Several recent efforts have highlighted the need for improving dynamic representation of human water management and demand in large-scale hydrological models (LHMs) to evaluate the vulnerability of water-dependent sectors. Generic rule-based representation of water management in large-scale hydrology models have been one means to account for reservoirs and their operations in earth system models. However, a remaining limitation of LHMs is their incorporation of water demands that are largely static in regards to adaptive human decisions such as crop type, area, and irrigation. To better account for the dynamic feedback between water availability and water demand, we integrate an economic model of adaptive crop and irrigation decision making with MOSART-WM, a large-scale water management model of surface water routing and generalized reservoir operations. The model is deployed across the United States at 1/8 degree spatial resolution. The irrigation decision making model adopts a positive mathematical programming (PMP) approach in which agricultural agents are treated as profit-maximizing firms with constrained water availability, calibrated using historical land use, price, and cost observations that are consolidated from various USDA datasets. The PMP-driven agricultural agents are coupled with MOSART-WM in a co-evolving fashion, with each model time step involving a dynamic exchange of: 1) grid cell-specific water availability from MOSART-WM to the agricultural agents, and 2) water demand from the agricultural agents to MOSART-WM, influencing reservoir release decisions and supply allocations. Preliminary stress test experiments including drought and price shock scenarios are conducted to gain insight into the behavior of the coupled human-natural water system under extreme conditions.