The challenge of managing complexity, volatility, and interconnection in critical urban infrastructure requires Information Systems Engineering (ISE) to adopt a systemic perspective that transcends a narrow focus on functionality and efficiency. Classic control paradigms operate exclusively with endogenous information (internal and efficiency-related), which is inadequate for reconciling operational optimisation with the social, environmental, and regulatory constraints of the urban ecosystem (exogenous data). This paper presents and validates the Context-Aware Optimised Control (CAOC) model, a hybrid control proposal that aligns operational efficiency with contextual social requirements. The model achieves this by incorporating a new layer of intelligent processing, known as the Context Awareness Engine (CAE). The CAE utilises Large Language Models (LLMs), combined with in-context learning and prompt design techniques, to integrate diverse, complex, and textual information sources, such as regulations or geological reports. This engine operates in parallel with the Efficient Control Engine (ECE), which utilises predictive AI on endogenous numerical data to generate cost-optimised operating patterns. The decoupling of both inference flows (ECE and CAE) promotes adaptability and maintainability, which are crucial aspects for modern ISI. Validation of the model in a critical drinking water supply infrastructure demonstrated its ability to select the most appropriate operational mode, accounting for contextual risks. One of them is aquifer vulnerability, a factor that is unfeasible for control systems based solely on efficiency. The proposal contributes directly to the areas of adaptive and context-aware IS within Cyber-Physical Systems and smart city management.