“If I had asked people what they wanted, they would have said faster horses.”
Let’s assume Henry Ford actually said it. The issue isn’t that users asked for faster horses. It’s that “what do you want?” is a terrible research question.
Of course they said faster horses. That was the only frame of reference they had. But dig into the actual problem and you hear something completely different. Not “faster horse,” but: long rides leave me sore, I can’t bring my whole family, the horse needs rest and costs money year-round. None of those answers mention a car. Every one of them points directly to what a car solves.
Good research doesn’t ask for solutions. It uncovers problems. And the same people who cite Ford to avoid user research are now using AI to build products faster than ever, which means they’re building the wrong things faster than ever.
Henry Ford couldn’t build a car in a weekend. You can ship a working product in hours. The barrier to building dropped to zero. The barrier to understanding what people actually need stayed exactly the same. Which makes one question the only one that still creates a real advantage: what are they actually struggling with?
The discipline to say it doesn’t belong
The skill that separates builders who ship great products from those drowning in features isn’t technical. AI can’t supply it.
It’s restraint: the discipline to say “we could build this, but it doesn’t belong here.”
Consider a focused client portal. Agencies love it. Then enterprise prospects start asking for invoicing, time tracking, project management. Each one is buildable in a weekend with AI. So you build invoicing. Support tickets shift from “how do I share a file” to “why isn’t my invoice syncing with QuickBooks.” You’ve absorbed a job that wasn’t yours to absorb.
The restraint move has three versions: build a clean integration instead of a native feature, partner with an invoicing tool rather than becoming one, or just say no and serve the need without absorbing it. Each requires more judgment than building the thing would have.
This applies internally too. My content operation has a lot of moving parts. One app for all of it sounds efficient. In practice it’s brittle and a nightmare to maintain. What works: purpose-built tools, each with one job, connected by agent skills. The restraint isn’t in choosing less capability. It’s in choosing the right surface area for each tool.
Plan mode in every major AI tool is a signal that spec-driven development is now standard. But jump straight into plan mode without the strategic thinking first, and you end up with a beautifully planned feature that shouldn’t exist. Before I open plan mode, I run a conversation that determines whether I should be planning this thing at all. That’s the step most builders skip, and it’s where restraint actually happens.
AI can pressure-test your thinking and surface trade-offs. But it can’t decide what your product should be. That strategic judgment is on us. And it’s the most valuable work we do now.
The bubble is expanding
For most of design’s recent history, taste was hidden behind execution. If you could use Figma well, deliver polished components, and handle handoffs cleanly, that was enough. The role was bounded. You could fill it without ever being asked to have a real opinion.
The execution layer is compressing now. The role is expanding. Strategy isn’t someone else’s job anymore. Neither is knowing which problem is worth solving, or having conviction about what will resonate with the people you’re building for.
The gap is more visible now precisely because execution no longer fills it. If you’re not doing the judgment work, it shows in a way it simply didn’t when craft could cover for it.
The designers who are thriving in this environment aren’t the ones who adopted the fastest new tools. They’re the ones who were quietly doing the judgment work all along and are now valued for it more directly. The ones who struggle are working within the old, smaller boundaries, and the space around them makes that obvious.
The role isn’t shrinking
Every industry where production costs collapsed followed the same arc: a flood of competent-but-generic output, followed by a sharp premium on what production can’t generate.
When photography made portrait painting cheap, illustration found new territory. When digital audio made music production cheap, the things that couldn’t be replicated — live performance, authentic voice, the specific sonic signature of a room — became more valuable, not less.
The same law is at work now. When AI makes interface generation essentially free, it doesn’t eliminate the need for someone who knows which interface to generate and why. It makes that person the only person who matters.
The work that’s always mattered most is suddenly the only work that’s clearly not going anywhere.
The question is who’s actually doing it.
Also in May: About Craft, Taste, and Speed · The Floor Just Moved · What Linear Did with Taste
Continues in June with Intent Engineering for AI Agents.
Daniel Arevalo is Product Experience Lead at Omny Security. He builds AI-native tools and writes about design, product leadership, and what it means to direct systems that execute at a speed humans never had before.