Is it AI or Agentic AI? Why it matters.

Greg Heffner Sept 5th, 2025

In the new age of working with all things AI and ways we can make our lives easier, we have come across a fork in a road that has drastically different outcomes based on the path that is taken.

When AI first started impacting engineers we saw quick responses to questions with a pretty large margin for error. We could ask models questions and for the most part we would get good feedback but it left more to be desired in most cases.

AI

In this example, you can see I asked AI to create me a simple flask application. Legacy AI works well but one step at a time. It can make determinations but only based on the information it has in front of it. While it made a very simple application in the app.py file correctly, the app isn't really created until all dependencies and prerequisites are in order and the script can be run. When AI finished, I immediately tried to run the script and the script failed. I had to first go into the terminal and create a virtual environment for my package installation, then after I had to remember to install flask and the correct command to install it, I was able to go back and run the script. Remember it created the application but thats as far as it can think. It cant also install packages. You can see the validation response I get in the new window. "Hello, flask!" showing not only the server was running but the values as we expected.

While AI is very handy when troubleshooting small problems, it's not very useful when you need more than just a little support.

Me: Hello, I would love a cookie.
AI: Here's a chocolate chip cookie.

Did I get what I wanted, yes. Did it include things I forgot to include, no. Will it work, Not really.

Agentic AI

In this example you see I have the very same file foundation. The difference is I am asking the Agentic AI agent (shown in the bottom right) to create a flask app. Once asked, the agent immediately thinks about what the goal is, it looks into the environment it is running in then makes a determination on what should be done. In this case, just like before, the script needs to be written and then in a virtual environment the dependent package of flask needs to be installed. This time the major difference we see is the agent is doing all of this (with our validation) and the only thing needed from us is to run the script. "Hello, flask!" it says.

Me: Hello, I would love a cookie.
Agentic AI: I read logs about you having a gluten allergy from previous requests. By looking through all the ingredients in the pantry, I was able to find gluten free flour and confirmed your allergy. Here's a warm gluten free chocolate chip cookie just for you with some milk. I know they are best served warm from my quick internet reference. If you would like a room temperature cookie I also can do that, just let me know.

Did I get what I wanted, yes. Did it include things I forgot to include, yes. Will it work, Yes.

The power of agentic AI is the ability to look at problems holistically. Initially you were limited to flat files or directories and AI had a hard time spanning across multiple references to make informed decisions. This led to scripts that worked sometimes or weird bugs that were very hard to find. Now, we are starting to get to a place where we can ask AI something, it not only provides a solution but is able to install the needed dependencies and test its solution before giving us an answer.

This is a very trivial example of how agentic AI can help but I wanted to be able to give an example that everyone could rationalize with the understanding that you can swap out flask apps for literally anything. Agentic AI to help with chat apps, troubleshooting large datasets, protein folding, etc...

I am looking forward to the future of agentic AI. It is going to shorten outage durations saving time and money for businesses/teams/individuals, it will reduce scripting fatigue allowing people to focus on the outcome and validations, and it could get to the point where agentic AI is able to react in real time with predefined playbooks like being able to respond to adversary attacks and manipulations.

It's more than just using AI to help get you from point A to point B. It's using AI to get you from A to B but you have a smart buddy along for the ride telling you where all the cops are.

When you hear anyone saying something about vibe-coding you can thank agentic AI. It was really easy for people to ask AI to create a single script for things but to be able to make a fully working program is game changing. Next up physical AI. Industrial use cases like what Agility Robots is doing or Autonomous travel solution like what Nuro is doing. Check out this link for more info.