Decision making is a dynamic process by which a person integrates sensory evidence with prior expectations to select an action to achieve a desired outcome. Often, this process requires working memory to hold and manipulate the relevant information. Working memory has several known limitations and models of implementation that have not been widely considered in the context of decision-making. For decisions regarding spatial stimuli, the dorsolateral prefrontal cortex (dlPFC) is likely to be essential. Activity in this area has been shown to relate to both decision making and working memory, and has led to concrete models of how spatial information may be stored by population activity. To investigate how this activity might be leveraged to implement several computational elements of decision making, we performed two related experiments.In the first, we had human participants perform a working memory task that required the reporting of a decision variable (average location) to determine how working memory limitations impacted decision precision. We also used models of working memory to predict and interpret what information was being actively maintained. We report not only the novel finding that decision variables held in working memory lose precision over time but also that the degree of precision loss depends on the strategy used to make the decision, which differed across participants and conditions. In our second study, we trained monkeys to perform an adaptive oculomotor delayed response task in which they had to integrate cue information with context to select the target most likely to be rewarded. The goal of this study was to investigate whether the tuning properties of working-memory related dlPFC neurons adjusted according to the statistics of the task, corresponding to adaptive behavior. We found that not all monkeys display adaptive behavior, but developed method for interrogating the dynamics of neural responses in those that do. This body of work contributes to our understanding of how working memory may implement the representation of a decision variable or prior information used to interpret incoming evidence. Such understanding may ultimately lead to more effective approaches to addressing disorders of maladaptive decisions, such as addiction and PTSD.