Agency theory in the context of using AI agents in financial intermediation
Purpose : to provide a theoretical and applied justification for the transformation of agency relationships in financial intermediation in the context of the AI agents use. Methods : the study utilized theoretical, comparative, and structural-analytical approaches, enabling a comparison of the agency theory tenets with the integration of AI in financial intermediation. Classic tenets of the agency theory are consistently correlated with the conditions for the transfer of functions to algorithmic systems. Results : it was established that with the AI integration, agency costs do not disappear, but their source changes. It was found that in algorithmic environment, residual losses can no longer be explained solely by the divergence of interests between the principal and the agent, as system behaviour is determined by training parameters, data quality, the chosen goal, and the possibility of subsequent verification of the result. A framework is proposed that allows distinguishing the sources of residual losses and observing how they develop at successive stages, using credit scoring as an example. It is also shown that the principal's position in such conditions changes, as the bank's past actions, embedded in the training dataset, themselves become a source of future deviations. Conclusions and Relevance : the study demonstrates that the use of AI agents requires a refinement of the agency theory as it applies to algorithmic agents, since the source of costs shifts from the personal interest to the training conditions, data composition, suitability of the previously trained system, and the ability to explain the program's actions. A comparison of coordinated and swarm-based AI agent organizations reveals that the difference between them lies not in the composition of agency losses, but in the way they arise, accumulate, distribute among digital agents, and are subsequently explained. The results of this study can be used in further research on AI agents in financial intermediation, as well as in the development of procedures for implementing algorithms in banking practices.