Skip to content

AI / LLM Integration

The 2026 reality. Microsoft Agent Framework went GA in April 2026, the official MCP C# SDK reached v1.0 in March 2026, Azure AI Foundry SDKs are at 2.0, and Microsoft.Extensions.AI has crystallized as the canonical .NET AI abstraction. This section covers the full stack — plus an explicit AI Models & Vendors Landscape subsection so you can hold an informed conversation about which model to pick for what workload.

This is the largest section in the guide because the surface area is wide and the material is freshest. Read it after the rest of the guide is internalized — senior AI work in .NET assumes solid grounding in DI, async, observability, security, and resilience.

Lookup & Reference

  •   AI Models & Vendors Landscape

    14 topics. OpenAI / Anthropic / Google / open-source / Phi; specialty models; selection matrix; pricing & cost engineering; benchmarks; AI coding tools (Copilot, Cursor, Windsurf, Cody, Continue, Claude Code); .NET connectivity per vendor; lock-in strategy.

  •   AI Layer Overview

    The 2026 Microsoft AI ecosystem map — how the pieces fit together.

  •   Deprecated & Outdated

    Things that were canonical in 2024 but should not be used in new code in 2026.

Microsoft Stack

Patterns & Storage

  •   RAG & Vector Stores

    7 topics. Microsoft.Extensions.VectorData, chunking & ingestion, store implementations (Azure AI Search / Qdrant / Azure SQL / Postgres pgvector / Cosmos), hybrid search & rerank.

  •   Local & Edge Inference

    3 topics. ONNX Runtime GenAI for Phi-3/Phi-4, Ollama + Aspire, perf trade-offs.

Hosting & Tooling

Operations

  •   AI Observability

    3 topics. OTel GenAI semantic conventions, prompt/response logging, agent tracing.

  •   AI Security

    3 topics. Prompt injection & shields, PII & output validation, jailbreak detection.

Why this order

Models & vendors landscape leads so you can talk fluently about OpenAI / Anthropic / Google / open-source choices before diving into how to call them from .NET. Then Microsoft.Extensions.AI because it's the contract every later piece implements. Agent Framework follows because it's the orchestration layer on top of Extensions.AI. MCP comes next because Agent Framework integrates it. RAG and vector stores follow because RAG is the most common production AI pattern. Local/edge inference, Foundry, observability, security, GitHub Copilot, and SK-legacy close the section.

Cross-references