In the last months, I’ve been hearing about Agentic-AI. The hype is enormous. I decided to delve deeper and understand what exactly Agentic-AI is and its relevance in today’s world.
As a first try, I went to Wikipedia, and found a very clear definition of Agentic-AI:
Agentic AI is a class of artificial intelligence that focuses on autonomous systems that can make decisions and perform tasks without human intervention.
Thus, agentic AI is a program that uses AI techniques to make a system autonomous. These systems can make decisions and perform tasks without human intervention. After reading this definition, I was reminded of my classes at the university about 25 years ago. Back then, we talked a lot about agents and intelligent agents in my artificial intelligence classes.
Wikipedia also has the definition of Intelligent Agent. It says:
An intelligent agent is an entity that perceives its environment, takes actions autonomously to achieve goals, and may improve its performance.
Can you spot the difference? No, there isn’t. We’re talking about a concept that’s more than 50 years old for which a lot of work has already been done. Take a look at this document from 1980, for example, where the concept of an intelligent agent was introduced. There are hundreds of papers about intelligent agents. Yet, companies have only started looking into this in recent years.
And why? Because of LLMs and how easily one can ask an LLM to perform a task. However, what I’ve seen in the industry is not an intelligent agent. We’re forgetting the “autonomous” part. We still need to take action and click buttons to start tasks.
To be autonomous, we need accuracy, precision, and recall. We also need solutions that do not hallucinate. Thus, my main complaint about this new trend in agents is the use of LLMs for reasoning when we have many other well-known algorithms that have been proven to make better decisions. Even rule systems written in Prolog and Lisp decades ago in expert systems would be more accurate and precise than LLMs.