The present study details evidence for a predictive coding account of attention control, which expands as a unified theory of perception, cognition, and action. On this account, virtually every aspect of human cognition serves the universal imperative of minimizing prediction error between implicit hypotheses about the state of the external world and data from sensory inputs, under a constraint that computational efficiency is prioritized. In a novel demand-selection visual search paradigm, support for this view was evident in patterns of objective performance, self-reports, and indices of neuromodulatory activity collected via eye-tracking. In particular, observers routinely exploited the task strategy associated with less cognitive demand, but their behavior was also flexibly adaptive and sensitive to alternative motives for goal-driven behavior. These and additional findings collectively offer a substantial contribution to several different scientific literatures, and they validate the foundation for a promising theoretical framework of attention control.