Large-scale neural recordings provide an opportunity to better understand how the brain implements critical behavioral computations related to goal-directed learning. Here, I will argue that re-visiting our understanding of the shape of the learning curve and its underlying cognitive drivers is essential for uncovering its neural basis. Rather than thinking about learning as either ‘slow’ or ‘sudden’, I will argue that learning is better interpreted as a combination of the two. I will provide behavioral evidence that goal-directed learning can be dissociated into two parallel processes: knowledge acquisition which is rapid with step-like improvements and behavioral expression, which is slower and more variable, with animals exhibiting rudimentary forms of hypothesis testing. This behavioral approach has allowed us to isolate the associative (knowledge-related) and non-associative (performance-related) components that influence learning. I will present probabilistic optogenetic and longitudinal two-photon imaging results that neural dynamics in the auditory cortex are crucial for auditory guided, goal-directed learning. Conjoint representations of sensory and non-sensory variables in the same auditory cortical network evolve in a structured and dynamic manner, actively integrating multimodal signals via dissociable neural ensembles. Our data suggest that the sensory cortex is an associative engine with the cortical network shifting from being largely stimulus-driven to one that is optimized for behavioral needs.