If you’ve recently started exploring Agentic AI, you might think that the idea of an “AI agent” emerged alongside Large Language Models (LLMs) like ChatGPT. In reality, the concept of an agent has been a cornerstone of Artificial Intelligence for decades.
One of the clearest examples comes from Russell and Norvig’s Artificial Intelligence: A Modern Approach (AIMA)—widely regarded as the standard textbook on AI. Interestingly, Chapter 2 of the book is devoted entirely to intelligent agents.
This is significant because AIMA is a general AI textbook, not a book about today’s Agentic AI. The fact that it introduces agents so early tells us something important:
The concept of an agent is fundamental to AI itself—not a recent invention of the LLM era.
The Classical Definition of an Agent
Russell and Norvig define an agent as:
Anything that perceives its environment through sensors and acts upon that environment through actuators.
This definition is intentionally broad. An agent doesn’t have to be a humanoid robot or a chatbot. It can be almost anything capable of perceiving and taking actions.
Examples include:
- A thermostat
- A chess-playing program
- A robot vacuum
- A self-driving car
- A medical diagnosis system
- A human being
Despite their differences, they all share the same basic architecture:
Environment ↓ Sensors ↓ Agent ↓ Actuators ↓Environment
The implementation may differ, but the abstraction remains the same.
Why AI is Built Around Agents
One of the key ideas in classical AI is that intelligence should be described by behavior, not by implementation.
Instead of asking:
“Does this system use neural networks?”
AI asks:
“Can this system perceive its environment and choose actions that help it achieve its goals?”
This shift in perspective makes the agent the fundamental building block of AI.
Whether the underlying technology is symbolic reasoning, search algorithms, expert systems, reinforcement learning, or neural networks is secondary. The important question is how the system perceives, reasons, and acts.
Then What Is Agentic AI?
If agents have existed for decades, why is everyone talking about Agentic AI today?
The answer is simple:
Agentic AI doesn’t introduce the concept of agents—it changes how we build them.
Traditionally, intelligent agents were implemented using techniques such as:
- Rule-based systems
- Search algorithms
- Planning algorithms
- Expert systems
- Reinforcement learning
- Robotics
Modern Agentic AI uses a different toolkit.
Today’s agents are often powered by Large Language Models and augmented with capabilities like:
- Natural language reasoning
- Planning
- Memory
- Tool use
- Retrieval
- Reflection
- Workflow orchestration
The underlying concept hasn’t changed. Only the implementation has.
A Modern LLM Agent
Consider a simple user request:
“Plan my trip to Japan.”
A modern AI agent might:
- Understand the request.
- Ask clarifying questions.
- Search for flights.
- Compare hotel options.
- Look up attractions.
- Generate an itinerary.
- Update the plan based on user feedback.
From Russell and Norvig’s perspective, this is still an intelligent agent.
It perceives information (user input and tool outputs), reasons about what to do next, and acts by calling external tools and producing results.
The difference is that an LLM now serves as the reasoning engine.
Classical AI vs. Agentic AI
A useful way to think about the relationship is:
Classical AI-------------------------Intelligent Agent | | implemented using |+-------------------------+| Rule-based systems || Search algorithms || Planning || Expert systems || Reinforcement Learning || Robotics |+-------------------------+Modern AI-------------------------Intelligent Agent | | implemented using |+-------------------------+| Large Language Models || Tool Calling || Memory || Retrieval || Planning || MCP || Multi-Agent Systems |+-------------------------+
The agent abstraction remains unchanged. The technologies used to implement it have evolved.
Why Agentic AI Feels Revolutionary
The excitement around Agentic AI isn’t because the idea of an agent is new.
It’s because LLMs have dramatically lowered the barrier to building general-purpose software agents.
Before LLMs, creating an intelligent agent capable of understanding arbitrary instructions, reasoning across domains, and interacting with different software systems required extensive hand-crafted logic.
Today, an LLM provides a flexible reasoning interface that can understand natural language, generate plans, invoke tools, and adapt to changing situations.
This has made building capable agents far more practical than ever before.
To Summarise – Agentic AI Not a New Idea
One of the biggest misconceptions in AI today is that Agentic AI represents an entirely new paradigm.
It doesn’t.
The idea of an intelligent agent has been central to AI since the early days of the field and has been taught in foundational textbooks like Russell and Norvig’s Artificial Intelligence: A Modern Approach for decades.
What has changed is not the definition of an agent—it is the technology used to implement one.
In that sense, Agentic AI is best understood not as a replacement for classical AI, but as its modern evolution. Large Language Models have become the reasoning engine that breathes new life into one of AI’s oldest and most enduring ideas.
Technology, Knowledge & Philosophy 
Leave a comment