The Optimization Trap: Why Technology Keeps Making Things Ugly
In 1961, Jaguar unveiled the E-Type at the Geneva Motor Show. Enzo Ferrari called it the most beautiful car ever made. The line wasn't just elegant — it was unnecessary. The curves did nothing for acceleration. The long hood added weight. The proportions were extravagant in a way that made no rational sense.
And yet.
Sixty years later, the best-selling vehicle in America is a pickup truck shaped like a brick. Tesla's Cybertruck looks like a first-year CAD student's attempt at a stealth bomber. The Waymo autonomous vehicle — once a friendly, rounded pod that looked like it belonged in a Pixar film — was redesigned into something more angular, more efficient, more forgettable. Safer, probably. More sensor-optimized, certainly. But you wouldn't call it beautiful.
This isn't about taste. It's about what happens when you optimize.
The thing optimization always kills first
Optimization is the defining logic of modern technology. Make it faster. Make it cheaper. Make it scalable. Make it measurable. It's how we build software, design products, run companies, train algorithms. And it works — until you step back and notice what's missing.
What's missing is almost always the same thing: the stuff that doesn't optimize well.
Beauty is one of those things. So is surprise. So is inefficiency that creates texture, products that age interestingly, designs that reward a second look. These qualities resist metrics. You can't A/B test your way to elegance. You can't run a regression on delight.
So we stop trying.
Consider the car. The 1960s and 70s produced vehicles with personalities: the Citroën DS, the Alfa Romeo Spider, the Datsun 240Z. They were shaped by designers who had opinions. Today's cars are shaped by wind tunnels, crash-test standards, and cost-per-unit calculations. The result is a global fleet that looks like it was extruded from the same machine.
When every variable is optimized separately, the whole becomes less than the sum of its parts.
This isn't nostalgia. It's a pattern. And it's accelerating.
What Waymo lost in translation
The original Google self-driving car — the one that looked like a friendly cartoon turtle — was polarizing. Some people loved it. Others thought it looked ridiculous. But no one forgot it.
Then came the next generation: Waymo's Jaguar I-PACE fleet, and later the fully custom Zeekr vehicles. More sensors. Better range. Cleaner integration. And visually? Forgettable. The rounded edges gave way to sharper lines. The whimsy disappeared. What replaced it was efficient.
The engineers will tell you this was necessary. The sensor arrays needed better sightlines. The aerodynamics needed work. The design had to communicate seriousness to regulators and riders who were already skeptical.
All true.
But here's what also happened: the optimization was multi-dimensional, and beauty wasn't one of the dimensions.
The vehicle was optimized for:
- Sensor coverage and redundancy
- Regulatory approval and safety perception
- Cost of manufacturing at scale
- Aerodynamic efficiency
It was not optimized for:
- How it made people feel when they saw it
- Whether it added something visually interesting to the street
- Whether someone would want to take a photo of it
You can't optimize for what you don't measure. And we've built an entire industrial system around not measuring the things that matter most.
The AI art problem is the same problem
In 2023, AI-generated art started flooding the internet. Midjourney, Stable Diffusion, DALL-E — tools that could produce a decent illustration in seconds. The results were often technically impressive. They were also visually homogeneous in a way that's hard to unsee once you notice it.
The same glassy-eyed portraits. The same over-saturated fantasy landscapes. The same cinematic lighting that doesn't come from anywhere in particular. It's not that the images are bad. It's that they converge on a narrow aesthetic optimum: the average of what the training data rewarded.
This is what optimization does when left unsupervised. It finds the local maximum — the thing that scores highest on the metric you gave it — and drives toward it relentlessly. In AI art, that metric is often "how closely does this match what people clicked on, saved, or upvoted before?"
The result is art that looks like art, the way margarine tastes like butter.
Compare that to human artists working in the same digital tools. The best of them are deliberately inefficient. They'll spend hours on a detail no one will consciously notice. They'll make choices that slow them down because the choice matters more than the speed. They're optimizing for something the algorithm doesn't have language for.
The things we call beautiful are often beautiful because they're suboptimal in some measurable way.
A hand-thrown ceramic mug is less structurally perfect than a factory-molded one. That's the point. The imperfection is where the person shows up.
How we got here: a short history of thing-making
This tension isn't new. It's as old as industrialization.
In the 19th century, the Arts and Crafts movement emerged as a reaction to mass production. William Morris and his peers argued that factory-made goods were soulless — that the efficiencies of the assembly line had stripped objects of their humanity. They were right, and also too late. The economic logic of mass production was overwhelming.
But for a while, there was a balance. Mid-century industrial designers — Raymond Loewy, Dieter Rams, Charles and Ray Eames — managed to work within the constraints of mass manufacturing and still produce objects that felt considered. A Braun coffee grinder. An Eames lounge chair. These weren't hand-carved artifacts, but they weren't optimized into oblivion either.
What made that possible was a different kind of constraint: the designer had authority, and the number of variables was smaller.
Rams could say "this knob goes here, at this angle, because it feels right" and the engineers would work around it. The manufacturing process was complex, but not so complex that every aesthetic decision could be second-guessed by a simulation.
Today, we have infinite variables. Every surface can be parametrically optimized. Every color can be A/B tested. Every curve can be justified — or questioned — with data. The designer's authority has been replaced by the optimization function's output.
And the optimization function doesn't care about beauty. It cares about the goal you gave it.
Why this keeps happening
There are three forces that make this pattern nearly inevitable:
1. Optimization is measurable. Beauty is not.
If you're building a car, you can measure 0-60 time, crash-test ratings, fuel efficiency, cost per unit. You can't measure "does this make someone stop and stare?" Not in a way that holds up in a board meeting.
So when trade-offs come — and they always come — the measurable thing wins.
2. Consensus is the enemy of distinctiveness.
The more people who have input on a design, the more it regresses toward average. This is true in committees, and it's especially true in algorithmic systems trained on large datasets. The algorithm converges on what's most broadly acceptable, which is rarely what's most interesting.
3. We've professionalized away taste.
Fifty years ago, a senior executive at an auto company might have had strong aesthetic opinions and the authority to enforce them. Today, those decisions are distributed across teams: design, engineering, marketing, legal, finance. Everyone has a valid concern. No one has permission to say "I don't care, it has to look like this."
The result is design by consensus, which is another way of saying design by optimization.
What we lose when we lose beauty
It's tempting to dismiss this as a luxury problem. Who cares if the Cybertruck is ugly? It's functional. Who cares if AI art looks generic? It's fast and cheap.
But beauty isn't ornamental. It's informational.
A beautiful object tells you someone cared. It signals that the maker thought about more than the minimum viable product. It creates a relationship between the thing and the person using it — a relationship that goes beyond utility.
When we strip that away, we're not just making things uglier. We're making the world less worth paying attention to.
Aesthetic blandness is a kind of learned helplessness. If everything looks the same, why bother looking?
This shows up in software, too. Every app now has the same rounded corners, the same sans-serif typography, the same flat design language. It's clean. It's modern. It's also impossible to tell apart. We've optimized for usability and accessibility — both good things — but in doing so, we've erased personality.
The same thing happened to airports. To hotel lobbies. To coffee shops. Optimization smooths out the local, the weird, the specific, and replaces it with the global average.
The way out (if there is one)
The problem isn't optimization itself. It's uncontested optimization — optimization that runs without a counterweight.
The companies and designers who still make beautiful things have figured out how to build that counterweight in. They treat aesthetics as a constraint, not a nice-to-have. Apple, for all its faults, still has a design team with veto power. Porsche still employs people whose job is to make sure the new 911 looks like a 911.
These aren't acts of nostalgia. They're acts of discipline. They're saying: this matters, even though it doesn't show up in the quarterly metrics.
The alternative is a world where everything works fine and nothing feels like anything.
We're not there yet. But scroll through a parking lot, or an AI image generator, or the App Store, and you can see it coming.