Agentic AI
A class of artificial intelligence systems designed to autonomously pursue complex goals and execute multi-step actions (such as software deployment or financial transactions) with minimal human intervention.
Définition
Agentic AI represents the shift from "passive" AI (which generates text or insights) to "active" AI (which interacts with the world). Unlike a standard Large Language Model (LLM) that waits for a prompt to produce an output, an Agentic system has agency: it can reason, plan, use tools (like APIs or browsers), and execute a chain of tasks to achieve an objective. This autonomy introduces new governance challenges, as the risk shifts from what the AI says (content) to what the AI does (actions).
Exemple concret
A travel booking company moves from a Generative AI chatbot (which only suggests hotels) to an Agentic AI system. The Agentic system doesn't just suggest the hotel; it autonomously checks the user's calendar, negotiates the best rate via an API, books the room, and charges the corporate credit card - all without the user clicking a button.
.svg)
