AI agents for curious beginners: A 3-level framework for non-technical minds

Breaking down the evolution from simple chatbots to autonomous AI systems using real-world examples that anyone can understand, even without a technical background.

Every time I scroll through LinkedIn these days, I see new technical terms being thrown around. "AI agents this" and "agentic workflows that" - it's overwhelming if you're not deep in the tech world.

Here's the thing:

Most explanations of AI agents are either too technical (making your eyes glaze over) or too basic (teaching you nothing valuable).

This article is for people like me - and perhaps you - who have zero technical background but use AI tools regularly and want to understand just enough about AI agents to see how they'll affect your work and life.

A simple 1-2-3 learning path for understanding AI agents

I've broken this down into three clear levels that build upon each other:

Level 1: Large Language Models (what you already know, maybe)

Level 2: AI Workflows (the bridge concept)

Level 3: AI Agents (what everyone's talking about)

Let's walk through each one with real examples you'll actually encounter in your daily life.

Level 1: Large Language Models - The foundation

Large Language Models (LLMs) are the technology powering the AI chatbots you already use - ChatGPT, Google Gemini, Claude, and others.

Here's how they work in the simplest terms:

→ You provide an input (your prompt) → The LLM produces an output based on its training data

For example, if I ask ChatGPT to draft an email requesting a coffee chat, my prompt is the input, and the resulting email (that's way more polite than I would ever be in real life) is the output.

🧠 The key limitations of LLMs are:

• They have restricted knowledge of proprietary information (your calendar, company data, etc.)
• They are passive - they wait for your prompt and then respond

To illustrate: If I asked ChatGPT, "When is my next coffee chat with Elon?" it would fail because it doesn't have access to my calendar.

Level 2: AI workflows - adding capabilities

Building on our example, what if I told the LLM:

"Every time I ask about a personal event, perform a search query and fetch data from my Google calendar before providing a response."

With this logic implemented, when I ask, "When is my coffee chat with Elon?" I'll get the correct answer because the LLM will first check my Google calendar.

But here's where it gets tricky.

What if my follow-up question is: "What will the weather be like that day?"

The LLM will fail because the path we defined only tells it to search my Google calendar, which doesn't contain weather information.

💡 This reveals a fundamental trait of AI workflows: they can only follow predefined paths set by humans (also called control logic).

Let's push this example further. What if I added more steps into the workflow:

1. Check my calendar for the meeting

2. Extract the date and location

3. Check a weather API for the forecast

4. Use a text-to-audio model to speak the answer

Even with all these steps, this is still just an AI workflow because a human (me) is the decision maker who defined each step.

🤔 Fun fact: Retrieval Augmented Generation (RAG) is just a fancy term for a type of AI workflow where the AI looks things up before answering.

A real-world example of an AI workflow

When I was building my audience for Whatever Matters (my podcast and newsletter), I created a simple AI workflow using make.com that:

1. Used Google Sheets to compile links to news articles

2. Used Perplexity to summarise those news articles

3. Used Claude with a prompt I wrote to draft LinkedIn and Instagram posts

4. Scheduled this to run automatically every day at 8 am

This is an AI workflow because it follows a predefined path set by me. If I tested the workflow and didn't like the final output (maybe the LinkedIn post wasn't funny enough), I'd have to manually go back and rewrite the prompt for Claude.

This trial-and-error iteration is being done by me, a human.

Level 3: AI agents - where the magic happens

Here's the one massive change that has to happen for an AI workflow to become an AI agent:

➡️ The human decision maker must be replaced by an LLM.

In other words, the AI agent must:

1. Reason: "What's the most efficient way to compile these news articles? Should I copy and paste each article into a Word document? No, it's probably easier to compile links to those articles and then use another tool to fetch the data."

2. Act: "Should I use Microsoft Word to compile links? No, inserting links directly into rows is more efficient. What about Excel? The user has already connected their Google account with make.com, so Google Sheets is a better option."

📊 Because of this, the most common configuration for AI agents is the REACT framework: • RE = Reason • ACT = Act

A third key trait of AI agents is their ability to iterate. Remember when I had to manually rewrite the prompt to make the LinkedIn post funnier? An AI agent would be able to do this autonomously.

In our example, the AI agent might:
• Draft V1 of a LinkedIn post
• Add another step where an LLM critiques the post based on LinkedIn best practices • Repeat until the best practices criteria are all met
• Produce the final output

Real-world examples of AI agents

One impressive example is a demo website created by Andrew Ng that illustrates how an AI vision agent works. When you search for a keyword like "skier," the AI agent in the background is:

1. Reasoning what a skier looks like (a person on skis going fast in snow)

2. Acting by looking at clips in video footage

3. Identifying what it thinks is a skier

4. Indexing that clip

5. Returning that clip to the user

Although this might not feel impressive at first glance, remember that an AI agent did all that instead of a human reviewing the footage beforehand, manually identifying the skier, and adding tags like "skier," "mountain," "ski," "snow."

When I'm developing product roadmaps at caisy.io, I often think about how AI agents could transform content management systems by autonomously making decisions about content organisation, metadata tagging, and even content creation - tasks that currently require significant human intervention.

The three levels are simplified

To wrap up, here's a simplified visualisation of the three levels:

Level 1 (LLMs):
• Input → LLM → Output
• Key trait: Basic question-answer functionality

Level 2 (AI Workflows):
• Input → LLM following a predefined path → Output
• Key trait: Human programs the path for LLM to follow

Level 3 (AI Agents):
• Goal → LLM performs reasoning → Takes action → Observes results → Iterates → Final output
• Key trait: The LLM is the decision maker in the workflow

Understanding these distinctions has helped me navigate the rapidly evolving AI landscape as I build and grow scriptbee.ai - an agentic GTM tool for lean teams. Whether you're a founder, marketer, or simply curious about technology, knowing the difference between these concepts will help you see through the hype and identify genuine opportunities.

What's next for you?

Are you using any AI workflows or agents in your work? I'd love to hear about your experiences in the comments below or in a reply to my Whatever Matters newsletter.