Systems are currently available across a range of autonomy levels, though many may never be fully autonomous (i.e., Level 5, Society of Automotive Engineers International). Consequently, autonomous systems will require human supervisory control & resource management, further requiring designed artifacts for interaction. Connecting human operator input to discrete operations used by autonomous systems enables levels of automation that allow for cooperation and performance augmentation [23, 24] and can facilitate analyses to improve operator efficiency (e.g., [12]). Here, we introduce methods for characterizing complex, goal-driven behavior and determining critical paths that optimize interactions between human operators and artifacts for the supervisory control of autonomous systems. We propose applying cognitive task analyses (CTA) and task activity networks to discretize sequential behavior and conditional generative modeling [2, 15] to identify and evaluate critical paths to offer potential improvements to human-autonomy interactions.