Agentic AI in Enterprise Systems: A Systematic Review of Human-in-the-Loop, Governance, and Autonomous Workflow Control

The convergence of advanced capabilities of large language models, coordinated interactions of multiple AI agents, and enterprise-level applications created a basis for emergence of novel systems known as agentic AI. In contrast with typical decision-making support systems, agentic AI systems are capable of performing autonomous goal-directed reasoning, action planning, and execution, minimizing human supervision required throughout the process. This paper conducts a systematic review of current state of art related to agentic AI applied to enterprise software. Three key dimensions of agentic AI are analyzed, namely the roles and characteristics of human-in-the-loop mechanisms as part of system governance, approaches to autonomous workflow control, and challenges in ensuring security and trustworthiness of such systems. Based on nineteen peer-reviewed articles published between 2022 and 2026, several patterns were detected in relation to agentic AI systems operating in enterprise resource planning, legal services, healthcare, retail industry, network automation, and e-commerce sectors. Key trends include a rapid proliferation of new agentic AI frameworks coupled with their limited maturity regarding memory management, adaptability, governance, and hardening of security. Architectural aspects involving human-in-the-loop and human-on-the-loop have been revealed to be crucial but implemented inconsistently. Finally, major gaps in adaptability engineering, compliance and security governance, and cross-agent trust management were identified. Future research should focus on developing metacognitive architectures, resilience assessment methods, and governance frameworks aligning with regulation requirements.