Agent-based Hybrid AI Models and Technologies: A systematic Literature Review

Personalized hybrid agent-based systems leverage data-driven and symbolic components to provide tailored, context-aware decision support in multiple domains. Yet, the field lacks a consolidated and evidence-based overview of current approaches, their maturity, and the open challenges they present. Method. This study presents a Systematic Literature Review (2018-2025) combining Kitchenham's protocol with a Goal-Question-Metric (GQM) framework. Searches in peer-reviewed and indexed repositories (e.g., Scopus, Web of Science, and IEEE Xplore) returned 9,733 records, reduced to 46 primary studies after an initial screening. The main research question targets how personalized, agent-based Hybrid AI are currently conceived and implemented. In particular, the study is organized around 10 Structured Research Questions (SRQs) focusing on demographics, abstraction, domains, objectives, users, hybridization, technologies, advantages, limitations, evaluation, and future challenges. Results. Three dominant integration strategies surfaced: (1) concatenated pipelines that serially couple ML outputs to rule engines; (2) shared-representation models that embed symbolic knowledge in neural architectures; and (3) agent-level orchestration where heterogeneous components interact via message passing. While recommendation and adaptive coaching are the main use-cases, 71% of contributions remain at the prototype-level and lack large-scale or longitudinal evaluations. Technical barriers include scalability, semantic interoperability, and explainability; only 15% of studies report user-centered validation. Conclusions. The review reveals a growing but fragmented landscape. We propose a research roadmap calling for: (i) publicly available benchmark datasets for hybrid personalization, (ii) standardized Hybrid Artificial Intelligence protocols for agent-to-human and agent-to-agent explanations, and (iii) design guidelines that combine symbolic guarantees with data-driven adaptability. Addressing these gaps is essential for a trustworthy deployment of Hybrid AI in sensitive settings.