Abstract: [Purpose/Significance] AI agents, represented by OpenClaw, are transitioning from conversational tools to autonomous actors, posing a challenge to the traditional established Human–Information–Technology tripartite analytical framework in information science. This paper aims to address the core question of "How agents, as novel autonomous actors, will transform information science?" and provide references for the theoretical development and practical response of information science in the era of intelligent agents. [Design/Methodology] Drawing on a review of the existing analytical perspectives and related theories in information science, this study takes OpenClaw, AutoResearch and Harness as analytical subjects, and discusses their impact along four dimensions: information behavior, knowledge production, human-agent role configuration and risk challenges. [Findings/Conclusion] This paper proposes treating AI agents as an independent element to construct a four-element analytical perspective for information science: Human–Agent–Information–Technology. Based on this analytical perspective, it is found that at the theoretical level, the agents pose a challenge to the existing premises of information behavior theory, knowledge organization, intelligence analysis, and information services. At the practical level, the manner in which the agents intervene in scenarios like knowledge production, intelligence analysis, and competitive intelligence has brought about a qualitative transformation in the nature of intelligence work, rather than merely an increase in efficiency. At the risk level, potential hazards such as prompt injection, information hallucination, information leakage, and the dilemma of subjectivity in knowledge production necessitate systematic research. [Originality/Value] This study advances beyond the limitations of prior research that views AI agents merely as auxiliary tools. Introducing intelligent agent as an independent element into the analytical perspective of information science can better reflect the newly evolving paradigm of current Human-on-the-loop.

Keywords: AI agent, OpenClaw, Information science, Human-on-the-loop, Harness