval is required to proceed. - The Release Candidate status, announced in February 2026, signals that the API is now stable and feature-complete ahead of the full version 1.0 General Availability release.
