Getting Started: Microsoft Agent Framework in Python

Building reliable agentic systems in Python requires moving beyond basic LLM API calls and understanding how to manage state, external capabilities, and complex execution graphs. This course teaches you how to construct and orchestrate production-ready AI agents using the Microsoft Agent Framework. You will start with the foundational concepts of LLMs and agents before writing your First Agent. We cover different execution environments, including Foundry Agents and Local Agents, and show you how to inspect their behavior using the DevUI. Because real-world agents need memory, you will learn how to manage Sessions, handle Persisting Sessions, and optimize context windows using Compaction. From there, we extend your agents with Tools, covering everything from passing Complex Tool Parameters to requiring Tool Approval and debugging tool execution directly in the DevUI.

To ensure your agents return predictable data, we explore Structured Output, Output Validation, and Prompt Templates, eventually Combining Tools And Structured responses. The course then moves into Orchestration, where you will design multi-agent Workflows using Workflow Builders, implement Conditional Routing, and manage Events And State. You will also learn how to execute a Concurrent Workflow and integrate a Human In The Loop for manual intervention. Next, we address production concerns through Observability, teaching you how to implement Tracing Agent Calls, mitigate Prompt Injection attacks, and apply Output Filtering. Finally, we tie these concepts together by Designing The System from scratch, walking you through Building The Foundation, Building The Workflow, and ultimately Hardening The System for real-world usage.