The use of gradient-based optimization in intelligent agents allows them to make reasoned decisions by predicting objective functions, leading to a more efficient approach than zero-th order, gradient-free optimization.