Artificial intelligence is increasingly embedded in task allocation, evaluation, knowledge production, decision support, and employment administration. Its organizational significance therefore extends beyond automation: it changes who may decide, who must justify decisions, what expertise remains consequential, and where responsibility is located. This integrative review examines artificial intelligence as work-design infrastructure by connecting evidence on employee autonomy, skill transformation and dependency, algorithmic control, and accountability across human and artificial actors. A transparent integrative-review approach combines theoretical, qualitative, mixed-method, field, experimental, and methodological scholarship while preserving differences in design, context, and level of analysis. The evidence suggests that artificial intelligence can simultaneously expand and restrict agency. It may increase access to information, flexibility, and task support while narrowing substantive discretion through opaque allocation, monitoring, evaluation, and recommendation systems. Short-term performance gains may coexist with altered learning opportunities, dependence on system-encoded knowledge, and weakened capacity to question outputs. Accountability remains organizational because design, deployment, oversight, and appeal arrangements shape the authority given to artificial systems. The review proposes an Agency-Redistribution Framework that organizes delegation, retained discretion, control exposure, skill development, responsibility transfer, contestability, and feedback over time. The framework is an original synthesis rather than a validated causal model. Its implications are conditional on task uncertainty, worker dependence, expertise, interpretability, institutional context, and implementation practice. Longitudinal, multilevel, comparative, and intervention research is required before general claims about organizational effectiveness or responsible deployment can be sustained.
