Why think in principles at all

Yan opens with an analogy to web development. HTML arrived in 1993; React arrived in 2013; by 2025 React and its descendants dominate. React won not just as a tool but as a philosophy, a pattern of reactivity and modularity that became the accepted standard. His point is that building agents in 2025 feels like the early days of raw HTML and CSS: no approach has become the agreed standard yet. So rather than copying a framework, he wants to reason up from principles. He is explicitly critical of libraries like OpenAI's Swarm and Microsoft's AutoGen for pushing multi-agent architectures, which he believes are the wrong default.

The whole article builds to two principles.