Cognitive processing relies on the brain's ability to balance flexibility for encoding new information with stability for maintaining it. We examined these dynamics in three magnetoencephalography (MEG) datasets of visuospatial working memory (vsWM) tasks. Across all tasks, we identified four distinct networks in the theta and alpha bands, which were used to define functional states. Optimal transitioning rate between states was associated with better cognitive performance. Further, two of the states were linked to flexibility and stability, respectively: an encoding state dominated by a posterior theta and a maintenance state dominated by a dorsal alpha. We simulated the states in an in-silico model with biologically realistic cortical connectivity. The model, featuring spiking and oscillatory cortical layers interacting via phase-amplitude coupling, demonstrated how frequency and spatial region could modulate information flow. Our findings suggest a cognitive control mechanism, where selective transitions between large-scale networks optimize information flow, enabling both stable and flexible visual representations.