Agentic AI continues to get attention and rightfully so, as it is quickly evolving. Generative AI and agentic AI are different, though. While GenAI is utilized for creating new content (code, images, etc.), agentic AI is focused on making decisions autonomously and, to some degree, with a purpose. When discussing AI, you have GenAI, agentic AI, and AI agents.
Wait, though, isn’t agentic AI just a complex application program? Maybe. Complex? Yes. Comprised of various components/modules? Yes. Rigid with input and specific output? No. Agentic AI is comprised of AI agents, which are considered “worker bees,” but the agentic AI “system” overall is different, as it makes decisions autonomously. That’s a big difference.
Like the various bees in a hive, there are various AI agents, currently categorized into five types (in increasing complexity):
- Simple Reflex Agents
- Model-based Reflex Agents
- Goal-based Agents
- Utility-based Agents
- Learning Agents
Simple Reflex Agents are basic types that are effective in structured environments. They work with predefined rules and do not store information, meaning they can make the same mistakes if their rules are insufficiently defined for their task.
Model-based Agents are more advanced, utilizing an internal model to make decisions. This type of agent can track past states (knowledge) and can use that data for future decisions. Goal-based Agents reach another level of complexity, since they consider their own ultimate objective and can plan their actions and consider different options to achieve the goal. They are limited, though, because they rely on defined strategies to evaluate goals, but they are widely used in robotics and autonomous vehicles.
Utility-based Agents are the type most applicable to complex or dynamic environments, as these agents are designed to consider a range of possible options (outcomes), assigning values to each potential outcome and subsequently determining the best action to take. These are difficult to design because it requires a significant effort to capture the various factors the agent must consider.
Lastly, Learning Agents differ in that they do not rely on predefined models. They continuously update their behavior based on environmental feedback. As you might guess, Learning Agents are highly flexible and designed to work in complex environments. The ability to learn makes them ideal when Natural Language Processing (chatbots, social media, etc.) is required.
Given this breakdown, it is correct to consider a particular agentic AI system as likely being comprised of many of the types listed above. There are additional items to consider, such as how all the agents used work together in the particular system, but detailing those would necessitate a lengthier article. The intent here is to provide readers with a little insight into agentic AI, especially if one or more of their teams are considering developing such a system. Thankfully, there are several tools and resources to aid in creating agents. Thoughtful planning and involvement of key personnel (e.g., data officer, AI/ML officer, enterprise architect, etc.) should be considered as mandatory in such an undertaking.
Dan Kempton is the Sr. IT Advisor at North Carolina Department of Information Technology. An accomplished IT executive with over 35 years of experience, Dan has worked nearly equally in the private sector, including startups and mid-to-large scale companies, and the public sector. His Bachelor’s and Master’s degrees in Computer Science fuel his curiosity about adopting and incorporating technology to reach business goals. His experience spans various technical areas including system architecture and applications. He has served on multiple technology advisory boards, ANSI committees, and he is currently an Adjunct Professor at the Industrial & Systems Engineering school at NC State University. He reports directly to the CIO for North Carolina, providing technical insight and guidance on how emerging technologies could address the state’s challenges.
Leave a Reply
You must be logged in to post a comment.