In smart city scenarios, the use of unmanned aerial vehicle (UAV) networks is one of actively discussed technologies. In this paper, we consider the scenario where carpoolable UAV-based drone taxis configure their optimal routes to deliver packages and passengers in an autonomous and efficient way. In order to realize this application with drone-taxi UAV networks, a multiagent deep reinforcement learning (MADRL) based algorithm is designed and implemented for the optimal route configuration. In the corresponding MADRL formulation, the drone-taxi related states, actions, and rewards are defined in this paper. Lastly, we confirm that our proposed algorithm achieves desired results.