We keep calling systems “agents” because they can use tools, remember context, and complete multi-step tasks. But how much of that is actual intelligence, and how much is simply better orchestration around a language model? At what point do you conside…