A Scientometric and PCA-Based Structural Mapping Study Identifying Research Fragmentation and Gaps in Agentic AI

Although Agentic AI has emerged rapidly as a distinct, goal-pursuing paradigm within artificial intelligence, its scholarship remains fragmented and conceptually unsettled. Most extant literature conflates functional AI agents with system-level Agentic AI and relies heavily on qualitative surveys lacking empirical structural validation. To address this gap, this study develops a data-driven scientometric and structural mapping framework using Principal Component Analysis (PCA). Analyzing $N = 170$ peer-reviewed publications indexed in Web of Science and Scopus between 2024 and 2026, text data across titles, abstracts, and keywords were standardized and coded into five primary themes: Technology & Architecture (T1), Ethics & Trust (T2), Governance & Policy (T3), Human–AI Interaction (T4), and Applications & Domains (T5). PCA reveals that three principal components capture 98.17% of the total variance. The thematic space is governed by two prominent orthogonal bipolar axes: PC1 (43.33% variance) contrasts deployment and interaction (T4, T5) with normative and policy dimensions (T2, T3), while PC2 (37.98% variance) sharply opposes core technical architectures (T1) against ethical governance (T2, T3). These findings expose systemic structural fragmentation, demonstrating that Agentic AI is predominantly investigated as isolated technical or normative problems rather than an integrated socio-technical system. By pinpointing critical research gaps at the intersection of agentic architecture, oversight mechanisms, and institutional accountability, this paper proposes a future research agenda centered on governance-aware agent architectures and cross-disciplinary research designs.