We describe an AI recommender system (RS) with machine learning to harness past user choices and large volumes of data, yet account for changes in weather and management decisions characteristic of agricultural systems. Our goal is to maximize the use of data relevant to solving agricultural problems and improve the efficiency of the scientific workforce while also improving the accuracy of estimates of the amount of food produced. Our example shows how the RS learns data analysis choices from user behavior for predicting agricultural production responses to rainfall and learns to identify classes of agroecosystem responses to alternative climate scenarios. We account for changes in relationships using spatial and temporal statistics. The RS provides a powerful approach to make use of the large amounts of data and scientific expertise in the agricultural enterprise to predict agroecosystem dynamics under changing environmental conditions.