Microsoft Agent Framework (hands-on)
Are you making raw API calls to language models and struggling with session management, tool invocation, error handling, and retry logic — all written by hand? Is your agent code growing into a tangle of JSON parsing, prompt construction, and fragile workarounds?
There is a better way.
The Microsoft Agent Framework is Microsoft's production-grade SDK for building reliable AI agents in Python and C#. Born from the fusion of AutoGen and Semantic Kernel, it gives you a clean abstraction over any LLM provider — Azure AI Foundry, OpenAI, Anthropic Claude, Ollama, and more — while handling the hard parts: streaming tool calls, session state, orchestration, observability, and deployment.
In this course you will learn to build, test, and deploy production-ready agents from first principles. You will work in both Python and C#, deploy to Azure AI Foundry, wire agents together into multi-agent systems, and ship them with observability, evaluation, and human oversight built in.
Course objectives
After completing this course, you will be able to:
- Build AI agents using the Microsoft Agent Framework
- Work with both Python and C# when developing agents
- Handle sessions, memory, and tool execution in agent-based systems
- Design multi-agent workflows and orchestration patterns
- Implement observability and evaluation for AI agents
- Deploy production-ready agents using Azure AI Foundry
Target audience
This course is intended for:
- Developers working with Python or C#
- Engineers interested in AI agent development
- Developers building applications with LLM integrations
Prerequisite
This course is intended for professional developers who are familiar with Python and/or C#. Experience with REST APIs and async programming is assumed. No prior experience with AI frameworks or LLMs is required.
Rick Beerendonk - Microsoft Agent Framework (hands-on)
Rick is a senior consultant and trainer from The Netherlands. He has over 25 years of professional experience while working in small, large and fast growing organisations. His passion is simplicity, well-written code and team dynamics. He is specialised in front-end technologies and AI engineering and speaks regularly about these topics at international events.
Part 1: Foundations
- What is the Microsoft Agent Framework and why it exists
- The problem with raw LLM calls
- Key abstractions: IChatClient, agents, tools, sessions
- History: AutoGen + Semantic Kernel → Agent Framework
- Installation and project setup
- Your first agent in Python and C#
- Running agents locally with Ollama; running against Azure AI Foundry
Part 2: Core Agent Capabilities
- Tools: calling functions from agent responses, automatic tool dispatch
- Sessions and memory: conversation history, in-memory and persistent stores
- Middleware: logging, retry, token limits, custom pipeline steps
- Providers: Azure AI Foundry, OpenAI, Anthropic Claude, Ollama, Foundry Local, custom providers
- Structured output: typed responses using JSON schema
Part 3: Orchestration and Workflows
- Workflows: sequential, parallel, and conditional agent pipelines
- Multi-agent orchestration: coordinator agents, specialist agents, delegation patterns
- Durable agents: checkpointing, resumable long-running tasks
- Human approval: requiring a human decision before an agent proceeds
Part 4: Quality and Observability
- Observability: OpenTelemetry traces, Azure Monitor integration, structured logs
- Agent evaluation: writing automated evaluators, keyword checks, tool-call assertions, custom scoring functions
- Developer UI (DevUI): real-time inspection of tool calls and session state during development
- Agent skills: reusable context providers, inline skills, resource and script attachments
Part 5: Production Deployment
- Foundry Hosted Agents: deploying agents to Azure AI Foundry with agent.yaml, auto-scaling
- CopilotKit and ChatKit: exposing agents over AG-UI to React front-ends
- Coding agents: CopilotAgent and ClaudeCodeAgent for automated code tasks, wrapping as orchestrator tools
- Agent Harness: sandboxed execution with controlled file system access and command allow-lists; stdio/SSE transport for CI/CD integration
Appendix
- OWASP LLM Top 10: prompt injection, training data poisoning, insecure output handling, and mitigations
- Vector stores and embeddings: semantic search, RAG patterns with Azure AI Search
Practical information
Duration: 3 days
Price: 20 900 NOK
Language: English
FAQ
Hvem passer dette kurset for?
Kurset passer for utviklere som ønsker å jobbe med AI-agenter og moderne LLM-baserte løsninger.
Må jeg kunne AI fra før?
Nei, du trenger ikke erfaring med AI eller LLM-er, men du må kunne Python eller C#.
Er kurset praktisk rettet?
Ja, kurset er workshop-basert med hands-on labs gjennom hele perioden.
Hvilke teknologier brukes i kurset?
Kurset bruker Microsoft Agent Framework, Azure AI Foundry og flere LLM-leverandører.
Dekker kurset produksjonssetting av løsninger?
Ja, du lærer hvordan du bygger og deployer produksjonsklare AI-agenter.
Other relevant courses
Other subject areas
