Agentic AI or Autonomous Agents - A No Jargon Explanation

One of my programming mentors at one of my earliest jobs was Rick (last name hidden). When I was somewhat of a newb at Java programming, Rick and I used to do some pair programming. Every time we started to work on a solution to a problem or complete a task. I would ask Rick, so, where do you think I should start? Rick would say “well, there are multiple ways to skin a cat” ??

He would encourage me to whiteboard a flow chart of the blocks of code I would have? to write to complete that task.

Agentic AI echoes the sentiments of that interaction. The sentiment that there may be many ways to solve a problem. At the time of this writing, Agentic AI is a system that achieves a goal by making one or more micro decisions and taking steps in a direction that gets you closer to the goal, eventually accomplishing it. This Agent because of AI is very different from automation such as BPA (Business Process Automation) where the steps are usually pre-determined. [I will cover the technical details of? this in my upcoming blog in Substack so consider subscribing to my substack]

To give you an example, every year as a business owner I am requested by my accountant to submit my mileage report for tax purposes. The miles driven should not include any of my commute to the place of my work but can include business meetings, events, seminars, networking meetups etc. To get this report done, as Rick would say there are many ways to skin that cat. At the end of the day, the objective is to produce that mileage report for the whole year.

When we consider that mileage report as the objective. We could potentially give that as an assignment to an AI Agent. The way you would do that is write that objective in plain English as you would email an Executive Assistant. The instructions may be something like.

• Look at my last year’s business calendar and extract all events that are not zoom or phone calls
• Exclude events that have my office as the location
• From that list, for each event, calculate the distance from my home address to? the? event location
• Create an excel sheet report with the event name, my home address, the event location and miles driven

Just as an Executive Assistant is able to understand the email and perform actions. The Agent that is given these instructions is capable of making several micro decisions to accomplish this task. Being able to make those micro decisions is what makes it Artificially Intelligent. For example, it may decide to scrape your calendar to extract the events?

Or decide to generate code to use the google calendar API (I know, pardon this one Jargon) to pull events from your calendar.

To summarize, an Agentic AI system is able to take textual instructions that communicate a goal, understand the instruction, plan the steps needed to accomplish the goal, access the required information/data by using tools, chain the entire execution of steps and to complete each step.?

To say it succinctly with a bullet list.? The characteristics of an Agentic AI or Autonomous Agent? are.

• Understand Textual user instructions or inputs
• Plan the steps to accomplish the goal by making micro decisions
• Access the required tools to complete steps
• Execute steps in a chain of tasks and evaluate completion

Now, for my tech innovator friends let’s take this a?step further to understand what the Agentic AI systems are made of. What are the components of an Agentic AI?? There are lots of rapid developments that are happening in terms of what components should be included to make it an Agentic AI.

At the time of this writing, the key components are

1. Language Models or LLMs
2. Orchestration System - Chain tasks in sequence for execution [Eg: Langchain, Llamaindex, Crew AI]
3. Tool Runtime - Access external systems to get information/data to perform tasks
4. Instruction Medium - Chat or text window or UI for user instructions to trigger the agent with the goal.

Since the blog is a No Jargon explanation of Agentic AI, I will stop here. The key takeaway is that Agentic AI systems are capable of making micro decisions and may not always take a pre-determined path to accomplish a goal.??

We can now imagine the future of Agentic AI as potentially having many many Agents working within a business augmenting the workforce.?

The agents that are built act, not just for you but as you.

If you are curious about Agentic AI and want to see Agents in action. For all the Denver AI Innovators, Technologists and Engineers, Attend Our Hands-On AI Labs Meetup!?

This upcoming Tuesday the 19th at 5.30pm.? Register here

https://www.meetup.com/meetup-group-zpqvmxup/events/304518353/?utm_medium=referral&utm_campaign=share-btn_savedevents_share_modal&utm_source=link