Coordination is a central challenge in self-organizing multi-agent and sociotechnical systems, especially as control shifts from centralized to decentralized forms, demanding that components remain adaptive. This paper introduces a novel and explicit approach to social intelligence centered on agents’ ability to form and act on expectations about others’ behavior. This capability is grounded in interpretable mechanisms that allow agents to adjust strategies based on observed behavior, empirical expectations, and reference networks, mirroring core aspects of human social reasoning. We evaluate this framework across four canonical coordination games. By modeling symbolically, we provide a diagnostic lens for understanding and engineering coordination without reliance on black-box learning techniques. Our results highlight the potential of explicit, symbolic, expectation-based reasoning as a foundation for robust, decentralized coordination.