Agentic AI is a concept that describes artificial intelligence systems designed to act with autonomy, making decisions and pursuing goals without continuous external direction. The term "agentic" originated in psychology, where Albert Bandura used it to describe self-organizing, proactive human behavior. The AI research community adopted the term in the 2010s to describe software systems that exhibit analogous qualities like initiative and independent goal pursuit. This evolution solves the problem of purely reactive AI, moving beyond content creation to enable goal-driven, autonomous task completion.
The mechanism of Agentic AI works through a continuous cycle of understanding a goal, breaking it into actionable steps, and evaluating the results. This process involves key components like Sensing (gathering information), Reasoning (processing data, often using a Large Language Model or LLM), and Planning (formulating action sequences). A crucial part of its function is the use of feedback mechanisms—often called reflection or refinement—to assess if its outputs align with expectations and adjust its approach, making the system self-correcting.
Agentic AI connects to and builds upon Generative AI (GenAI), but differs by shifting from content generation to autonomous action. It combines technologies like Large Language Models (LLMs), Natural Language Processing (NLP), and Machine Learning (ML) to understand intent and execute multi-step workflows. The concept has changed recently, with the period of roughly 2024–2025 marking a shift where LLMs were transformed into workflow engines capable of structured reasoning and tool use. The core idea of agency—the capacity to act independently toward a goal—remains the constant foundation.