From Software Agents to Agentic AI

From Software Agents to Agentic AI

Before completing my formal education in computer science, where I studied computing principles and programming in depth, I had the opportunity to research available technologies and computing power for a tank-farm automation project. During this research, I encountered a company specializing in industrial automation, which offered me a job even before I graduated. With the support of top management, I became a software architect and contributed to the development of an enterprise-level SCADA (Supervisory Control and Data Acquisition) software architecture.

I vividly recall our CEO's vision of pushing the boundaries of expert SCADA systems and redefining how they function. He often spoke about an enterprise-level software architecture and a platform designed to manage and optimize corporate information workflows. Unfortunately, at the time, our development teams struggled to grasp this ambitious concept. However, about two years after he first introduced the idea, Lotus Software, founded by Mitchell Kapor and Jonathan Sachs, launched Lotus Notes. Later acquired by IBM in 1995, Lotus Notes embodied many of the principles our CEO, Dr. Hari Gunasinghem, had been trying to explain to us. I still remember him pointing at Lotus Notes and saying, "This is the foundation of what I was envisioning—plus a little more."

Despite the initial challenges in understanding his vision, we managed to adopt some of his concepts, such as building software agents that could autonomously traverse enterprise networks, perform tasks, and achieve goals. This went beyond simple rule-based automation—it introduced goal-oriented autonomous processing. Dr. Hari’s idea was to create a document management platform where a software agent could be assigned to a task, similar to how a person is entrusted with handling a client request or project proposal. This agent would then interact with various stakeholders, obtain approvals, and execute necessary actions.

We successfully implemented this idea in SCADA and building automation systems. The Intelligent Building Management System (IBMS) we developed was designed to integrate with document management systems, respond to meeting requests, and facilitate corporate resource management. Our efforts led to the creation of an agent-based software architecture, which ultimately earned our team a U.S. patent.

Agentic AI: An Old Concept with a Modern Facelift

The concept of Agentic AI is not new; it has been evolving since the early days of artificial intelligence. In the past, AI primarily focused on task-oriented workflow management and intelligent processing—elements that were part of our invention. However, early intelligent software lacked the ability to learn autonomously, process natural language efficiently, or apply learned reasoning. Our systems could spawn software agents and deploy them, but they did not incorporate advanced reasoning or self-learning, but natural language processing (NLP).

Understanding AI "Reasoning"

AI does not "reason" like humans. Instead, it processes input by recognizing patterns learned during training. This internal process—often conceptualized as a "chain-of-thought"—allows AI to predict the next logical step in a sequence, leading to coherent and contextually relevant outputs.

• What It Does: AI generates text or decisions using statistical associations and probability assessments, which create the illusion of reasoning but are actually complex predictive models.

• What is NLP? Natural Language Processing (NLP) is a subset of AI that enables computers to understand, interpret, and generate human language by combining linguistics with machine learning techniques.

The Rise of Generative AI

As AI evolved over the decades, advancements in NLP led to the emergence of Generative AI—a significant leap forward in artificial intelligence. The development of models like ChatGPT and DeepSeek highlighted how AI could not only understand conversations but also generate meaningful, human-like responses.

What is Generative AI?

Generative AI creates new content—text, images, music, videos, and even code—based on the data it has been trained on. Unlike traditional AI, which primarily analyzes or classifies existing data, Generative AI produces entirely new outputs that mimic human creativity.

The Shift to Agentic AI

The industry is now shifting towards Agentic AI, despite the fact that agent software and goal-oriented AI have existed for decades. The key difference today is that Agentic AI now integrates NLP and Generative AI as core tools, enabling more sophisticated autonomous agents.

Generative AI vs. Agentic AI

Feature	Generative AI	Agentic AI
Function	Generates text, images, music, etc.	Acts autonomously to achieve goals
Primary Capability	Content creation	Workflow and decision management
Example	ChatGPT generating an article	AI autonomously managing a corporate workflow
Key Components	NLP, deep learning	Generative AI + Workflow automation + Decision-making

Why Agentic AI Matters

Agentic AI is more than just a reactive text generator. It is designed to function as a self-directed system capable of managing workflows, making decisions, and achieving complex goals. By building on the foundation of Generative AI, it expands AI’s capabilities into autonomous task management and rule-based decision-making.

So What it means?

Some might argue that agentic AI is like old wine in a new bottle—an old idea revived with a facelift, now equipped with powerful tools that deliver tangible, goal-oriented results, continuously improving through self-learning

While Generative AI excels in content creation, Agentic AI represents the next evolution of AI—an era where AI can act independently, manage tasks, and drive intelligent automation. The concept of software agents, which we worked on decades ago, has now resurfaced with a modern facelift, incorporating self-learning, NLP, and advanced AI reasoning to create systems that can think, plan, and act autonomously.

Written by : Sanjaya Gunasiri

Copyright © 2023 Pragmatic Engineering. All rights reserved.