The Hype Train

Every new tech cycle brings an insane amount of hype, and everyone jumps onto the Hype Train. I’ve seen it with VR/AR, 3D printing, IoT, blockchain, and now AI agents. In the beginning, there’s an enormous wave of excitement, but after a while, there appear to be fewer practical applications or it makes things unnecessarily complex.

Choo choo buckle up, because we’re going to jump on the hype train.

First stop: Blockchain Central Station, or should I say Decentralized Station? When this era started, I talked with clients who wanted to use blockchain to replace a traditional database. Why? Because blockchain was in the news. I think we can all agree on the complete overkill. They wanted decentralization and transparency for... customer records that only they would access. A regular PostgreSQL database would’ve done the job in a tenth of the complexity and cost. But it wouldn’t have been “innovative”.

Next station: VR. Expectations were enormous: virtual offices, mass gaming adoption, and educational revolutions. But the hardware remained expensive, people got motion sick, and the number of truly compelling applications was limited.

Do you hear that? The conductor blows the whistle, we’re moving on. Next stop: 3D printing. “Within ten years, everyone will be printing their stuff at home,” they said. Startups promised revolutions in manufacturing, construction, and even food.

I’m not saying these technologies aren’t useful, on the contrary! There are great use cases for all of them. But they mainly remain niche.

Oh, the train is leaving for the next hype.

The Definition of an AI Agent

Finally, the train pulls into AI Agent Station. But wait? Is this actually a new destination, or just the old AI Station with a fresh coat of paint?

Here’s my definition of an AI Agent: An autonomous loop that senses, decides, and acts towards a goal.

When I explained it to my wife, I used a different analogy: Think of a railway maintenance crew with a toolbox including a hammer, wrench and measuring tape. When they see a loose rail, they know to grab the wrench. When they spot a broken sleeper, they reach for the hammer. They assess the situation and choose the right tool autonomously.

The pitch seen online sounds incredible: autonomous agents that work 24/7, handle complex tasks, and adapt to any situation. Just like blockchain would “revolutionize every industry” and VR would “replace physical offices.”

The demos? Flawless. A slick agent booking meetings, researching competitors and drafting reports: all autonomously. It looks like magic.

But then you deploy it in the real world.

A customer uses “urgent” instead of “high priority,” and your agent routes it to the wrong department. Someone writes “I think we should cancel” in an email, and your agent cancels a €10,000 order. A prospect asks a slightly unusual question, and instead of admitting uncertainty, your agent hallucinates a confident answer about features you don’t have.

Each failure chips away at trust. And in our business, trust is everything.

The problem isn’t the technology, it’s the promise. We’re being sold general-purpose autonomous agents that can “do anything,” when what actually works are narrow, well-defined automations with AI components.

The Crucial Difference

Here’s what most people get wrong:

AI Automation: A predefined workflow with AI steps. Think: “If email arrives → Extract info with AI → Validate format → Send to system A.” The path is fixed, the output is predictable. It’s a train following the tracks.

AI Agent: Autonomous decision-making. It chooses which tools to use, when, and can take unexpected paths to reach its goal. Think: “Here’s your goal, figure out how to achieve it.” It’s a conductor that can change the route mid-journey.

The agent decides. The automation follows.

90% of the so-called “AI Agents” I see being posted on social media are just regular AI automations that call an LLM. That’s not necessarily wrong and will definitely create value as well, but they’re throwing more coal into the shiny AI-powered steam locomotive without understanding what they’ve actually built.

Why Simple Wins

Most companies don’t need complex, shiny AI systems, what they need is something simple yet reliable.

The Post-it Note Rule: if someone can’t explain their AI agent on a single Post-it, it’s probably overcomplicated.

When Agents Actually Make Sense:

• Living inside tools you already use (Slack bot that answers questions, Jira assistant that triages tickets, CRM helper that enriches contacts)
• Small, repetitive tasks where occasional oddness is acceptable
• Always with a human in the loop

Take our order extraction project at OWOW. We started with an autonomous agent: give it access to email, the ERP API, and let it figure out the rest. Sounds efficient, right?

In testing, it looked brilliant. In production with 1.000 orders per month? Chaos. The agent would decide to send data to the wrong system, call the wrong API endpoints, or confidently push incomplete information to tools that weren’t designed to handle it. It made autonomous decisions, great! Just not the right ones..

We rebuilt it as a strict automation:

1. Extract data using AI
2. Validate against schema
3. Flag anything uncertain for human review
4. Only send confirmed orders to ERP

Boring? Absolutely. Reliable? 100%.

The AI still does the heavy lifting: parsing messy emails, handling different formats and understanding context. But the process is predictable. When something goes wrong, we know exactly where. That’s what you need for critical workflows.

Start Building

The best way to understand the difference? Build both and see for yourself. Let me show you two simple examples: one automation, one agent.

Example 1: AI Automation (Fixed Flow)

// AI Automation: Fixed path, predictable
const orderAutomation = async (email) => {
console.log(”Step 1: Extracting order details...”);
const orderData = await llm.extract(email, orderSchema);

console.log(”Step 2: Validating data...”);
const isValid = validateOrder(orderData);

if (!isValid) {
console.log(”Step 3: Flagging for human review”);
return { status: “needs_review”, data: orderData };
}

console.log(”Step 3: Sending to ERP”);
await sendToERP(orderData);
return { status: “processed” };
};

Example 2: AI Agent (Autonomous)

// AI Agent: Makes its own decisions
const researchAgent = new Agent({
model: llm,
tools: [searchWeb],
instruction: “Research this company and their main products”
});

const result = await researchAgent.run(”Tell me about Stripe”);

// In the console you’ll see the agent’s autonomous decisions:
// > Agent: I need to search for information about Stripe
// > Tool selected: searchWeb(”Stripe payment platform”)
// > Agent: I found they do payments. Let me search for their product offerings
// > Tool selected: searchWeb(”Stripe products API services”)
// > Agent: Now I have enough information to summarize...

See the difference? In the automation, YOU decide every step. In the agent, IT decides which tools to use, how many times to call them, and what information it needs. The agent saw the first search result, determined it needed more specific information, and autonomously decided to search again with a refined query.

Try building both. Watch the agent work. It’s genuinely magical when it makes smart choices. Then watch it fail in creative ways you never anticipated. Both lessons are valuable.

Start small: build a tool that helps YOU, not production systems. Once you’ve seen an agent confidently make a terrible decision, you’ll understand exactly when to use one and when to stick with boring, reliable automation.

Final Stop

So here we are, at the end of our journey on the AI Agent hype train. We’ve passed through Blockchain Station, VR Central, and 3D Printing Plaza. Each promised to revolutionize everything. Each found its niche.

AI Agents will follow the same path. They’re not useless, far from it. But they’re also not the autonomous digital employees that every pitch deck promises, yet.

Before you buy your ticket on this hype train, ask yourself:

• Can I describe the AI Agent on a Post-it? If not, you need automation.
• Am I solving a real problem, or chasing innovation for innovation’s sake?
• Would I trust this to run unsupervised with my company’s reputation on the line?

The most valuable AI implementations I’ve seen aren’t the flashiest. They’re the boring, reliable automations that save 20 hours a week. The simple Slack bots that answer repetitive questions. The extraction tools that handle the tedious work so humans can focus on what matters.

Don’t chase the hype. Build what works.

And when the next hype train pulls into the station? Take a moment before jumping on. Ask the hard questions. Demand the real-world proof, not just the polished demos.

Because in a few years, there’ll be another train, another revolution, another technology that promises to change everything.

The question isn’t whether to board, it’s whether you’re heading somewhere that actually matters.

Choo choo. 🚂