Generative AI for Beginners .NET v2: Rebuilt for .NET 10 with Microsoft.Extensions.AI

Microsoft has updated Generative AI for Beginners .NET to Version 2. The course is free, open-source, and now rebuilt entirely for .NET 10 with a significant architectural change: Semantic Kernel is out as the primary abstraction, replaced by Microsoft.Extensions.AI (MEAI).

The Shift to Microsoft.Extensions.AI

Version 1 leaned on Semantic Kernel for orchestration and model access. Version 2 standardizes on MEAI’s IChatClient interface, which ships as part of .NET 10 and follows the same dependency injection conventions as ILogger.

The registration pattern will be familiar to any .NET developer:

var builder = Host.CreateApplicationBuilder();

// Register any IChatClient-compatible provider
builder.Services.AddChatClient(new OllamaChatClient("phi4"));

var app = builder.Build();
var client = app.Services.GetRequiredService<IChatClient>();

var response = await client.GetStreamingResponseAsync("What is AOT compilation?");
await foreach (var update in response)
    Console.Write(update.Text);

The interface is provider-agnostic. Swapping OllamaChatClient for an Azure OpenAI implementation requires changing a single line. The course uses this deliberately — the skills transfer between providers rather than locking you into one vendor’s SDK.

What the Five Lessons Cover

The restructured curriculum runs in five self-contained lessons:

1. Foundations — LLM mechanics, tokens, context windows, and how .NET 10 integrates with model APIs
2. Core Techniques — Chat completions, prompt engineering, function calling, structured outputs, and RAG basics
3. AI Patterns — Semantic search, retrieval-augmented generation, document processing pipelines
4. Agents — Tool use, multi-agent orchestration, and Model Context Protocol (MCP) integration using .NET 10’s built-in MCP client support
5. Responsible AI — Bias detection, content safety APIs, and transparency guidelines

The agent lesson is particularly relevant if you have been following .NET 10’s MCP support. The course connects multi-agent orchestration directly to that feature using the Microsoft.Extensions.AI.Abstractions MCP client, so you can run samples against local or remote MCP servers without framework gymnastics.

Migrating from Version 1

The eleven Semantic Kernel samples from Version 1 are moved to a deprecated folder inside the repo — they still run, but are no longer presented as the recommended pattern. If you worked through Version 1, the core concepts remain the same. The migration is mostly a swap at the API layer: replace Semantic Kernel’s Kernel and IKernelBuilder with IChatClient and the standard IServiceCollection extensions.

The course repository is at github.com/microsoft/generative-ai-for-beginners-dotnet. The course itself starts at aka.ms/genainet.