Agency Redistribution Dataset: Task-level coding of generative AI delegation in B2B commercial work

Description

Generative artificial intelligence (GenAI) is reshaping professional cognitive work, yet existing research has largely assessed its effects through aggregate productivity and performance outcomes. Less is known about how GenAI redistributes the constituent elements of professional agency between humans and machines. This study develops a task-level framework for analysing and measuring this redistribution. The proposed Conceptual Model of Agency Redistribution distinguishes three redistributable components—decision, execution, and judgment—and a transversal accountability component that remains anchored in the human profession and is operationalised through supervision. The model is translated into a coding instrument capturing delegation degree, decision structuredness, supervision strategy, task weight, and error criticality. An illustrative application to 236 tasks across 26 professional roles and five functional areas in a large B2B technology organisation demonstrates the analytical potential of the framework. The results show deeper AI delegation in execution than in judgment, substantial penetration into semi-structured decisions, and a subset of tasks characterised by high delegation, high criticality, and weak preventive supervision. We define this configuration as the accountability risk zone, where the divergence between operational control and human accountability is greatest. The study contributes a theoretically grounded and operational framework for examining Human–AI Agency Redistribution and establishes a measurement architecture for subsequent validation across organisations, occupations, and sectors.

Institutions

• Universitat de València

  Valencia, Valencia
• Universitat Politècnica de València

  Valencia, Valencia

Categories

Licence