Microsoft Releases ‘Microsoft Agent Framework’: An Open-Supply SDK and Runtime that Simplifies the Orchestration of Multi-Agent Methods

Microsoft launched the Microsoft Agent Framework (public preview), an open-source SDK and runtime that unifies core concepts from AutoGen (agent runtime and multi-agent patterns) with Semantic Kernel (enterprise controls, state, plugins) to assist groups construct, deploy, and observe production-grade AI brokers and multi-agent workflows. The framework is out there for Python and .NET and integrates immediately with Azure AI Foundry’s Agent Service for scaling and operations.

What precisely is Microsoft transport?

• A consolidated agent runtime and API floor. The Agent Framework carries ahead AutoGen’s single- and multi-agent abstractions whereas including Semantic Kernel’s enterprise options: thread-based state administration, sort security, filters, telemetry, and broad mannequin/embedding help. Microsoft positions it because the successor constructed by the identical groups, quite than a substitute that abandons both undertaking.
• First-class orchestration modes. It helps agent orchestration (LLM-driven decision-making) and workflow orchestration (deterministic, business-logic multi-agent flows), enabling hybrid methods the place inventive planning coexists with dependable handoffs and constraints.
• Professional-code and platform interoperability. The bottom AIAgent interface is designed to swap chat mannequin suppliers and to interoperate with Azure AI Foundry Brokers, OpenAI Assistants, and Copilot Studio, decreasing vendor lock-in on the utility layer.
• Open-source, multi-language SDKs underneath MIT license. The GitHub repo publishes Python and .NET packages with examples and CI/CD-friendly scaffolding. AutoGen stays maintained (bug fixes, safety patches) with steerage to think about Agent Framework for brand new builds.

The place it runs in manufacturing?

Azure AI Foundry’s Agent Service offers the managed runtime: it hyperlinks fashions, instruments, and frameworks; manages thread state; enforces content material security and id; and wires in observability. It additionally helps multi-agent orchestration natively and distinguishes itself from Copilot Studio’s low-code strategy by concentrating on advanced, pro-code enterprise eventualities.

However how is it related to ‘AI economics’?

Enterprise AI economics are dominated by token throughput, latency, failure restoration, and observability. Microsoft’s consolidation addresses these by (a) giving one runtime abstraction for agent collaboration and gear use, (b) attaching manufacturing controls—telemetry, filters, id/networking, security—to the identical abstraction, and (c) deploying onto a managed service that handles scaling, coverage, and diagnostics. This reduces the “glue code” that usually drives price and brittleness in multi-agent methods and aligns with Azure AI Foundry’s model-catalog + toolchain strategy.

Architectural notes and developer floor

• Runtime & state: Brokers coordinate by way of a runtime that handles lifecycles, identities, communication, and safety boundaries—ideas inherited and formalized from AutoGen. Threads are the unit of state, enabling reproducible runs, retries, and audits.
• Features & plugins: The framework leans on Semantic Kernel’s plugin structure and function-calling to bind instruments (code interpreters, customized features) into agent insurance policies with typed contracts. (
• Mannequin/supplier flexibility: The identical agent interface can goal Azure OpenAI, OpenAI, native runtimes (e.g., Ollama/Foundry Native), and GitHub Fashions, enabling price/efficiency tuning per process with out rewriting orchestration logic.

Enterprise context

Microsoft frames the discharge as a part of a broader push towards interoperable, standard-friendly “agentic” methods throughout Azure AI Foundry—in keeping with prior statements about multi-agent collaboration, reminiscence, and structured retrieval. Anticipate tighter ties to Foundry observability and governance controls as these stabilize.

We like this route as a result of it collapses two divergent stacks—AutoGen’s multi-agent runtime and Semantic Kernel’s enterprise plumbing—into one API floor with a managed path to manufacturing. The thread-based state mannequin and OpenTelemetry hooks tackle the standard blind spots in agentic methods (repro, latency tracing, failure triage), and Azure AI Foundry’s Agent Service takes on id, content material security, and gear orchestration so groups can iterate on insurance policies as an alternative of glue code. The Python/.NET parity and supplier flexibility (Azure OpenAI, OpenAI, GitHub Fashions, native runtimes) additionally make price/perf tuning sensible with out rewriting orchestration.