The proliferation of generative models and AI agents is transforming the architecture of managerial processes by changing the distribution of functions, authority, and responsibility within management decision support systems. This transformation raises the research problem of ensuring well-founded delegation of authority to AI agents while preserving human agency and autonomy in decision-making. Despite extensive research on algorithmic management and AI ethics, management science still lacks a methodological framework for determining the permissible level of AI-agent autonomy with due regard to the characteristics of managerial tasks and the organizational and technological environment. The study aims to develop a conceptual model of human–agent interaction in decision support systems that provide a basis for delegating authority to an AI agent and determining its level of autonomy. The study employs a systems approach, comparative analysis of existing models of intelligent-system autonomy, content analysis of the academic literature, conceptual modelling, and logical and structural synthesis. Its methodological foundation is an interdisciplinary approach integrating decision theory, humanistic and algorithmic management, organizational theory, and applied AI ethics. Drawing on the humanistic theory of organization, the study substantiates the distinction between AI-agent autonomy and human autonomy. On this basis, a human autonomy – AI-agent autonomy matrix is developed. It identifies four types of human-agent interaction and specifies management mechanisms for preserving human autonomy within each type. The study formulates methodological principles of responsible interaction and delegation of authority to an AI agent, establishing the priority of human professional judgement in decision-making. It also defines the components of the conceptual model of human–agent interaction and the relationships among them. The model incorporates mechanisms for monitoring compliance with the conditions of delegation and providing feedback. An AI-agent autonomy scale is developed to reflect the extent of the agent’s involvement in managerial decision-making. A procedure for determining the permissible level of autonomy is formalized by aggregating expert assessments of the parameters of the managerial task and the technological maturity of the AI agent. The scientific novelty of the study lies in the theoretical substantiation of dynamic and context-dependent delegation of managerial decision-making authority to an AI agent. AI-agent autonomy is treated as a manageable parameter determined by task characteristics and the organizational and technological environment. Human autonomy is regarded as a humanistic value and as the basis for distinguishing the functional autonomy of an AI agent from human moral agency. Humans retain priority in exercising control and making the final decision. The findings provide a methodological basis for substantiating the delegation of authority to AI agents across different types of managerial tasks. They may support the responsible integration of AI into managerial processes and the development of internal regulations and codes of ethics. Further research should empirically validate the findings in different organizational contexts and elaborate the mechanisms for human control and the preservation of human autonomy in hybrid decision support systems.