In this blog post I want to bring your attention to the evolution of Agent technology in AI. When I started in AI in the 1990’s, intelligent agents were introduced. Intelligent Agents were software systems that can perceive their environment, reason, and act autonomously to achieve goals. This AI technology was developed using various programming languages. Early in my career in AI I learned LISP and Prolog programming languages to build Expert Systems and Intelligent Agents.

Details about Lisp, Prolog and other notable languages include:

Lisp: This language was widely used for artificial intelligence and agent-based programming. Specifically, the AGENT-0 interpreter, developed in the 1990s, was implemented in Lisp, allowing agents to communicate using speech act theory.

Prolog: Prolog was another popular language that I used in the 90’s. This language was developed in the 1970’s and was also used for intelligent agents to do reasoning and problem-solving like humans. However, it is not used much now, it is still a valuable tool in many specific AI areas such as expert systems and natural language processing.

AgentSpeak(L): This language was created in 1996 and is a logic programming language based on the BDI (Belief-Desire-Intention) model and serves as an abstraction of implemented BDI systems. It’s used to interpret agent programs as horn-clause logic programs.

Procedural Reasoning System (PRS): Developed and used from 1980’s though the late 1990’s. PRS is a framework for building intelligent agent real-time reasoning systems capable of handling challenging tasks in dynamic environments. It is the idea of an intelligent agent that uses the belief-desire-intention software model and is rational. In the development of a PRS, a set of knowledge areas is typically defined as a user application and provided to a PRS system.

Knowledge areas mentioned here identify an early connection between AI and Knowledge Management. These two concepts are intertwined, as knowledge areas provide the building blocks for effective knowledge management practices. Also implemented in Lisp, PRS defines a system capable of reasoning about processes and procedures, enabling agents to select intentions for achieving particular goals.

These languages and frameworks were instrumental in developing intelligent agents that could reason, learn, and interact with their environment. Let’s take a look on how Intelligent Agents have evolved into AI Agents and Agentic AI. A brief exploration of the evolution and differences between Intelligent Agents, AI Agents, and Agentic AI follows.

Intelligent Agents (1990s–early 2000s): Reactive to Rule-Based Autonomy

Intelligent agents are software entities designed to perceive their environment, process input, and act autonomously based on a fixed set of rules or heuristics to achieve specific goals. Their intelligence is typically reactive or rule-driven, and their autonomy is tightly bounded by human-defined parameters. Examples include thermostats, simple expert systems, and basic robots that respond to environmental cues. However, they are limited in adaptability, lack contextual awareness, and cannot learn from data.

AI Agents (early 2000s–2010s): Intelligence-Infused Autonomy

AI agents evolved as intelligent agents empowered by artificial intelligence techniques such as machine learning, natural language processing, and symbolic reasoning. They can interpret ambiguous input, learn from interactions, and improve performance over time. Examples include virtual assistants, recommendation engines, and chatbots. While they exhibit more flexibility than traditional intelligent agents, their decision-making is still bounded by human-defined objectives and constraints.

Agentic AI (2020s–present): Emergence of Autonomous Decision-Makers

Agentic AI represents a paradigm shift where systems not only act intelligently but also possess agency—the capacity to independently set goals, plan, adapt, and reflect on their actions. They exhibit high contextual awareness, long-term memory, recursive learning, and can orchestrate multiple tools or agents to complete tasks. Examples include AutoGPT, BabyAGI, and LangChain agents. These systems operate with minimal human intervention and can self-direct toward complex objectives.

Key Differences

1. Autonomy: Intelligent agents can act autonomously within predefined rules, while AI agents and Agentic AI exhibit increasing levels of autonomy, with Agentic AI being more adaptive and self-directed.
2. Complexity: Agentic AI systems are more complex and capable of handling dynamic environments and uncertain situations, whereas intelligent agents and AI agents might be more limited in their scope and adaptability.
3. Goals and Decision-Making: Agentic AI can set its own goals and make decisions based on complex reasoning, whereas intelligent agents and AI agents typically operate within predefined objectives and rules.

Conclusion

The evolution from intelligent agents to Agentic AI reflects advancements in AI research, enabling systems to become more sophisticated, autonomous, and capable of complex decision-making. Agentic AI can act independently and make decisions to achieve goals, without consistent human-in-the-loop (HITL) guidance/oversight. AI Agents will be able to do a lot more and this is just the beginning. However, Agentic AI introduces new challenges in ethical design, emergent behavior, regulatory oversight, and safety. Developers must design systems that align with human values, integrate robust monitoring and explainability, and adopt new governance frameworks. What we have to figure out is how do we incorporate a HITL component(s) to ensure the agents are working properly. Otherwise, Agentic AI could lead to systemic risks, particularly in critical domains like healthcare, legal systems, and national security.