News

Semantic Kernel + AutoGen = Open-Source 'Microsoft Agent Framework'

• By David Ramel
• 10/01/2025

With recent advances in agentic AI, Microsoft is revamping its related dev tooling, combining the capabilities of the discrete properties Semantic Kernel and AutoGen to offer the new open-source Microsoft Agent Framework to simplify the building, orchestrating and deploying of AI agents and multi-agent workflows, supporting Python and .NET.

"Developers asked us: why can't we have both — the innovation of AutoGen and the trust and stability of Semantic Kernel — in one unified framework?" Microsoft said in an Oct. 1 blog post. "That's exactly why we built the Microsoft Agent Framework."

Semantic Kernel is a lightweight, open-source SDK designed to help developers build intelligent AI agents and orchestrate complex workflows using large language models (LLMs).

AutoGen is an open-source framework for building multi-agent AI systems, designed to simplify the orchestration of LLMs, tools, and human inputs.

With the new offering, Microsoft's AI-related dev tooling space looks something like this:

"With Semantic Kernel, we gave developers a stable SDK with connectors into enterprise systems, content moderation, and telemetry," Microsoft said. "With AutoGen, pioneered in Microsoft Research, we opened the door to experimental multi-agent orchestration patterns that inspired the community. Both had passionate users -- but each had gaps."

Those gaps are reportedly being filled with the comprehensive new framework, which the company said aims to make the building of AI agents as easy as creating a web API or console app.

"Building AI agents shouldn't be rocket science," the company said in another Oct. 1 post. "Yet many developers find themselves wrestling with complex orchestration logic, struggling to connect multiple AI models, or spending weeks building hosting infrastructure just to get a simple agent into production."

Key takeaways of the new framework as presented by Microsoft include:

• Simple by Design: Get started with just a few lines of code. Create your first agent in minutes, not days.
• Scales with You: Start with a single agent, then easily add workflows, tools, hosting, and monitoring as your needs grow.
• Built on Proven Technology: Microsoft Agent Framework brings together the best from AutoGen and Semantic Kernel. It builds on Microsoft.Extensions.AI, a unified foundation for modern AI development, to deliver a robust and cohesive experience for .NET developers.
• Production Ready: Deploy using familiar .NET patterns with built-in observability, evaluation, and hosting capabilities.

Agents and Workflows
Microsoft defines agents as "systems that accomplish objectives," equipped with reasoning and decision-making, tool usage, and context awareness. Capabilities are powered by LLMs, APIs, and context sources such as vector stores or enterprise data. Workflows, meanwhile, define the steps required to achieve objectives, ranging from requirement gathering and design to testing and deployment. "When agents are equipped with reasoning, tools, and context, they can optimize workflows," Microsoft explained.

Building on Proven Foundations
While Microsoft Agent Framework brings together Semantic Kernel, AutoGen, the company's Microsoft.Extensions.AI is added to the mix to help create a cohesive developer experience. According to Microsoft, "By combining these technologies, Agent Framework offers reliability, flexibility, and a developer-friendly API. This allows you to build and deploy powerful AI agents quickly and efficiently." It supports both deterministic business workflows and dynamic multi-agent orchestration patterns pioneered in Microsoft Research.

Four Pillars of the Framework
The Azure AI Foundry team described four pillars underpinning the new framework:

• Open Standards & Interoperability -- Support for Model Context Protocol (MCP), Agent-to-Agent (A2A) messaging, and OpenAPI-first design, enabling cross-runtime portability.
• Pipeline for Research -- Experimental orchestration patterns from AutoGen, including group chat, debate, and reflection, now available with enterprise durability.
• Extensible by Design -- Modular architecture with connectors for Azure AI Foundry, Microsoft Graph, SharePoint, Elastic, Redis, and more. YAML and JSON declarative agent definitions allow version-controlled workflows.
• Ready for Production -- Native observability with OpenTelemetry, Azure Monitor integration, Entra ID authentication, and CI/CD support via GitHub Actions and Azure DevOps.

Getting Started with .NET and Python
The GitHub repo highlights quick setup for both languages. Developers can install via pip install agent-framework for Python or dotnet add package Microsoft.Agents.AI for .NET. Examples include a "HaikuBot" agent that generates poems and a "Hello World" multi-agent workflow connecting a writer and editor. Microsoft emphasized, "In just a few lines of code, you have a fully functional AI agent."

Workflows and Tools
Beyond simple sequential pipelines, the framework supports concurrent, handoff, and group chat workflows. Developers can enhance agents with external tools using MCP servers, hosted interpreters, or APIs. For instance, one demo connects a writing agent with an editor agent to automatically refine output, while more complex workflows support customer service or research pipelines.

Enterprise Momentum
Microsoft showcased early adopters of the framework across industries:

• KPMG -- Using Agent Framework to automate audit testing and documentation. "The governance and observability in Azure AI Foundry provide what KPMG firms need to be successful in a regulated industry."
• BMW -- Deploying multi-agent systems to analyze terabytes of vehicle telemetry. "Durability and observability are key for our operations... engineers don't just access data -- they get insights they can act on immediately."
• Commerzbank -- Piloting avatar-driven customer support. "The new Microsoft Agent Framework simplifies coding, reduces efforts and fully supports MCP for agentic solutions."
• Fujitsu -- Embedding the framework into integration services to balance human and AI collaboration.
• Citrix, TCS, TeamViewer, Elastic, and others are also actively adopting or integrating the framework for customer, IT, and developer-facing scenarios.

Migration Path from Semantic Kernel and AutoGen
Microsoft stressed continuity for existing developers. "Semantic Kernel users replace Kernel and plugin patterns with the Agent and Tool abstractions," while AutoGen users map the AssistantAgent to the new ChatAgent, benefiting from checkpointing, simplified messaging, and stronger durability. The company said, "Microsoft Agent Framework is not a replacement for what came before -- it is the natural evolution that unites innovation and stability."

Availability
The Microsoft Agent Framework is available now under the MIT license on GitHub, with documentation, samples, and Microsoft Learn modules. Developers can also test agents via GitHub Codespaces and follow tutorials for both Python and .NET.