Microsoft has merged Semantic Kernel and AutoGen to create the Agent Framework, stirring the AI Agent waters.
Microsoft Agent Framework is an open-source development kit for building AI agents and multi-agent workflows for .NET and Python. The big deal is that, with the new framework, Microsoft provides the tooling necessary for orchestrating multiple agents that work on complex multi-step tasks.
Those orchestration scenarios include:
Sequential Agent orchestration in scenarios where step-by-step workflows are needed.
Concurrent orchestration in scenarios where agents need to complete tasks at the same time.
Group chat orchestration in scenarios where agents can collaborate together on one task.
Handoff Orchestration in scenarios where agents hand off the task to one another as the subtasks are completed.
Magnetic Orchestration in scenarios where a manager agent creates and modifies a task list and handles the coordination of subagents to complete the task
But Agentic systems are not just about orchestration. There's the aspect of Autonomy in that the agent decides by itself which tools and when to invoke, as well as the aspect of Observability where the developer is able to see what the system does in order to evaluate it. The Agent Framework does 3 out of 3. Orchestration, Autonomy and Observability.
Microsoft Agent Framework is also focused on being inter-operable by:
Being Cloud-agnostic - Agents can run in containers, on-prem and across multiple different clouds.
Being Provider-agnostic - Agents can be created through your preferred SDK including Azure OpenAI and OpenAI
Integrating Open Standards - Agents can utilize protocols such as Agent-to-Agent(A2A) and Model Context Protocol (MCP) to discover and use other agents and tools.
Plugins and Connectors - Connections can be made to data and memory services such as Microsoft Fabric, SharePoint, Pinecone and Qdrant.
Coding time! The framework is available for both Python and .NET, so to install it :
#Python
pip install agent-framework --pre
#.NET
dotnet add package Microsoft.Agents.AI
Now after you authenticate with Azure CLI, you can write your first basic Azure Responses Agent that writes a haiku about the Microsoft Agent Framework:
The project's Github repository features many more code samples to get you started. There you'll find code for creating agents:
using Azure AI Foundry Agent Service with various tools including function tools, code interpreter, MCP integration, and thread management
using Microsoft Copilot Studio with streaming and non-streaming responses, authentication handling, and explicit configuration options
using Azure OpenAI APIs with multiple client types (Assistants, Chat, and Responses clients) supporting function tools, code interpreter, thread management, and more
using OpenAI APIs with comprehensive examples including Assistants, Chat, and Responses clients featuring function tools, code interpreter, file search, web search, MCP integration, image analysis/generation, structured outputs, reasoning, and thread management
using Anthropic models through OpenAI Chat Client configuration, demonstrating tool calling capabilities
custom and chat clients by extending the base framework classes, showing complete control over agent behavior and backend integration
If you're more of a visual builder then you'll be delighted to know that workflows can be authored and debugged visually through the VS Code Extension or Azure AI Foundry, then deployed, tested, and managed in Foundry alongside existing solutions. That is a cloud Foundry Agent Service feature, however.
A child-sized robot has achieved lift-off and stable flight after nearly ten years of development. The developers say this is the first successful vertical takeoff of a jet-powered flying humanoid rob [ ... ]
Linus Torvalds seems to have had a bit of a relapse lately in his efforts to moderate his comments on code. One of his latest outbursts merits more than a surface analysis.
