Microsoft Agent Framework (public preview): enterprise-ready open-source SDK & runtime for multi‑agent apps

Key update

Microsoft published the Microsoft Agent Framework as an open‑source, public preview and integrated it with Azure AI Foundry’s Agent Service. The framework unifies the research orchestration patterns from AutoGen with the enterprise features of Semantic Kernel into a single SDK and runtime (with first‑class support for .NET and Python), adds Open Standards interoperability (MCP, Agent‑to‑Agent, OpenAPI), and delivers production primitives such as long‑running durable workflows, human‑in‑the‑loop approvals, built‑in OpenTelemetry tracing, and CI/CD/monitoring integration for scaling agentic applications.

Why it matters

This is a practical inflection point for developers building agentic systems. Until now teams traded experimental orchestration patterns for enterprise readiness; Microsoft is packaging both in one supported, open‑source stack and aligning it with a cloud runtime. For engineering teams that already use Semantic Kernel or AutoGen the migration path is explicitly covered and should be low friction: .NET namespaces map into Microsoft.Extensions.AI, Python gets a pip package, and many AutoGen orchestration patterns map to the new Workflow API. That reduces boilerplate and gives teams durable checkpoints, pause/resume, and observability that are essential for real production workloads.

Operationally, the framework changes how you design systems: agents become first‑class components that call tools via OpenAPI or MCP and communicate across runtimes via A2A. The built‑in OpenTelemetry support and Azure Foundry hosting mean debugging, tracing, and compliance are feasible at scale rather than being ad‑hoc. For security‑sensitive or regulated environments, the framework’s Entra integration, content safety hooks, and approval workflows lower the bar for compliance-ready deployments.

Immediate developer actions are straightforward: evaluate the SDK for any workflows that would benefit from durable orchestration or tool‑calling, experiment locally with the VS Code AI Toolkit updates, and plan observability and approval flows into your CI/CD pipelines. Be mindful of cloud coupling versus portability; the use of MCP/A2A/OpenAPI improves interoperability, but leveraging Foundry hosting and Microsoft connectors will still introduce operational ties to Azure. Overall, this is a meaningful, pragmatic step toward making multi‑agent architectures production‑ready rather than research demos.

Source

• Source