Concept of Agentic AI, explaining what agents are, how they work and the role of tools

An AI agent is a system that doesn’t involve inherently in text to text response generations but interacts with it’s environment to achieve certain objectives - this is possible by giving access to internet to agents that allows them to surf through the internet.

Any AI agent has 3 main components:

1. Perception: Taking in information from the world

2. Reasoning & Planning: Deciding the sequence of steps needed to follow to achieve goal

3. Action: Executing those steps ensuring completion and calling next steps when needed.

   An LLM - Large Language Model can generate text but it has no real world knowledge after it’s training cutoff. By connecting it to tools we give it its arms and legs being able to act efficiently.

   One should note that this is very different than RAG - Retrieval Augmented Generation System - it only allows the LLM to access latest - up-to-date information and provide answers accordingly.

eg) if a user asks about the number of moons of jupiter - RAG system would be activated but if a user asks the agentic ai workflow to subscribe to a channel - RAG system won’t work. Therein, a AI agentic workflow would be required.

Agentic AI represents a natural evolution of language models — from text generators to goal-driven autonomous systems.