Important decisions often involve choosing between complex environments that define future item encounters. Despite its importance for adaptive behaviour and distinct computational challenges, decision-making research primarily focuses on item choice ignoring environment choice altogether. Here we contrast previously studied item choice in ventromedial prefrontal cortex with lateral frontopolar cortex (FPl) linked to environment choice. Furthermore, we propose a mechanism for how FPl decomposes and represents complex environments during decision-making. Specifically, we trained a choice-optimized, brain-naive convolutional neural network (CNN) and compared predicted CNN activation with actual FPl activity. We showed the high dimensional FPl activity decomposes environment features to represent the complexity of an environment to make such choice possible. Moreover, FPl functionally connects with posterior cingulate cortex for guiding environment choice. Further probing FPl’s computation revealed a parallel processing mechanism in extracting multiple environment features.