a flight?
Chances are high that at some point — maybe for vacation, maybe for work — you have. At the airport, as you hand over your luggage, it disappears into the hidden world of baggage handling. Then, in most cases, your luggage magically reappears at your destination. Not much to say about this, actually.
But before you reach the service counter, you have to get your luggage there. And airports are large. If you ever had to walk across any major hub such as Dubai, Frankfurt, Heathrow, Istanbul, or Beijing while dragging bags, you know what that feels like.
Now imagine doing that without a trolley bag.
At some point, somebody had the idea of putting small wheels on a suitcase and added a handle. That’s it; nothing fancy. No machine learning, no distributed systems, no “world’s hardest problem”-level problems. Just wheels on a bag. Yet, this simple idea changed how millions of people move through the world, not just when flying.
We almost never think of the trolley bag as an “innovation.” In our minds, innovations always seem to be transforming the world, going zero to one in an instant. But the trolley is an invention — and, like many others, an important one.
The Trolley Bag Problem
What does the trolley have to do with AI? Well, right now, AI is often hailed as the missing ingredient to solving humanity’s biggest and hardest challenges.
X (Google’s moonshot factory) and similar organizations advertise theirs focus on renewable energies, clean drinking water, reliable and healthy foods. In many of these pitches, AI shows as an essential tool: optimizing energy grids, modeling crop yields, improving medical diagnosis. These are noble goals and I don’t doubt the sincerity or technical depth behind many of these efforts*.
But there’s a gap.
Much of the AI narrative lives in the realm of moonshots: spectacular, press-ready, “this could change everything” stories. That is appealing, and we want to believe in such announcements. Just think about bringing a man to the moon.
Yet the inventions that quietly move society forward are often those that are mundane, nearly invisible improvements:
- Wheels on luggage
- The barrow
- The zipper
- The lighter
- Standardized plugs
- Road signs
These are boring inventions, yes, and nobody considers them as such. But they are also massive. They reduce friction for billions of people every day.
Most AI projects today don’t target this level of, well, boring usefulness. They either:
- Optimize experiences we probably don’t want to optimize much further (“better” content recommendations, slightly faster ad click prediction), or
- Aim at huge global challenges where impact is real but slow, uncertain, or heavily constrained by non-technical factors.
What is missing from this listing is the AI-days equivalent of the trolley bag: simple, reliable, inventions that day-in, day-out remove friction for you and me in ways we barely notice — but would miss immediately if taken away.
Overestimating AI’s Everyday Impact
For most people, I’d wager that the following fundamentals still matter more than the latest AI model releases (even if they go from 100 billion to 500 billion parameters):
Good relationships.
Good food.
A safe and stable home.
Good health.
At the current stage, AI doesn’t dramatically improve these in the way the hype sometimes suggests. Not for the majority of people, and not yet at the “wheels-on-a-bag” level.
Sure, AI systems can:
- Tell you when to water your plants.
- Suggest a new YouTube video.
- Draft an email or summarize a document.
These are nice, but mostly marginal and often require additional human re-tweaking. You don’t fundamentally upgrade your life by having a model remind you to water your plants. On your deathbed, you (hopefully) won’t think: “I wish I’d had better content recommendations.”
This doesn’t mean AI is useless. It’s already valuable in many workflows, including mine (think about coding assistance, for example!). But we should rethink our expectations: such tooling improvements are not the same as civilization-shaping inventions. And right now, a lot of AI attention is biased towards spectacular narratives, and away from the quiet, structural, boring improvements.
What We Miss When We Only Chase Moonshots
When research, funding, and talent all converge around the “world’s hardest problems”, three things can happen:
- Boring problems stay unsolved. Annoying paperwork processes, hospital workflows, municipal services, accessibility issues, logistics quirks — areas where small, robust AI tools could remove daily pain — get less attention.
- Potential is mistaken for presence. We talk as if transformative AI impact has already fully arrived, when much of it is still conditional: on policy, infrastructure**, economics, adoption.
- We overestimate how much AI matters for a good life. We risk treating AI literacy or AI enthusiasm as more important than basic, human, offline things that actually drive wellbeing (such as friendship or food).
The trolley bag metaphor can serve as a sanity check: if an AI system disappeared tomorrow, would people feel it like losing wheels on their luggage? In some narrow cases***: yes. In most cases: no, not at all.
What Does This Mean for You?
To anchor this perspective in your day-to-day thinking, I suggest three ways:
1. Be cautious with AI claims
When you see bold AI promises — containing “revolutionize,” “disrupt,” “solve X forever” style claims — do a quick mental check:
- Does this improve something concrete in daily life, or is it mostly a demo?
- Is the bottleneck here really intelligence (which could be solved by an advanced AI system), or is it policy, incentives, logistics, or basic infrastructure?
- If this system vanished, who would actually be worse off, and how?
Keep in mind: you do not need to be a cynic; just have well-intended skepticism.
2. Compare AI to mundane, boring inventions
Use everyday inventions as a reference class:
- Does this AI system simplify life as clearly as, for example, lighter vs. matches, zipper vs. buttons, trolley bag vs. carrying?
- Is it robust, cheap, and potentially boring enough that people will rely on it without thinking?
If the answer is “not even close,” treat the announcement accordingly: interesting, maybe useful, but probably not world-reordering.
3. When choosing your own AI projects, consider going anti-hype
If you’re working in ML or AI:
- Look for problems that are unsexy but real: scheduling, documentation, accessibility, internal tools, error reduction, safety checks, forms, billing, routing, maintenance.
- Aim for tools that people stop noticing because they just work.
- Optimize for reliability over impressiveness.
Ask yourself: Is this closer to a trolley bag or to a launch trailer? If it’s the former, you’re probably on a good track.
Closing Thoughts
I’m not arguing against ambitious AI research. That would be counterproductive to my own work. We should definitely explore what’s possible and apply it to hard problems. But we should also recognize a gap here:
Right now, most AI hype lives far away from the quiet, structural improvements that shape everyday life.
As individuals — researchers, engineers, users — we can respond by staying skeptical of inflated promises, by valuing mundane but meaningful improvements, and by intentionally building tools are more like a set of wheels on a bag.
Those are the kinds of changes that, over time, move the world forward.
* In fact, behind such inventions, are easily 10+ years until notable improvements, not considering the many proverbial shoulders upon which these years are built themselves. From what I recall, David Silver started researching Reinforcement Learning around 2000 — but it would take a decade and more until we heard about AlphaGo!
** If you ever had to deploy a model, you know that infrastructure really is a paint point. Or, think about the energy costs used to train these models. The money spent on training the AI could probably already have solved the problem without needing AI…
*** Mostly work, I suppose. For day-to-day interaction with people, AI is not needed.