Microsoft Agent Framework
The Microsoft Agent Framework is an open-source framework for agents and agent workflows with tool calling, memory, multi-agent orchestration and observability hooks, available for .NET and Python.
The Microsoft Agent Framework merges the AutoGen and Semantic Kernel lines into one framework for single- and multi-agent systems in .NET and Python. AutoGen itself is, per Microsoft, in maintenance rather than active development.
Explanation
Two Microsoft projects pursued the same goal: AutoGen as the research line for agent conversations, Semantic Kernel as the enterprise-facing library. The Agent Framework brings both together: research ideas about orchestration combined with what enterprises demand — telemetry, approval steps, integration with existing .NET services. For shops whose line-of-business applications run on .NET, this is the first agent framework that does not look like a Python port.
Marketing Relevance
For organisations on a Microsoft stack the framework lowers the entry barrier considerably, because identity, logging and operations match what is already there. The usual caution applies: a framework from your own cloud ecosystem is convenient and increases lock-in to that ecosystem.
Example
An inside sales team lets three agents handle enquiries: one researches the CRM, one checks availability in the ERP, one drafts the reply. An approval step holds back any reply that names a price until a human releases it.
Common Pitfalls
Multiple agents are almost always slower and more expensive than one with good tools — splitting needs a business reason. And migrating from AutoGen should be treated as a project, not a dependency bump; concepts were renamed and recut.
Origin & History
AutoGen appeared in 2023 and shaped the discussion about agent conversations; Semantic Kernel addressed enterprise integration in parallel. In 2025 Microsoft announced the merge, and AutoGen has been in maintenance since. The message for decision makers: new projects start in the Agent Framework.
Further Resources
Marketing Use Cases
Agents inside existing .NET line-of-business applications with enterprise identity.
Multi-step flows with an approval step before customer-facing actions.
Orchestration of several specialised agents with shared telemetry.
Frequently Asked Questions
Should we still start with AutoGen?
For new projects, no. Microsoft continues development in the Agent Framework and only maintains AutoGen. Existing systems need not switch immediately but should have a path.
Does it work without Azure?
The framework is open source and not tied to one provider. In practice identity, telemetry and model access are aligned with the Microsoft environment; outside it, integration effort rises.
Related Services
Go deeper: Agentic AI Hub · Model comparison 2026