The tremendous complexity of modern distribution systems calls for alternative coordination architectures, supported by smart, self-adaptable, and, to a large degree, environment-agnostic algorithms. In this article, we discuss decentralized and distributed coordination architectures for the operation of active distribution grids aiming at effectively coping with their complexity. We present relevant methods and algorithms under the framework of multiagent systems (MASs) and decentralized decision-making associated with handling different parts of the optimal grid operation. The decision-making models are based on distributed optimization algorithms using consensus/gossip models, bioinspired algorithms from the field of population dynamics, and a method for decomposing the power-flow model. The developed techniques aim at matching production with demand in microgrids, settling the short-term energy imbalances at the distribution grid level, mitigating voltage deviations, and resolving distribution grid congestions in real-time operation. The algorithms are implemented as MAS-based software platforms, able to aggregate diverse DG units and flexible loads. Results are provided from the theoretical simulation-based models and demonstrations of the operational techniques in actual pilot sites. The applied implementations have been performed in a smart grid pilot site, for which MAS platforms have been developed and tested.