Welcome to 2026 🚀 With the new year, I am starting a regular news update on AI topics in the .NET world, focussing on AI use cases, software architecture, and production-grade patterns.
If you’re working hands-on with .NET, AI, or relevant enterprise architectures, feel free to connect 👋, follow 🔔, or jump into the discussion.
• .NET 10 positions agent-based AI as a first-class architectural pattern: With .NET 10, Microsoft consolidates agent orchestration patterns (Semantic Kernel / AutoGen) directly into the platform, including templates and runtime support. This shifts AI integration from ad-hoc service calls toward structured, stateful workflows with explicit orchestration. For production systems, this has implications for observability, lifecycle management, and concurrency control.
🔗 https://devblogs.microsoft.com/dotnet/announcing-dotnet-10/
• Microsoft.Extensions.AI introduces stable abstractions for generative AI clients: The new AI abstractions (IChatClient, IEmbeddingGenerator, etc.) formalize how generative AI integrates into modern .NET applications. By aligning with dependency injection and OpenTelemetry, they enable provider-agnostic architectures with proper monitoring and governance.
🔗 https://learn.microsoft.com/dotnet/ai/microsoft-extensions-ai
• Clear separation of concerns: Semantic Kernel vs. Microsoft.Extensions.AI: Recent guidance from Microsoft clarifies Semantic Kernel focus on orchestration, tools, and memory, while MEAI handles low-level client abstractions.
🔗 https://devblogs.microsoft.com/dotnet/generative-ai-with-large-language-models-in-dotnet-and-csharp
• Production-ready agent architectures emerge beyond demos: Community implementations show how Semantic Kernel can be embedded into clean or hexagonal architectures with explicit boundaries for prompts, tools, and domain logic. This matters as long-term maintainability, not model quality, is often the limiting factor in real AI systems.
🔗 https://dev.to/sebastiandevelops/building-production-ready-ai-agents-with-semantic-kernel-and-clean-net-architecture-4oeg