What is Agentic AI?

So I keep hearing this term. Agentic AI. Agentic AI is going to be a significant iteration of innovative leaps in AI. Even NVIDIA is betting on agentic AI. First lets share what types of AI there are, then I can share what agentic AI is and what it’ll do for AI in general.

Types of AI

So when people think “AI” now they are probably most familiar with the generative AI portion of it. But there are a few others that are important to understand. Here’s a list including how you may interact with it. Yes, I used generative AI to help create this list for me.

Type of AI	Description	Use/Product
Generative AI	Creates new content like text, images, music, and video based on training data	ChatGPT, Claude, DALL-E
Predictive AI	Uses historical data to make predictions about future events or outcomes	Netflix watch recommendation, weather predictions
Reactive AI	Makes decisions based solely on current input without memory of past experiences	Bots in a computer game
Limited Memory AI	Uses recent past data along with pre-programmed information to make decisions	Smart thermostat
Natural Language Processing (NLP)	Focuses on understanding and generating human language	Siri, Alexa
Computer Vision	Processes and analyzes visual information from the world	FaceID, security camera alerts
Expert Systems	Emulates decision-making of human experts in specific domains	GPS routing, tax prep recommendations
Robotics AI	Controls physical systems and interacts with the real world	RoboRock vacuum,
Recommendation Systems	Predicts user preferences and makes personalized suggestions	Spotify music, Amazon recommendations
Reinforcement Learning	Learns optimal actions through trial and error	Genetic algorithms, schedule optimizations
Agentic AI	Autonomous systems that can perceive, decide, and act independently in their environment	Salesforce Agentforce, NVIDIA AI Agent Blueprints

It’s quite the list. We’ve had many of these AI’s for decades, some going back to the 50’s. The reason why it’s not been front and center is because it hasn’t been sexy for most people…

If I told you FaceID works by using a Vertical-Cavity Surface Emitting Laser that is read by an infrared camera and uses an advance machine learning neural engine that performs real-time biometric authentication to analyze the depth map from the mathematical model of the users face to let the user in AND has a false positive rate of 1 in 1 million…

Oh, um, I lost you? Shoot. Well on to agentic ai.

Agentic AI Capabilities

Agentic AI is basically a smart assistant, one that can act on your behalf to complete more complex tasks, linked tasks, or tasks that require interaction with multiple services.

How is this different than what generative AI can do?

Agentic AI is an iteration of generative AI. Think of generative AI as a child that knows how to spew information. Is it correct? Who knows! But Bobby said that geologists um eat um say you should uh eat um 1 rock per day and that cheese is um is stuck on pizza with uh nontoxic glue.

With children we just laugh and gently correct them. With AI, we call that hallucinating.

You may have experienced an form of “agentic” AI with chatbots that can make some basic “decisions” from their training data, sometimes its has some good results but most of the time it makes you huff in frustration and mash that 0 button in hopes you can get a real person to resolve your issue.

With agentic AI, once implemented, could allow you to say I want a vacation in Disneyland, the agentic AI will find which days are the best, make decisions based on your likes, dislikes and social expectations, plan arrival and departure based on your calendar and most optimized flight days, and schedule your trip for you, sharing the details with your friends agentic AI that is also scheduling the same vacation, but coming from a different location. The end goal will be sharing the details with you with what it found and you being able to describe or adjust changes without having to interact with UI, you can just describe what you want. Then when you get back, share your likes, dislikes, or adjustments for a future vacation, that may not be at Disney, but the AI can use it to better hone your next vacation.

Or an even more optimistic outlook; traffic management can connect with cars on the road and suggest different routes based on congestion and each cars individual routes, additionally connecting with traffic lights to improve traffic flow. Or even suggest that you stay an additional 20 minutes at the shopping mall to halve your travel time home because of a traffic accident.

Agentic AI Development

This kind of agentic AI requires a different methodology of development. When generative AI came out, many companies basically slapped an API with ChatGPT’s generative AI and created a new product category without doing deep research into how to make it truly improve efficiency be a category to further invest in.

With AI based workflows, I imagine the new formula will be rock solid API documentation and implementation with less effort on the beautification of the product, as some products will only be API integrations or versions of agentic AIs that can integrate with other AIs.

Or better yet, having integrations that can allow for you to make your own dashboard with agentic AIs feeding into the data pool, only showing the most important items for the day, making decisions on what information it shows based on how your day looks.

While this may seem a bit like algorithms controlling your every move, as a social being, a lot of our likes and dislikes are based on our social circles and patterns, so in some ways there is a benefit, but definitely something to look out for.

A long term goal for companies should be how to make agentic AI and all AI share its blindspots algorithmic tendencies and outcome methodology, without sharing the secret sauce behind its algorithms which companies will defend to their last breath.

Final Thoughts

Agentic AI will be huge, and will be the next iteration of AI, being able to pull all the different types of AI, information and tools into one expansive and helpful dashboard productivity center, making much of the stupid, poorly written or confusing UI/UX and putting it behind something humans are good at: talking, explaining, and describing. This is why human call centers are still so ubiquitous since human to computer talk is not a natural skill, and very much a learned skill.

Interfacing with computers have not changed in over 50 years, and AI, even agentic AI will begin the slow transformation of how we interact with computers to something science fiction has only been able to dream of.