Freshwater scarcity has increased the demand for precision irrigation, and root water uptake (RWU) models are now essential for field-scale water management. Conventional macroscopic models estimate uptake capacity from root length density. While computationally efficient, these frameworks overlook root topology, xylem resistance, and rhizosphere heterogeneity and often fail to simulate compensatory uptake under heterogeneous soil moisture such as partial root-zone drying. Hydraulic architecture theory addresses this gap by representing roots as conductance-based networks, in which compensatory uptake arises from physical principles. Though fully resolved 3D models capture microscale rhizosphere processes, practical implementation is often limited by high parameter demands and computational overhead. Recent advances in root imaging, stochastic modeling, upscaling, and machine learning guided by physical principles have eased bottlenecks in simulation across scales. Belowground digital twins open further opportunities for precision irrigation. This review charts the evolution of RWU models and evaluates recent progress in parameterization and upscaling, with the aim of guiding drought-tolerant breeding and water-saving irrigation design. An illustrative case study demonstrates the practical consequences of the modeling choice for irrigation design.