Upcoming Events:
🏫 Course: AI for Product Designers | Aug 25 - Sep 4
🎙 Talk: From Mockups to Merged PRs: An AI-Native Designer’s Playbook | Sep 4
Today’s newsletter features a guest post by Dominik.
He is a hands-on designer building products fully with AI for US clients, and bootstrapping a profitable business made possible by AI. He’s spent 14 years across design and product roles in consumer electronics, preventative health, decentralized finance, and Agriculture.
In today’s article, he explores “ontology”, the structure that defines what exists, what things mean, and how they relate to each other. Using the boarding pass as an example, he shows why understanding the systems behind a product may matter more than designing the interface itself in the age of AI.
And he even recorded the voiceover himself. Check it out at the top.
Let’s dive in.
“AI can probably do your job now, right?”
An executive asked me that two years ago, leaning against the white stone countertop in his kitchen — but kindly, the way you’d tell someone they’ve got something in their teeth. And regarding the part of my job he could see—the screens, the flows, the copy—he was right.
Now I will not claim the moral high ground here, arguing “human taste and judgment” because that can’t be measured. Nor will I bury you in arguments about how a human brain can run all day on a sandwich, while machines guzzle rivers of electricity to do less.
It took me more than a year of working with AI, every single day, to find the proper answer I owed him.
But here is the promise that you and I must cling to across the 2000 words that follow: I will reveal to you what—after fourteen years of building, managing teams in five countries, and surviving months of monomaniacal experiments and dead ends—I have determined to be the single greatest artifact of system design you will ever throw away.
I’m here to show you the part of your job that is entirely invisible—and hard for AI to copy.
The evidence I needed to answer that executive is something so obvious we’ve been using it for sixty years. I’m pretty sure I have one crumpled in my coat pocket right now. It’s yours, too—unless you’ve already thrown it away.
Like a nest of complicated wires without labels
At 7:46 a.m.—67 minutes after arriving—I had crossed some of the most regulated and controlled spaces on Earth: Heathrow’s departure area.
I’d survived the quick-draw showdown at the security scanner, a routine now fine-tuned to seconds: one shoulder strap, bag open on my chest, liquids and laptop locked and loaded. After buying two bottles of rosé from duty-free I was reminded to scan my boarding pass and, I joined the momentary standstill of passengers looking up at the almighty departures display, that altar we all face hoping to find our flight number and gate.
With time, my “elite” boarding group C, was called. I made my way to seat 16A and fell asleep.
Not one person told me where to stand, when to move, or what to do—yet it all worked. I’d just been funnelled through a series of highly regulated events so complicated, our ancestors would be left in awe.
Though the number can vary between 9 and 14 million passengers are moved safely around the world every single day. Huge numbers, and they are completely unaware of the hundreds of impeccably choreographed systems built to carry them.
For some, traveling on planes has become as routine as public transit; for others, it’s a stressful, demanding ordeal. It’s a dense web of cultures, languages, and emotions—a place where exhausted parents herd kids past colourful candy, a retiree tests her freedom after knee surgery, and an overworked middle manager stresses over a deadline on a flight back from an unexpected funeral. All mixing into a system that must accommodate everyone to keep travelers safe, calm, and on time.
No small task.
So how does a piece of paper direct around five billion passengers a year without talking to them?
Before you dismiss a boarding pass as just a barcode and some text, let’s break it down.
Take the pass apart, and you stop seeing a ticket. You start seeing decisions. A flight is a thing. A passenger is a thing. A seat is a thing. A date is a thing. A bag belongs to its owner; a flight belongs to a route; a kid cannot sit in the exit row because, in an evacuation, that seat isn’t bonus legroom—it’s a job.
All these decisions—what exists, what it means, and how things are allowed to relate—form an ontology. And before you go off looking up that word, here’s short history:
Ontology comes from philosophy as a study of what exists, then was formalized in the 20th century as “applied ontology” for modeling domains of knowledge. Then was adopted in computer science as structured, machine-readable specifications of concepts and relations.
Ontology now underpins the “Engin-Nerding” good stuff of today.
Knowledge graphs, the semantic web, AI reasoning, and data integration. It’s important but I haven’t ever had a conversation about it with an engineer without my eyes glazing over and asking them to repeat but slower this time. Put simply. Ontology is “structured relevant context”: what exists, what it means, and how things are allowed to relate.
The simple interface of a boarding pass, which is so easy to critique visually, is quite literally a minimalist user interface hiding one of the most complex, globally standardized ontologies on the planet.
To the passenger, it just looks like a few letters and numbers. But to the airline’s system, every single item printed on that pass represents a deep web of constraints, rules, and relationships.
Why grandma isn’t allowed to board last
Drinking a steaming hot coffee, I had a commercial pilot walk me through the boarding process, some of which I will try not to butcher here. But note: the amount of thought that’s gone into every minute makes anything you’re imagining — simpler than reality.
Consider boarding groups. When you think of boarding groups you might think of status, the elderly and babies. Yet in addition to grouping in-flight service to one part of the plane, your boarding group is designed to reduce congestion at gates. That congestion causes delays, increases work and fatigue for staff, and leads to financial loss for an airline.
The business class is normally up front on jetliners. Filling them first through the higher boarding group affects the aircraft center of gravity by moving it forward, allowing the aft cargo hold to be filled earlier. When arriving, the business group can move faster to their next flight or transportation. Incidentally the aft cargo hold is emptied first to keep the center of gravity forward and keep the aircraft from tipping.
Group 0 assists passengers with reduced mobility and families with small children who would otherwise hold up the flow of boarding. Because they board first, gate staff have a buffer to do other tasks while the crew, catering is boarded.
Group C (a much later boarding group) tells the staff something else entirely: these passengers have no larger overhead baggage left to stow. The system has already accounted for the cabin space and these bags likely need to be loaded as luggage underneath the plane.
That one letter on the boarding card represents a number of cascading decisions. The boarding pass is efficient because the ontology beneath it is sound.
It’s merely the tip of an iceberg called the Passenger Name Record—a 60-year-old system of logic. Let’s visit this to make sense of why this helps you in the world of AI today.
In 1953, as your grandmother’s favourite game, Scrabble, first became popular, and while the coronation of Queen Elizabeth II was being watched by 20 million people on television for the first time, two men had a chance encounter on a flight from Los Angeles to New York.
C.R. Smith, president of American Airlines, sat next to R. Blair Smith, an IBM salesman. Having the same surname the conversation turned to seat bookings, and American Airlines’ manual problematic “request and reply” system.
At that time, Seat inventory was controlled from the departure city using a rotating file of handwritten cardboard cards. When a seat was booked, a team of operators found the flight card, marked the seat, and wrote the ticket by hand. A single reservation took an average of 90 minutes; Blair Smith claimed IBM could build a machine to track it all.
That conversation led to SABRE (Semi-Automated Business Research Environment) eleven years later, in 1964. SABRE forced American Airlines to pin its rules down — All the knowledge that used to live in the clerks’ heads was captured in a structure a machine could read. Then within a decade, every major airline had done the same.
“You can’t polish a turd, but you can roll it in glitter.”
It took the airline industry sixty years, millions of dollars, and thousands of people to refine the ontology that moves people safely and seamlessly everyday.
The hard work wasn’t improving the surface, the boarding pass, rather everything it represented. With a solid ontology the interface could evolve over time, from physical printed cards to digital screens on our phones, it made almost no difference to the end user. It just felt easier. The medium changed, but the underlying system didn’t move at all.
If you spend your time defining an underlying ontology it naturally produces an incredibly effective interface. Contrast this, if that foundational logic is fundamentally broken, no amount of aesthetic polish can save the system from failing.
I don’t think anyone would say a boarding pass is pretty yet it’s insanely effective.
Yet there’s a lot of pretty designs that are absolutely worthless.
One of the opportunities for Product People today (designers, PM, etc) is to extract what is in the heads of a company, map the processes and system. Define the protocols behind decisions, and bring everyone along with you for the ride.
“Twenty seconds is a huge saving in a twenty-minute turnaround.”
I was sitting in a meeting room with the owner of an aviation analytics company and a commercial pilot. We were watching a video feed of a Boeing 737-800 approaching a gate.
“Stop—go back,” the pilot asked. He’d spotted a problem.
Had one of the ground staff stood just five meters to the left, ready to connect the external AC-power the exact moment the plane’s parking brake engaged, they would have saved twenty seconds. For an airline, on-time performance is the primary metric because a single bad turnaround ripples out across the entire network. One delay leads to the next, crossing airlines, breaking later connections for passengers and crew members, and cascading into thousands of ruined days, missed birthday parties and sad kids. We don’t want sad kids.
A system this complex is a living thing, constantly fighting new friction. Sixty years after the industry defined the architecture of air travel, this pilot was still actively refining its ontology: What counts? What is wasted? Who stands where? We need to adapt to events and the entropy of the world as it changes: For instance, 9/11, and Covid-19 both change how we get on a plane.
Most of us, certainly me, have never designed a product with the high-stakes and physical constraints of aviation. However after the last year of building with agentic systems everyday.
I believe that the level of care that was put into aviation is becoming accessible to every company, and needed for your business to stand out from others.
Why Camouflage is no longer working
In 2019, I was on a team that wasted 14 months building a completely useless product. Endless discussions and wireframing, service blueprint, and customer meetings. All kept the team busy and acted as a camouflage. Giving us the illusion of progress hiding the lack of deep thinking, and decisions being deferred.
As more people discover AI prototyping, wasting that much time is going to be a lot harder today. The industry is undergoing a massive shift from static interfaces and tools to Agentic AI—systems that don’t just display information, but actively perform tasks and make decisions. As of today, AI is making the generation of the surface layer (the boarding pass, screens, the code, the UI) nearly free. In doing so, it hasn’t replaced designers but it’s pulling away the camouflage of busy work and to bring out the real work.
For an AI agent to be truly helpful without hallucinating, it cannot just guess what the user wants based on statistical text patterns.
If a flight is delayed, a generic AI without an ontology can only generate a polite apology email. It doesn’t inherently understand that a “Passenger” is a human who now needs a “Hotel,” which must be within a “10-mile radius” of an “Airport,” and that their “Baggage” must be rerouted.
Ontology is the map that gives AI that understanding.
Context, just way better
As many companies try to apply AI today, the pattern is the same as in any software project: engineers build for feasibility, but often lack the human-centric lens that makes a product or process feel natural. That is where you come in. AI labs talk about “context” — which can feel wildly vague. The more valuable version of context isn’t feeding a model more documents; it’s sharing an ontology with an AI. Think of ontology as structured context.
You don’t need a degree in data science to design for AI. You just need to shift your focus from designing surfaces to defining relationships.
On a recent client project we had an AI model extract key details from lengthy legal documents. On its own it was okay but inconsistent. So we spent time understanding the nuances of the client’s world, then gave the model a clear ontology — the language used in this office, the legal frameworks, the laws, the narrow use case we wanted it to work within. And the output was significantly more accurate in edge cases and cheaper to run.
Ontology + tool use reduces token costs.
Think of AI like a new hire. If you don’t give it the context, and time to understand the relationships of your work they’ll never quite do what you want. The AI remains general. Generic, hallucinatory, and shallow.
You cannot automate what you do not understand. And designers are already trained to do exactly this. When you map user journeys, service blueprints etc. You’re defining processes and objectives. You are understanding goals and reducing the friction to get there. When you do object mapping, you are defining entities and properties. When you research behaviour, you are defining relationships, actions, meaning and constraints.
The highest leverage design work of the future is helping companies define their goals and ontology.
Two years ago in that kitchen an executive asked me if AI could do my job.
I finally have my answer. AI can make the boarding pass. It cannot yet understand or design what the boarding pass means. It cannot decide that a child can’t sit in the exit row, or that Group C already has their bags below deck, or that the ground crew should stand five meters to the left.
That work of defining ontology — deciding what exists, what it means, and what the system must never do — is still the job.
Now that AI accelerates the rest of our work, we can finally invest in depth. Bring the same level of care we see in Aviation to our own companies.
We define the world. The machine moves inside it.
So the next time you throw away a boarding pass, remember — It’s evidence of the valuable work still worth doing. And you already have the skills to get started.
That’s it for this week.
If you enjoy this article, Dominik also writes a Substack where he shares thoughtful insights about building products. Subscribe to Dominik here.
See you next time.
Xinran
-
P.S. What are your takeaways on ontology? Or what’s a similar example you can think of? Let me know in the comments.
Hey, love your thinking and the article. Thank you. A couple of observations/comments:
1. Why would you say this work needs to be done by designers? Why not PMs, for example?
2. The discussion is slowly shifting from customer obsession to system thinking, where actors are not only clients but also stakeholders, company earnings etc and the entire context in which a product or service functions.
3. An example. I'm just checking if I get it right. If the universe was my own substack, the relationships and their success that I need to think about, ontology should focus on:
- readers and audience, maybe different types of readers, and how they sequentially could move through different pieces
- collaboration with other writers
- priorities and time involvement I have
- biggest ROI, and so on
What do you think?
Excellent article 👏. I thoroughly enjoyed it. At last I found an article that defines Ontology that is basis of good domain design. It provided very detailed explanation on aviation industries complex rules by taking boarding pass as example / what delay on boarding would cost them. Thanks for this great views.