The Agentic AI Buzz & Simple Ways to Harness It

By Gregg Wright

Agentic AI is the biggest buzzword in the technology and any AI community. In order to utilize and explore the area of Agentic AI, we have to set some parameters on what it is and what is it not.

There are three main types of AI: Non-Generative/Traditional, Generative, and Agentic.

Non-Generative/Traditional

Non-Generative/Traditional AI focuses around the inputting of specific data and information where the tool cannot go and explore options other than what it’s been given.

This type of AI has the most guardrails and structure. Think of when you’re trying to troubleshoot an issue with your internet connection: You end up talking with a bot before you can get to a person. Those bots are non-generative AI. They have been fed with information, questions with answers, policies, procedures, and much more. The bot searches this information to (hopefully) find you what you need to self-service. The downsides are the limitations of the content and that it’s searching this content based only on those parameters.

Another part of Traditional AI is that it’s fully self-created. The people who program these bots and similar functionalities have to program and think of any iteration to a question, spelling, phrasing, and so forth for the tool to try to give a response. If you’ve been frustrated with a bot before, this would be why.

Generative

Generative AI is free to search large amounts of information, internal and external, in order to “predict” what it should give you back.

For example, when you enter a prompt about wanting to find a recipe for chocolate-chip cookies, it’ll give you a recipe it predicts based on its ability to search all the information on the internet for chocolate-chip cookies.

Will the recipe work? It absolutely could, but it’s only predictive. When you’re looking at the recipe, and it says “3 eggs,” that’s based on its search results, whereas a traditional recipe would call for 2 eggs. These “hallucinations” are very common, which is why we need to validate the output of Generative AI to ensure we’re getting the right information. Generative AI never knows what it’s going to create two to five words or items ahead; it’s only looking at what comes next, like each next word, based on its own programming and understanding. It’s extremely helpful in brainstorming and ideation, but for effective execution, there still has to be that human-in-the-loop.

Agentic

Think of Agentic AI as the bridge between Non-Generative/Traditional and Generative AI. Agentic AI uses the same parameters of Traditional, as in you’re usually telling it generally what resources and boundaries it has to have, but you’re giving it a specific goal to meet. It will generate and adjust its information in order to reach that goal, or when the goal is adjusted.

Agentic means “agency” in that you’re giving parameters and boundaries to the tool, but you’re allowing it to have the agency to predict and make decisions around that ultimate goal for the task at hand.

Think about an Excel Macro. They break so easily, but they are built around a process and steps. In a macro, those steps have to be followed exactly—no changes to the data, no additional rows of information or columns—otherwise the whole thing breaks down.

Agentic AI would be a “smart macro” where you still have these parameters and guidelines, but knowing the goal is to take this data and transform it into a dashboard with certain criteria based on the information, the agent will expand and contract the inputs as necessary—without breaking.

Use Cases for Agentic AI

Agentic AI can be a super useful tool in individual and leadership performance support. You can use Generative AI to create the background of certain personas (i.e., an angry customer, high performer, struggling leader, etc.). These personas can then be used by the Agentic side to be an “actor” where you set the goal of “I want you to take on the persona of Bill (here is all of Bill’s persona information from the AI), and the goal of today’s talk will be to help identify some skill growth opportunities. I will be Bill’s manager.”

We have given the Agentic prompt the guidelines and the goal, but everything else requires it to think through and seek to meet our goal, adjusting as our conversation happens. This is a great practicing opportunity for any crucial conversations or even performance support, allowing learners to ideate ahead of time.

Another use case for Agentic AI is around marketing of learning offerings and materials. Give the AI a goal of skills or values you’re trying to build, have it run through your learning library, and it will go through the content, descriptions, titles, etc. and help develop a plan based on the criteria.

This agent can be used over and over again, or set up in a way that it constantly is curating this skill path as things adapt and change. This agent can also draft the communications around these new AI-built programs and send them out to learners, based on employee criteria you’ve given it boundaries for.

AI from A to Z

Don’t miss your chance to explore AI from every angle at DevLearn 2025 Conference & Expo, November 12-14 at the MGM Grand in Las Vegas! Join conversations about Agentic AI, see demos, and participate in hands-on BYOD sessions to try your hand at working with AI, Register today!

Image credit: Vertigo3d