Intermittent androgen deprivation therapy (IADT) is an attractive treatment approach for biochemically recurrent prostate cancer (PCa), whereby cycling treatment on and off has been shown to limit toxicities and reduce cumulative dose. While IADT has been shown to delay the development of treatment resistance, underlying mechanisms and actionable biomarkers are required to accurately predict when resistance will emerge and how to maximally delay time to progression. We developed a quantitative framework to simulate the enrichment of prostate cancer stem cell dynamics during treatment as a plausible mechanism of resistance evolution. Using simulated dynamics of cancer stem cells and non-stem cancer cells, we derive longitudinal blood serum prostate-specific antigen (PSA) concentrations that were calibrated to and validated with data of 70 patients undergoing multiple cycles of IADT. Model analysis suggests that cancer stem cell proliferation patterns correlate with PSA dynamics and patient outcomes.  Learning these dynamics adaptively from the earlier treatment cycles in individual patients can predict the evolution of resistance in subsequent IADT cycles with a predictive power of 90%, which warrants prospective evaluation in a clinical setting. The model is used to simulate and predict alternative treatment options. In particular, simulations show that response dynamics from the first IADT cycle can be used to identify patients who would benefit from concurrent docetaxel.