International Journal of Science, Engineering and TechnologyAuthors: Atharva Pisal, Soham Pawar, Revan Chenna, Swaraj Dalvi
Abstract: Agentic AI represents a paradigm shift in which intelligent systems autonomously reason, plan, and act within dynamic environments to achieve complex goals with minimal human intervention. This paper presents a comprehensive study of agentic AI architectures integrating large-scale foundation models and multimodal Visual Language Models (VLMs). Key architectural components including the LLM reasoning core, memory modules, planning engines, tool interaction layers, and multi-agent orchestration mechanisms are analyzed in depth. Latency challenges arising from model inference, framework processing, communication overhead, and system inefficiencies are systematically examined, and optimization strategies including model compression, speculative decoding, KV-cache management, and edge deployment are evaluated. Applications spanning indus-trial automation, healthcare, energy systems, smart infrastructure, and financial services demonstrate significant performance im-provements. Open challenges relating to computational complexity, hallucination, interpretability, privacy, and ethical alignment are discussed, followed by future research directions toward efficient, trustworthy, and scalable agentic AI systems.
DOI: https://doi.org/10.5281/zenodo.19878647
