Artificial intelligence is increasingly embedded in organizational tasks that once depended primarily on human judgment, memory, interpretation, and discretion. Yet describing work as AI assisted says little about who retains meaningful authority, which cognitive operations are delegated, or who remains answerable when consequential outcomes follow. This Original Propositional Model Article develops an evidence-informed account of those distinctions. It integrates work-design, agency, delegation, algorithmic-control, judgment, cognitive-offloading, and responsibility scholarship to propose an Assisted-Autonomy Model organized around four linked allocation processes: authority allocation, judgment allocation, cognitive allocation, and responsibility alignment. The model distinguishes assistance from judgment substitution, routine or execution offloading from evaluative-reasoning offloading, and formal accountability from practical evaluative control. It further introduces retained evaluative agency and autonomy–responsibility decoupling as proposed constructs for explaining why workers may remain formally responsible even when their capacity to interrogate, reject, or reformulate AI-generated judgments has narrowed. The framework also separates technological capability from legitimate organizational authority and nominal human presence from consequential oversight. The article argues that autonomy effects are conditional on task verifiability, relative human–AI capability, opacity, workload, organizational control, and meaningful opportunities for intervention. The model and its propositions are not presented as validated causal relationships or universal prescriptions. Their contribution is to specify falsifiable mechanisms, rival explanations, and empirical boundaries for research on AI-assisted work.
