Artificial intelligence agents and agentic AI systems: Architectures, capabilities, applications, challenges, and future directions

The development of the Agentic AI and Artificial Intelligence Agents has reshaped the traditional experience of artificial intelligence as the passive prediction systems into the Autonomous Intelligent System. The issue discussed in this literature review is the disjointed concept of Agentic Architecture, Multi-Agent Systems and LLM-based Agents and as such has led to inconsistent design practice, risks to security and the lack of scalability to real-life applications. Recent progress indicates that agent-based systems support goal-oriented decision making, the use of tools, and joint problem solving, making Agentic AI an enabling technology of future digital ecosystems. This paper adheres to the PRISMA to conduct a systematic review of the recent studies on AI Orchestration, Distributed AI, Cognitive Architectures, AI Automation and Human-AI Collaboration. The emerging ideas, like Hybrid Neuro-Symbolic AI, Explainable AI, AI Governance, and AI Safety, are also analyzed as the new requirements to the trustful use of autonomous agents. It is found in the review that current Agentic AI Systems are based on layered architectures that incorporate perception, reasoning, memory, and action modules that are typically orchestrated by Multi-Agent Systems and workflow orchestration mechanisms. Its uses include healthcare, finance, robotics, cybersecurity, education, and enterprise automation and major challenges such as interoperability, reliability, ethical adherence and vulnerability of a system to security in highly autonomous settings. The results suggest that scalable Agentic Architecture, standardized communication frameworks, solid governance strategies, and hybrid reasoning systems that integrate symbolic and neural intelligence represent the future of Next-Generation AI Systems and can be used to provide safe and reliable large-scale implementations of autonomous AI ecosystems.