The interaction modalities between human decision-makers and AI systems are currently under-specified, lacking clear definitions of mutual expectations and role allocation. While decision-making topologies are becoming increasingly complex—ranging from direct Human-AI dyads (e.g., User(H)-to-tool(AI)) to extended, heterogeneous sequences involving multiple agents (e.g., Trainer(H)-to-learner(AI), Trainer(H)-to-learner(H)-to-tool(AI), or Trainer(H)-to-learner(AI)-to-tool(AI))—system designs fail to account for these nuances. Critically, while the sequence always starts with a single human decision-maker, the variable number of embedded learners, users, and AI tools in these chains complicates the flow of information. This interaction ambiguity contributes to decision-makers placing inappropriate levels of trust in AI outputs, as the systems are not designed to support the transparency, feedback loops, or retrospective analysis required for human learning and effective decision oversight.