AGENTIC AI - THE RISE OF AUTONOMOUS INTELLIGENT AGENTS IN THE ERA OF LLMS

Agentic AI refers to AI systems that autonomously set and act towards these goals over time. The emergence of large language models (LLMs) has renewed interest in agentic architectures as LLMs are a “brain” that provides human-level reasoning capability for agents. This survey reviews the state of the agentic AI research area. We examine agentic AI’s definition and historical foundations, the theoretical underpinnings of agency, system architectures, and applications. We consider some of the leading LLM-agenting frameworks (Auto-GPT, BabyAGI, LangChain agents) and the essential components that facilitate agency, e.g., memory, planning, tool usage, and feedback mechanisms. We evaluate benchmarks for agents (AgentBench and TheAgentCompany), and outline applications in various domains from software engineering to scientific discovery. We discuss major challenges faced by agentic AI (alignment, control, interpretability, safety), and the potential paths forward, including metrics for agency, dynamic alignment mechanisms, and governance. Throughout, we illustrate how agentic AI takes us further along the journey from narrow AI to general, goal-driven autonomy.