Multisensory environments require the brain to flexibly integrate or segregate information across sensory modalities to guide efficient behavior. Recent work has sought to clarify the mechanisms underlying multisensory evidence accumulation, integration, and their variability across development and clinical populations. Using combined neural and behavioral measures, evidence indicates that auditory and visual signals are initially accumulated by distinct processes, which can converge sub-additively onto a shared motor decision mechanism depending on task demands. This challenges simplistic accounts of early sensory fusion and highlights the role of decision-level integration. Developmental and clinical findings further demonstrate that multisensory processing is dynamic: while children and adolescents with autism spectrum disorder (ASD) show deficits in basic audiovisual integration, these impairments are not evident in adults, suggesting delayed maturation rather than permanent dysfunction. Computational modeling reveals a developmental transition from competitive to facilitative multisensory interactions that occurs later in ASD and is influenced by sensory weighting and prior cross-modal context. Finally, classic behavioral evidence for multisensory facilitation, such as the redundant signals effect, may be confounded by attentional switching costs in commonly used experimental designs. When modality-switching demands are controlled, apparent multisensory speeding largely reflects unisensory slowing rather than true co-activation. Together, these findings underscore the importance of task structure, developmental stage, and cognitive context in shaping multisensory processing and challenge prevailing assumptions about obligatory multisensory integration.