What AI Agents Truly Are

I was asked a lot about what AI agents really are. And what’s surrounding the whole hype.

I get it: It’s hard to find your way around the buzz that’s being upheld by all the LinkedIn AI cowboys.

They hear a new word and immediately add it to their vocabulary.

That might also be a reason why economic bubbles happen. People just like to repeat (and think) what others have said.

That’s why words from one language find themselves in surrounding languages. Think French in the 18th century in Europe or English today.

Let’s get something straight first: AI agents aren’t new. The first idea of designing programs having common sense was in 1958, and the idea that agents will become the core of AI dates back to 1983.

But what’s new today is the actual technical capability of making them real. Compare it to rockets: there have been many types of rockets of different sizes ever since 13th century China, and many visionaries have dreamed of flying to the Moon with them. But it wasn’t until the 1960s that the technology was ripe to take them there.

One of the definitions of AI agents is “anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators.” (from the influential AI book by Russel and Norvig)

There you go. It’s that easy.

Two central concepts here ar the “perception” and “acting” pieces. AI agents get some information and act on it.

The perception piece wasn’t the hardest part. But acting was REALLY hard to achieve, because all systems just couldn’t “think”. Researchers would wing it by creating all sorts of rules, but none of them came anywhere near to what Large Language Models (LLMs) can do today.

To understand why 2022 (ChatGPT public launch) was a breakthrough, take look at this short chat:

Here we have it:

1. Perception: I’ve told the agent how much coffee is left in the cup.

2. Action: It knows what to do based on that

This example is super basic. But to put it into context, what you’re seeing in this screenshot is the Saturn 5 rocket (that took humans to the Moon) and everything else that came before it were the gunpowder rockets from 13th century China.

But there’s another thing that’s even more powerful about these LLMs. The “action” step can be taken to the real world via tools. That is, instead of the assistant saying “Making coffee…” it ACTUALLY makes it.

Most of us can’t afford a robot that walks around the office (or wouldn’t want to due to safety – we’re still in 2025!). But what you CAN do is connecting this to your database, CRM system, WhatsApp business chat, website chat box, or even your phone!

Here’s what it does:

Suppose you want to add people you meet at conferences to your CRM. You’re with your phone and on the go, so you dictate all the info into ChatGPT.

If you connect it via your CRM’s API, every action that involves adding a new contact (the purple dots in the diagram), actually performs that! It doesn’t just output it, like in the coffee example earlier.

And that’s the main breakthrough: AI agents can plan AND use tools!

So essentially the impact is real and tangible. But AI cowboys don’t know the reason. They just take everything that has been posted more than two times on LinkedIn and regurgitate it. Which is okay – that’s how marketing works. But they can’t be possibly making any deep transformations in your company’s manual processes without knowing the fundamental basics of how this stuff works.

Now, there’s a ton of platforms and services that allow you to create these agents from scratch. Ultimately all agents boil down to the same thing: they know when to perform an action and how to call the necessary tools.

If you get these fundamentals, you’ll be able to drill through any problem.

We cover these tools and how to create your first AI agent without any coding knowledge inside the Make Work Obsolete community.

Imagine creating an automation that saves you 80 hours of work per month, while having real tech experts behind you, who can solve any problem along the way.

Cheers,
Robert