What is Microsoft Agent Framework?
Microsoft Agent Framework (MAF) is Microsoft's open-source, MIT-licensed framework for building AI agents and multi-agent workflows in Python and .NET, and the successor to both Semantic Kernel and AutoGen.
Checked
For two years Microsoft had two agent projects: Semantic Kernel for enterprise .NET and Python apps, and AutoGen for multi-agent research. Microsoft Agent Framework merges them. Semantic Kernel's README now says it is Microsoft Agent Framework, and AutoGen's says the project is archived and no longer maintained, with no bug or security fixes from Microsoft.
MAF reached 1.0 for Python and .NET on 3 April 2026, with stable APIs and a commitment to long-term support. Microsoft's message to existing Semantic Kernel and AutoGen users was direct: now is the time to migrate.
How it works
Agents and connectors
Single agents over Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic Claude, Amazon Bedrock, Google Gemini and Ollama, with middleware hooks for content safety, logging and compliance.
Workflows
A graph-based engine with branching, parallel fan-out, checkpointing, streaming, human-in-the-loop approvals and time-travel for long-running processes.
Multi-agent orchestration
Sequential, concurrent, handoff, group chat and Magentic-One patterns, with streaming, checkpointing and human-in-the-loop approvals.
Declarative and connected
Agents and workflows defined in YAML, MCP tool support, A2A for cross-framework calls, memory providers (Foundry, Mem0, Redis, Neo4j) and built-in OpenTelemetry tracing.
Version and status, checked 1 November 2026
Versions
Python agent-framework 1.24.0 and .NET Microsoft.Agents.AI 1.27.0, both released 30 October 2026. Releases land roughly weekly in each language.
Adoption
14,460 GitHub stars and 396K PyPI downloads a week for agent-framework-core, in the week to 1 November 2026. Go lives in a separate repo, microsoft/agent-framework-go.
Still in preview at 1.0
DevUI, Foundry hosted agents, Foundry tools, memory, observability and evaluations, AG-UI, CopilotKit and ChatKit adapters, Skills, the GitHub Copilot SDK and Claude Code SDK as agent harnesses, and a local Agent Harness. Preview APIs may change before they are marked stable.
Predecessors
Semantic Kernel still ships releases (.NET 1.81.0 on 6 October 2026) but points new work to MAF. AutoGen's last release was 0.7.5 on 30 September 2025.
When to pick it, and when not to
Pick it for
.NET teams, organisations standardised on Azure and Microsoft Foundry, and any team running Semantic Kernel or AutoGen today.
Also a fit when
The design needs named multi-agent patterns such as handoff or group chat, with checkpoints and approvals, in a framework with a long-term support commitment.
Look elsewhere when
The product is TypeScript-first: the main repo ships Python and .NET only. Teams wanting a ready file-and-shell harness may find a lab SDK closer to the job.
Weigh carefully when
Python is the only language. MAF reached 1.0 in April 2026, about six months after LangGraph, so community answers and third-party examples are thinner.
Gotchas from production
Check stable versus preview
Several features teams expect, including the DevUI, Skills and hosted agents, were preview at 1.0. Each should be checked in the current release notes before a design depends on it.
Migration is not a version bump
Moving from Semantic Kernel or AutoGen means rewriting against new abstractions. Microsoft publishes separate migration guides for each, and only release-candidate MAF users got a simple package bump.
AutoGen is a security risk to stay on
With no security fixes from Microsoft, an AutoGen agent in production carries unpatched dependency risk. AG2 is a separate community fork, not Microsoft's path.
Pin versions
Weekly minor releases in two languages mean a floating version range can change behaviour between builds.
How Motiontic approaches it
Motiontic's engineers are vendor-neutral. They run a short spike on the client's real task before committing to a framework or store, keep prompts, tool definitions and evals outside the framework so a later switch is a refactor rather than a rewrite, and hand the eval suite to the client's team so the next tooling decision can be measured rather than argued.
Microsoft Agent Framework, its predecessors and LangGraph, checked 1 November 2026
| Microsoft Agent Framework | Semantic Kernel | AutoGen | LangGraph | |
|---|---|---|---|---|
| Status | Active; 1.0 since 3 Apr 2026 | Active; README points to MAF | Archived; no Microsoft fixes | Active; 1.0 since 17 Oct 2025 |
| Languages | Python, .NET (Go separate) | .NET, Python, Java | Python, .NET | Python, JavaScript / TypeScript |
| Licence | MIT | MIT | MIT per PyPI metadata | MIT |
| Latest release | Python 1.24.0, .NET 1.27.0 (30 Oct 2026) | .NET 1.81.0 (6 Oct 2026) | 0.7.5 (30 Sep 2025) | 1.2.19 Python (30 Oct 2026) |
| Multi-agent model | Workflows plus named orchestration patterns | Agent and process frameworks | Conversational multi-agent | Explicit graphs |
| Fit for new work | Yes, Microsoft's forward path | Migrate to MAF | Migrate to MAF or AG2 | Yes, model-neutral |
Common questions
Is AutoGen deprecated?
Its README says the project is archived and no longer maintained, with no bug fixes, security fixes or new features from Microsoft, and directs new users to Microsoft Agent Framework.
What replaced Semantic Kernel?
Microsoft Agent Framework. Semantic Kernel's README describes MAF as its enterprise-ready successor, though Semantic Kernel itself still receives releases.
Does Microsoft Agent Framework work with non-Microsoft models?
Yes. The 1.0 release lists connectors for OpenAI, Anthropic Claude, Amazon Bedrock, Google Gemini and Ollama alongside Microsoft Foundry and Azure OpenAI.
Does Microsoft Agent Framework support TypeScript?
Not in the main repo, which covers Python and .NET. A Go implementation lives in a separate repository.
Related
Google ADK
Google's open-source agent framework, now with graph workflows, in five languages.
CrewAI vs LangGraph
Role-based crews and flows compared with an explicit state graph.
Multi-agent system
Several agents splitting one job, with some way of coordinating who does what.
Agent orchestration
The logic that decides what runs next, with what context, and who combines the results.
AI agent development
Agents that do real work for users, with the evals, traces and guardrails to keep them in production.
Sources
- Microsoft, Microsoft Agent Framework Version 1.0 (2026-04-03)
- GitHub, microsoft/agent-framework (14,460 stars; releases to 30 Oct 2026)
- PyPI, agent-framework (1.24.0, 30 Oct 2026)
- AutoGen README (archived, no longer maintained)
- Semantic Kernel README (now Microsoft Agent Framework)
- LangGraph docs, overview
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.