The hardest thing to build in agentic AI isn't capability. It's restraint. What Taoist non-action taught me about designing agents that know when to stop.