Suhail Gupta wrote an essay (opens in a new tab) earlier this year telling designers that taste won’t save them from AI. His claim is broad: AI will soon have better taste than you do, and designers are fooling themselves by betting their careers on it.
Look at his evidence, though: sign-up pages, forms, hierarchy, spacing, the best of what’s already been shipped. He’s not wrong about any of it. He’s just fighting a battle that ended more than a decade ago.
I also wouldn’t call what’s being lost there taste. It’s convention: the steady leveling-up of patterns people already agreed on. Someone decided what a good sign-up flow looks like. That decision got rolled into UI kits and pattern libraries, and now into AI. Humans still made it. You just don’t have to make it again every time.
That’s what’s really happening. The universe of things already decided keeps expanding, and the work designers do keeps moving out to its edge.
We solved this years ago
Ten years ago, if you needed a solid sign-up flow (if you knew what you were doing), you didn’t invent one. You opened one of a dozen pattern libraries and picked the best one. The same went for forms, onboarding, checkout, settings screens and empty states. The best answers to the common problems were already documented, compared and free to copy.
My own teams went further. We kept internal Docker setups loaded with the front-end libraries, components and fonts we used over and over. Some of those libraries we designed ourselves; some came from the community. Either way, nobody re-decided a form field from scratch on each project.
If you needed a simple, good-looking marketing site, you picked a Squarespace template. It took about thirty seconds. A WordPress theme took not much longer. A UI kit gave you a polished system of pages and components for a few dollars.
Here’s the part nobody mentions: picking a Squarespace template is still as fast as getting AI to build a simple marketing page. You prompt, review, re-prompt, fix the spacing, swap the fonts, regenerate the hero. It’s impressive, but it isn’t quicker for building something simple than choosing from a gallery of finished options.
So AI didn’t take this work from designers. It was already gone. Kits and templates took it long before AI showed up. AI has just made it impossible to pretend otherwise.
Why we keep fighting it anyway
If the battle ended years ago, why are we still having it?
Partly because “taste” has been a comfortable story. Designers talk about it as a gut feeling, an eye you either have or don’t. Gupta pokes at this, pointing out that most designers’ taste is borrowed from a few dozen well-known apps. I’d go further.
The mystery behind taste is solved. Show me a person’s background, the work they’ve seen, the projects that succeeded and failed, the people who trained them, what they were rewarded for, even their temperament, and we can deduce why they land where they land. It’s inputs and mechanism. We call it instinct because we can’t see the process from the inside.
That’s exactly how we talk about AI. Show me a model’s training data and I can tell you how it thinks. Show me a designer’s history and I can tell you the same about them. Both are amazing. Neither one is magic.
That doesn’t make either one worthless. Explaining the mechanism removes the mystery, not the value.
But once you see taste as mechanism, defending it on sign-up pages looks like what it is: holding onto a craft the market moved past years ago. Nobody writes essays defending darkroom editing. It’s a hobby now: real, even beautiful, but not where the photography industry works. Arguing about whether a human eye matters on a login screen is the same mourning, just late.
Where the real battle is
So if the conventional layer is settled, where’s the fight now?
It’s in the work that doesn’t have a pattern library yet.
Gupta’s own argument points here. He notes that AI learns mostly from shipped work, designs that have already been refined over many rounds. That’s why its baseline is higher than people expect. It’s also why it’s so good at conventions, the same way the old kits were. It’s great at opening up options you wouldn’t have considered, too. What it can’t do is point to a proven answer for something nobody has shipped yet.
This has happened before. I came up in the heyday of Flash, a few years before CSS Zen Garden changed everything. (Shout out to Dave Shea.) Many of the people I worked alongside came from print, and they were skilled at it. But they couldn’t get their heads around designing a digital system: a set of rules that would generate pages nobody had laid out by hand. To them, a page was something you composed. Designing the rules and letting the pages follow felt like giving up the craft.
It wasn’t giving up the craft. It was the new craft, and it took more judgment, not less. The people who made the jump stopped designing pages and started designing the systems that made them.
Each jump didn’t just make the old work cheaper. It created new problems that nobody had a template for.
One trend I’m watching, and it’s only one of several, is headless content evolving into agentic systems that know everything a brand or organization believes about itself: what it stands for, how it speaks, what it will and won’t promise. Those systems won’t just publish pages. They’ll talk to other systems, including other organizations’ agents, under the surface, where nobody ever sees a layout.
Nobody has an established pattern for that yet. There’s no pattern library to pull from, and no AI can tell you definitively how to do it, because the answer isn’t in its training data. Pieces exist, but it hasn’t been shipped at scale. Deciding what an organization believes, encoding it so a machine can act on it, and setting the rules for how it represents itself to other machines is design work. It’s the same leap my print colleagues faced, one level up. Then it was moving from composing pages to designing the rules that made them. Now the audience on the other end is often another system.
The contradiction at the end
Gupta closes on ownership and agency. Stop worrying about which role AI replaces first. Dissolve the lines between design, code and product. Own the whole product. Be the person in control of the agents.
I agree with nearly all of that. I’ve worked across design and development my entire career, and the most valuable people I’ve hired were never the ones who stayed in their lane. They were curious, self-directed and willing to own the whole problem.
But along the way he says skill will become “as immaterial to quality output as taste.”
That can’t be right if the answer is to be the one in control of the agents. Controlling the agents takes skill. Knowing which of thirty generated options deserves to exist, spotting where the output went wrong two decisions back and steering it somewhere better: that’s judgment, and it matters most exactly where there’s no pattern to fall back on.
And it doesn’t stop at one layer. Agents will start directing other agents, and then the question becomes who directs that one. Every time a layer gets automated, the judgment moves out to the next edge. There will always be an edge.
What’s actually left
Strip the mystery off AI, and strip it off human talent too. What remains is someone deciding what should exist, why it should exist that way, and being accountable for the result.
That was always the job. Templates and kits took the parts that had already been decided. AI is taking more of them, faster. What it can’t supply is the deciding itself, because it has no point of view of its own. The person in control of the agents brings that, or nobody does.
So Gupta is right: taste won’t save you on the sign-up page. It hasn’t for a decade.
The question is whether you’re still fighting for the sign-up page, or working on the problems nobody has solved yet.
Written by
Micah Slavens
Principal, Lift
