Agentic AI refers to systems designed to pursue broad goals autonomously, as opposed to 'passive' AI that simply responds to a user's prompt. An agentic system perceives its environment, reasons about how to solve a problem, breaks it down into a plan of sub-tasks, and executes actions (like browsing the web, writing code, or calling APIs) to achieve the goal. It often includes a self-reflection loop to critique and correct its own work.
Rooted in classical AI planning and Reinforcement Learning, but surged in popularity with LLM-based agents (like AutoGPT and BabyAGI) in early 2023.
A major focus of 2024 AI development, with frameworks like LangChain, CrewAI, and OpenAI's Assistants API making it easier to build autonomous software engineers and research assistants.