The application of LLM agentic frameworks as a bridge between social and technical domains in Integrated Water Systems management.

Socio-hydrology has emerged as a critical framework for Integrated Water Systems (IWS) management, emphasising the need to bring governance, policy, and institutional dynamics alongside physical and technical processes. Achieving meaningful integration between social and technical domains presents both epistemological and methodological challenges, requiring the coexistence of quantitative and qualitative forms of knowledge and the development of hybrid, co-creative modelling approaches. Recent advances in Large Language Models (LLMs) offer new opportunities to address this socio-technical divide. These systems can process heterogeneous data sources such as text, images, and video, and translate qualitative information into representations that can be systematically analysed alongside numerical model outputs. LLMs can be embedded within agentic frameworks and be applied to emulate aspects of human reasoning, interaction, and decision-making. This enables the exploration of social behaviours and institutional responses within complex IWS contexts. Our study reviews existing applications of LLMs in IWS research and categorises them according to their use of LLM affordances, particularly in terms of agentic autonomy and the nature of simulated social interactions. Building on this review, we propose a novel approach that employs LLM-based agents as interpretative intermediaries between IWS simulation outputs, policy documents, and stakeholder relationships in decision-making scenarios. Applying such systems to real-world water management problems and conducting statistical and comparative analyses can reveal previously inaccessible relationships and dependencies between social and technical components of IWS, supporting more integrated and adaptive water governance.