dotnetJune 23, 20261h 1m

On .NET Live - An In-Depth Introduction to Agentics

TL;DR

A one-month-old AI book already had outdated code: Jesse says Microsoft Agent Framework is changing so quickly that an MVP-built tool called MAF Doctor now updates stale examples every Thursday to keep pace.
He rebuilt the same app twice to compare ecosystems directly: the original blog-writing workflow used Python with LangChain and LangGraph, then got ported to C# with Microsoft Agent Framework so he could see both the similarities and the tradeoffs side by side.
Multi-agent systems should stay small unless the task truly needs them: Jesse's app uses a researcher, author, reviewer, and blogger, but he says production systems should generally keep agent counts to 3-5 because complexity explodes fast.
Prompt engineering is not a side skill, it's central to the work: he argues the first 80 percent is deciding what you want, shaping system and user prompts, and setting constraints, while the actual app code is often the last 20 percent.
Coding with AI and building AI products are two different disciplines: using AI as a coding partner is separate from designing an application where AI is a core runtime component, and he says people often blur those together.
Traditional keyboard-first coding may shrink fast: Jesse predicts that within a year, developers will spend far less time hand-writing code and far more time directing AI assistants, reviewing output, and designing AI-native applications.

The Breakdown

A book published in May was already obsolete a month later, and that became the perfect frame for Jesse Liberty's big point: learning agentic AI right now feels like changing the tires on a moving car. He walks through porting a multi-agent blog-writing app from Python's LangChain and LangGraph stack into C# with Microsoft Agent Framework, while arguing that prompt design, planning, and AI fluency are quickly becoming more important than hand-writing every line of code.