Building AI Agents with .NET — Part 2: Advanced Agentic Architecture and Production Workflows

Build advanced, production-oriented AI agent systems with C#, .NET, and Microsoft Agent Framework.

Building AI Agents with .NET — Part 2 continues the journey from Part 1, moving beyond foundational agent development into the architectural challenges that emerge when AI agents operate across workflows, tools, services, and other agents.

Through practical implementations, this book explores human-in-the-loop workflows, Model Context Protocol (MCP), dynamic multi-agent orchestration, Magentic orchestration, durable workflows, OpenTelemetry-based observability, Agent-to-Agent (A2A) communication, and production-ready context engineering.

You will learn how to:

• Design human-in-the-loop workflows with deterministic approval policies
• Expose and consume enterprise capabilities using MCP
• Build dynamic multi-agent systems with specialized agents and supervisors
• Apply multiple orchestration patterns for complex agent workflows
• Build durable workflows that survive long-running business processes
• Add workflow tracing and observability with OpenTelemetry
• Enable distributed agent communication using A2A
• Engineer bounded, relevant, and explainable context for production multi-agent systems
• Test agentic architectures while clearly separating deterministic behavior from model-driven reasoning

The examples use C#, .NET, Microsoft Agent Framework, and Azure OpenAI , with an emphasis on architecture, reliability, observability, testing, and explicit production boundaries.

This book is intended for .NET developers, software architects, AI engineers, and technical leads who want to move from individual AI agents toward more sophisticated, production-oriented agentic systems.

Companion source code is available on GitHub , with chapter-by-chapter implementations and tests.