Agentic AI Across Domains: A Comprehensive Review of Capabilities, Applications, and Future Directions

Abstract

The emergence of large language model-based agents represents a transformative shift in artificial intelligence, enabling autonomous systems capable of perceiving environments, reasoning about complex tasks, and executing multi-step actions. This comprehensive review examines fundamental capabilities underpinning agentic AI, analyzes cross-domain applications spanning software engineering, scientific discovery, and healthcare, and identifies critical technical challenges. Through systematic analysis of recent advances, we establish a unified framework encompassing perception, reasoning, and execution while documenting performance metrics. The synthesis reveals persistent challenges in reliability, evaluation methodology, and safety governance requiring coordinated research efforts. Our findings indicate that while agents demonstrate remarkable capabilities in constrained domains, achieving robust autonomy demands fundamental innovations in coordination, reasoning, and decision-making protocols.