Ai-Agents

Use middleware to wrap agent runs, intercept tool calls, and redact PII before the LLM sees it in both Python and .NET safely.

Add per-request context with focused providers, compose them before each LLM call, and plug in TextSearchProvider for simple RAG.

Most AI agent tutorials end at “hello world.” You build a single agent with one or two tools, it answers a few questions, and that is it. The gap between that tutorial and a production system with multiple agents, authentication, observability, and a real frontend is enormous.

Revised with .NET examples — A newer version of this article, covering both Python and .NET, is available as part of the MAF v1: Python and .NET series: MAF v1 — 01-first-agent, MAF v1 — 02-add-tools.

Revised with .NET examples — A newer version of this article, covering both Python and .NET, is available as part of the MAF v1: Python and .NET series: MAF v1 — 24-prompt-engineering. The newer version pulls role-specific behavior out of the system prompt and into a context provider, so you don’t re-instantiate the agent per role. The five concerns and YAML composition below are still the canonical architecture; the new chapter shows the modernised wiring.

Revised with .NET examples — A newer version of this article, covering both Python and .NET, is available as part of the MAF v1: Python and .NET series: MAF v1 — 02-add-tools.

Stream tokens as they arrive and reuse a session across turns so the LLM sees the full conversation. ~60 lines of code per language.

Serialize an AgentSession to JSON, persist it, reload in a fresh process, and have the agent pick up exactly where it left off.

Build the smallest useful Microsoft Agent Framework program: one LLM call, one response, and the same shape in Python and .NET.

Install uv, .NET 9, Docker, OpenAI keys, and a verify script so Chapter 1 starts with both Python and .NET environments ready.

Give an agent a callable function, let the LLM decide when to use it, and compare Python and .NET tool loops side by side.

Every vendor selling software right now claims their product is “agentic.” I’ve seen chatbots with a system prompt called an agent. I’ve seen a scheduled Python script described as autonomous AI. I’ve also shipped actual agents to production, at an insurance company, handling FNOL triage, policy lookup, and claims routing. The gap between what gets marketed as an agent and what you’d actually build in production is wide, and it’s expensive to get wrong.