Let’s frame our thinking about AI and its progress with some historic references.

AI was developed originally as “expert systems” in the 80s. It was a so called deterministic machine learning approach, where you can with confidence determine that 1+1 = 2.

In 2022 GPT first came out with Large Language Models (LLMs) with the introduction Microsoft GPT.  This rush to generative AI led to the next set of big advancements. This LLM can be characterized as a probabilistic version of AI. For example, ere 1+1 might = 2 99.3% of the time.

Now with the release of agentic AI in 2024 we have AI agents that are able to make decisions. The AI agent is a “decision engine” that can make decisions as it continues to automate.  These agents can learn, they can reason, they can make decisions.

Most of us see the agentic AI through the lens of past models. In other words a one to one relationship. I will have my healthcare agent. It will be one agent for me. I will have my financial advisor agent and it will be another single agent for me.

We should rethink the ratio of agents in a corporate setting. Now in a corporate setting, we are going to automate an entire department.

The biggest and most crucial decision that needs to be made is where do you insert the “human-in-the-loop.” You have to go through your business process flow and build a decision tree to identify where you will insert the human. For example you might send in the human because you need the human to be in close touch with your high value customers. Or you may have the human inserted because of a regulatory and reporting requirement…

We should now think that we are replacing at a rate of 1 to 10, and soon 1 to 100 people with agents.

Ray Wang shared a little history on what Meta (Facebook) said that last year that they replaced 17% of their finance operations organization with agentic automated agents. This year Meta expects the 35% of their finance department will be automated. Next year 50% of finance will be replaced with agents.

Staff functions like Accounts Payable, Accounts Receivable, etc. are very easy to automate. We have an existing models where we outsourced these back office functions to India in the past.

Looking forward if you are not out in the field, such as sales where you touch the customer or in product development where you create new experience/products (hardware or services) you will likely be at the front end of functions that are at the top of the list to be automated.

As the renowned technology futurist Ray Wang says, we are in the age of exponential efficiency. Unless you can do something 10 times better or 10 times cheaper when you look forward in your corporate trajectory, you will be vulnerable to competitors who are deploying agentic AI at scale.

Read original article