Traditionally, business process management has automated work processes in a handcrafted and explicit manner through process models and decision tables. Robotic process automation has enabled agentic lightweight automation with software robots mimicking human users. Despite these technological advances, certain areas have remained off-limits for automation due to their complexity and required cognitive abilities. In this context, intelligent systems based on self-learned analytical models have become increasingly versatile to automate work previously not automatable. With this shift from deterministic to probabilistic automation, supervision and control of automated processes needs to be revisited. That is, a key issue in the development of intelligent automation is not only applying the appropriate level of automation, which ranges from human control to fully automated systems, but also determining how human-machine interaction should be designed in a reproducible and standardized manner. Process patterns provide a means to guide this interaction by ensuring balance between automation efficiency and retention of human control as well as – inversely – machine-in-the-loop safety. We propose 12 process patterns for human-machine interaction based on previously established levels of automation to enable the design of process-aware systems with varying degrees of independence. We have consolidated these levels using an integrative literature review and evaluated the results through an interview study.