The .NET team, including Bruno Capuano and Justin Yoo, provides an enterprise-focused walkthrough of AI agent architectures in .NET, showcasing Microsoft’s Agent Framework with deployment on Azure.

.NET AI Community Standup: Real-World AI Agent Architecture in .NET

This community standup session dives deep into what’s required to build real-world AI agent systems using .NET, moving beyond simple examples to focus on production-grade architectures suitable for enterprise workloads.

Technologies and Patterns Explored

  • Microsoft Agent Framework (MAF): Fundamentals and usage for orchestrating multi-agent systems within .NET applications.
  • Microsoft Foundry as Model Backend: How the open-source Interview Coach sample leverages Foundry for advanced model hosting and integration.
  • Model Context Protocol (MCP): Facilitating powerful tool integration and model abstraction within AI agent workflows.
  • Aspire: Enabling orchestration, service topology, health checks, and observability for containerized .NET AI systems.
  • Deployment: Strategies for deploying agent-based applications to Azure Container Apps for scalable, cloud-native operations.

Key Architectural Topics

  • Handoff vs Agent-as-Tools Patterns: Analyzing when to use different agent composition strategies in production workloads.
  • Service Topology Design: Structuring AI systems for modularity, scalability, and reliability.
  • Telemetry & Health Checks: Ensuring robust monitoring, logging, and system health with Aspire.
  • Model Abstraction via IChatClient: Creating flexible interfaces for interacting with AI models and backend systems.

Use Cases

  • Interview Coach Sample: An open-source .NET project serving as a reference for production-level agent system design.
  • Enterprise Workloads: Guidance on adapting these patterns for complex, real-world AI applications in business environments.

Further Resources

This session is especially valuable for developers designing AI systems in .NET who want actionable architectural guidance straight from the community and product teams.