The increasing demand for agricultural productivity coupled with labor shortages and the need for sustainable farming practices has accelerated the development of autonomous harvesting technologies. Traditional single-robot systems face fundamental limitations in coverage, throughput, and fault tolerance when deployed across large-scale agricultural fields. Swarm intelligence, inspired by collective behaviors observed in natural systems, offers a paradigm shift toward decentralized, scalable, and robust multi-robot harvesting operations. This review examines the application of swarm intelligence algorithms and architectures to autonomous harvesting systems, emphasizing decentralized coordination mechanisms, adaptive task allocation strategies, and real-time collision avoidance protocols. We analyze core swarm intelligence principles including stigmergy, self-organization, and emergent behaviors as applied to agricultural robotics, and evaluate their performance across diverse harvesting scenarios encompassing fruit picking, grain harvesting, and vegetable collection. Critical communication strategies, sensing modalities, and distributed decision-making frameworks are systematically reviewed. Field implementations demonstrate that swarm-based approaches achieve superior scalability, operational resilience, and adaptability compared to centralized control architectures. However, significant challenges remain in energy management, heterogeneous fleet coordination, and real-time computational constraints. We conclude by identifying pathways toward commercial deployment, including standardized communication protocols, hybrid control architectures, and integration with precision agriculture ecosystems, positioning swarm robotics as a transformative technology for sustainable agricultural intensification.