Design Got Cheap. Taste Got Expensive.

I started making things on the internet in 2005, redesigning MySpace layouts and hand-writing HTML in my bedroom in Florida. Back then, a decent-looking page was a small act of will. You had to want it. You sat with the tags until the thing looked right, and if it looked right, that meant something, because most pages didn't.

Twenty years later, I can describe a page out loud and have a working version in a couple minutes. So can you. So can the roughly 8 million people building on Lovable, a company that crossed $500M in annual recurring revenue this year and sees more than 100,000 new projects created a day (Sacra). Most of those builders are not engineers. About 63 percent of people using AI app builders have no coding background at all (Hostinger). A quarter of Y Combinator's Winter 2025 batch shipped codebases that were 95 percent or more AI-generated (TechCrunch).

Design is no longer scarce. It is effectively infinite. And that changes the job in a way most product teams have not caught up to yet.

The floor came up, and the floor became the look

Here is what nobody tells you about abundance: it flattens everything it touches.

When everyone can generate a clean interface, clean interfaces stop being a signal. Teams report shipping features 40 to 60 percent faster once they lean on prompt-to-UI tools (Drawbackwards). The output is real. Components align, spacing is consistent, the accessibility checks pass. And yet if you open five AI-native products side by side right now, you often can't tell them apart. Same dark gradient. Same glowing orb where the product should be. Same rounded cards, same system font, same tasteful and anonymous nothing.

There's a good reason for this, and it isn't laziness. A generative model is a prediction engine, not a taste engine. Ask it for a landing page and it hands back the most statistically probable one, which is the average of everything it was trained on (Managed Code). The average is, by definition, the least distinctive option available. Left to its defaults, the tool doesn't give you your taste. It gives you the same taste it gives your competitor.

So the floor came up for everyone at once. And when the floor rises for everyone, clearing it stops being an advantage. It becomes camouflage.

Design was never the hard part

Sameness is the shallow version of the problem. The deeper one is older than AI.

Making something look designed was never what made a product great. We have decades of evidence for this. Pendo studied anonymized usage across hundreds of software products and found that 80 percent of features are rarely or never used, with just 12 percent of features driving 80 percent of daily activity (Pendo). The Standish Group put the figure at 64 percent rarely to never used. Pendo estimated that publicly traded cloud companies were collectively spending up to $29.5 billion a year building features almost nobody touched.

Sit with that. Long before a single interface was generated by AI, most of what teams carefully designed, built, reviewed, and shipped went straight into a drawer nobody opened. Those features had specs. They had mockups. Plenty of them looked great. They were still a waste, because looking finished and being worth building are two different things.

AI did not fix that. It emphasized it. We just got much, much faster at producing beautiful things that do not need to exist.

The bottleneck moved. Again.

The Misfit team has written before about how AI changed how fast companies get built without changing what makes them work. That is exactly right, and it holds at the product level too. The constraint used to be production. Can we make this, can we make it look good, can we make it fast enough. AI dissolved that constraint. What is left is the part it cannot touch.

Judgment.

Judgment is knowing which of the twenty interfaces you just generated is the one, and being able to say why. It is knowing that the feature everyone is excited about is headed straight for the 80 percent nobody uses, and killing it before it ships. It is understanding whose problem you are actually solving well enough to notice when a technically correct solution quietly misses the point. It is the distance between a prototype that works and a product that is right, which, as anyone who has shipped knows, is most of the actual work.

Product taste is the new superpower.
What is worth building?
What's the right way to built it?

I’ve spent the last few years building governance and evaluation products for enterprise AI. That entire category is, in a sense, judgment turned into software: helping teams decide whether a model's output is good enough to trust, and then act on it. What I took from it is simple. The evaluation is the product. The generation is the cheap part. That was true for AI systems long before it became true for the tools we now use to design them.

Taste is a decision you make on purpose

I have always believed design is how you think, not just what you make. That belief matters more now, not less, because the making got outsourced and the thinking is what is left to set you apart.

The good news is that judgment is not a mystical gift. It is a muscle, and abundance is genuinely good for building it, if you use the tools the right way. Generate ten directions instead of one, then have a real opinion about all ten. Use AI to explore, not to decide. Treat the first output as raw material, never as the answer. Say no more often than you say yes. Get close enough to real users that you can feel when something is off before you can explain why.

That last one is the whole game. The teams winning right now are not the ones generating the most. They are the ones with the clearest view of the problem, the sharpest sense of what to leave out, and enough conviction to ship something that reads like a decision instead of a default.

Anyone can make infinite designs. That used to be the hard part, and now it is free. What is scarce, and getting scarcer, is the person who knows which one to ship and has the taste and the spine to stand behind it.

AI made it free to make something that looks like a product. It did not make it any easier to make something worth using. Those were never the same thing. Now it’s the only advantage left.

Ashley Nader is a 0-to-1 product leader and AI builder. She has led AI governance and evaluation products, founder of the AI health startup fini, and has spent more than a decade building at the intersection of AI, product, and design. Learn more at ashleynader.com.

---------

Sources

• Lovable revenue, users, and project volume: Sacra, "Lovable revenue, funding & growth rate" (sacra.com/c/lovable) and TechCrunch, March 2026
• Share of AI app builder users with no coding background; YC W25 AI-generated codebases: Hostinger, "AI app builder statistics 2026" (hostinger.com/blog/ai-app-builder-statistics), citing Hostinger vibe coding data and TechCrunch
• Prompt-to-UI speed gains and design sameness: Drawbackwards, "AI-Generated UI Is Making Every Product Look the Same" (drawbackwards.com)
• Models return the statistical average, not taste: Managed Code, "Why AI-generated websites all look the same" (managed-code.com)
• Feature usage and wasted R&D spend: Pendo, "The 2019 Feature Adoption Report" (pendo.io); Standish Group Chaos research
• On the bottleneck moving: Misfit Labs, "The bottleneck moved" (misfitlabs.vc/blog/the-bottleneck-moved)

‍