The Second Tap

This article is part of a series by Intuiface CEO Mathieu Yerle. In this edition, he explores why the true measure of an interactive experience isn't the first touch—it’s the second. Subscribe to his Substack, Screen Heresy, for early access to his latest posts.

‍

Every week, someone shows me a tool that builds a working app from a sentence. Type a prompt, get a website. Describe what you want, get a functioning interface. The demo always works. The demo always looks like magic.

Then you try to use it for something real, and it falls apart.

These are app builders and website builders - Lovable, v0, the whole prompt-to-app wave. Nobody is showing me a tool that builds an interactive experience for a physical space, because almost nobody is building one. That’s the gap I keep thinking about. Everyone is racing to generate software from a sentence. Almost no one is asking what that means for the screens people actually walk up to.

I’ve watched this pattern before, and not just in signage. Every creative industry that AI touches goes through the same arc. First the hype: AI will replace the creators. Then the disappointment: the output is generic, soulless, wrong in ways that take longer to fix than starting over. Then, eventually, the quiet truth: AI didn’t replace anyone. It removed the friction between an idea and a working version of it.

That third stage is the only one worth building for. And it’s where the interesting argument lives.

The wrong question

The question everyone asks is: can AI make the content?

It’s the wrong question because it assumes content was ever the bottleneck. It wasn’t. The bottleneck was never a shortage of images, text, or video. Walk into any organization with screens in its lobby, and you’ll find more content than anyone can use. Marketing has folders of it. Agencies produced it. The problem was never making more.

The problem was turning content into something a person could actually interact with. That’s where projects died. Not in the design. In the gap between the design and the deployed, working, responsive experience that does something when someone engages with it.

That gap is technical. It’s wiring a screen to a live data source so a product wall reflects real inventory. It’s the state and behavior logic that decides what happens on tap, on idle, when a second person walks up. It’s keeping content current without someone manually swapping files every week. It’s the unglamorous plumbing that separates a beautiful mockup from a thing that works in a museum on a Tuesday with 4,000 visitors.

For years, crossing the gap meant one of two things: hire a developer or commission a long, expensive agency engagement. Either way, the cost of building interactivity was too high for anything but flagship projects. Not because nobody wanted it. Because it didn't pencil out. No-code platforms started changing that math a decade ago - the first real alternative to writing code or outsourcing the whole thing.

The underlord

Here’s a word I picked up from Laura Burkhauser, now CEO of Descript, in a talk on AI strategy. She didn’t talk about AI as an overlord. She talked about an underlord.

It stuck with me because it inverts the thing everyone fears. We keep imagining AI as the overlord - the system that takes the creative work away and leaves us watching. The useful version is the opposite. It works underneath the experience, in the plumbing, handling the parts no human wanted to do anyway.

None of that is the creative act. Nobody got into experience design because they love wiring data sources. The underlord tackles the friction, not the meaning. It frees the people who are good at meaning - story, emotion, design, the reason anyone should care - to spend their time on the part that actually matters.

Why prompt-to-everything falls apart

Here’s the part the hype keeps getting wrong, and it’s not really about AI at all. It’s about how people interact with anything.

Natural language is a brilliant way to express something general. “Show me a layout for a product wall that feels premium and lets people compare three models.” A prompt like that gives a machine room to work, and the result can be a genuinely good starting point. Prompting is the right tool for the abstract, exploratory move.

Now try to use that same prompt to nudge one object slightly left. Or change a single color. Or put this element on top of that one. Suddenly you’re writing a paragraph of awkward instructions to do something your hand could do in half a second. Anyone who has tried to refine an AI-generated image knows the feeling - great for the overall look, miserable for the precise adjustment. The same prompt, run twice, gives you two different results. You fight the tool instead of using it.

This is the second tap. The first prompt gets you something that looks finished. Then you go to adjust it - the second move, the precise one - and the tool that felt like magic a moment ago turns into a fight. The demo only ever shows you the first tap.

This isn’t a flaw you can prompt your way out of. It’s a property of the modality. Natural language, used as a prompt, is built for the general and the abstract - it gives a model room to interpret. Language can be precise when a human reads it, of course. But a prompt isn’t read, it’s interpreted, and the more exact you try to be, the harder you fight the gap between what you typed and what came back. Direct manipulation - pointing, dragging, clicking, touching - is built for the specific and the precise. Forcing one to do the other’s job is the actual mistake hiding inside “generate the whole experience from a sentence.”

The interesting future isn’t prompt-only or editor-only. It’s the combination. Language to rough out the shape. Direct control to make it exact. Maybe both at once - point at a thing and say what you want done to it. We already learned this lesson once, the slow way, watching voice-only interfaces try to do everything and fail at the precise parts. The tools that win won’t pick a modality. They’ll put each one where it’s actually intuitive.

What changes when the gap closes

Think about what becomes possible when the cost of building interactivity drops by an order of magnitude.

The flagship museum installation stops being a once-a-decade capital project and becomes something a mid-sized institution can attempt. The retail experiment stops needing a six-figure agency budget and becomes something a regional brand can test in one store. The corporate briefing center stops being a thing only the Fortune 100 can afford.

The reason most physical spaces still broadcast is economic, not philosophical. As I argued a few posts back, the technology for interactivity has been here for years - touch is commodity hardware, sensors are cheap. The barrier was never the screen. It was the cost of making the screen do something worth doing.

That cost has been falling for a while. No-code tooling already closed much of the gap between an idea and a working experience, long before AI entered the picture. That’s worth being clear about, because the easy version of this argument - “AI finally makes interactivity possible” - isn’t true. Interactivity was already getting cheaper to build. What AI does is push the cost down another order of magnitude, and collapse the slowest part of the work: getting from a blank canvas to a structured first version.

Lower the cost that far, and interactivity stops being a luxury reserved for the one space a company spends real money to impress people. It becomes the default. That’s the shift. Not better content generated faster. The barrier that priced most organizations out of interactive projects, finally low enough that the experiment is worth running.

Where the hype gets dangerous

I want to be honest about the risk, because the underlord frame can be abused.

The danger isn’t that AI does too little. Getting from zero to one is a real gain, and worth saying so plainly - a tool that turns a blank canvas into a working draft in minutes is a genuine productivity leap, even when that draft isn’t the finished thing. The trap is what’s underneath the draft. Most prompt-to-app tools optimize for the demo, and to hit it they scaffold in whatever way is fastest - ad hoc data models, improvised structure, code and logic nobody would have designed on purpose. It works for a prototype, then collapses the moment you try to scale or maintain it. This is the problem I spend my days thinking about at Intuiface: generation is only worth having if what it produces is structurally sound - built by agents trained on years of architectural best practice, not improvised to win a demo. The prompt-to-app world is full of companies racing on speed. Far fewer are asking what happens after the prototype, and almost none are asking it about physical spaces and digital signage workflows - which is exactly where we’ve chosen to plant ourselves. The question was never whether AI can get you to one. It’s whether what it hands you at one is built to become two.

There’s a second danger, the one that connects this back to everything else I’ve written. If AI in physical spaces becomes a machine for generating more passive content faster, we will have used a genuinely transformative tool to make the original problem worse. More screens with nothing to say. Just produced at lower cost. We’ve watched one industry already learn that lesson - content nobody chose to make, generated to fill a space, gets the engagement it deserves, which is none.

The point was never speed. Speed that produces forgettable work, or throwaway structure, is just a faster way to be ignored. The point is removing the friction that stopped good ideas from ever getting built - and building them in a way that can actually last.

The line worth holding

AI doesn’t replace the creative act. It removes the reasons the creative act never made it out of the room.

Keep it underneath the experience, doing the work nobody wanted, and it’s the most useful thing to happen to this industry in a decade. Put it on top, generating the meaning and forcing language to do a job that belongs to a human hand, and it’s just a faster path to a wall full of screens that nobody looks at.

The screens were always the easy part. What happens when someone stands in front of one is still the whole game. AI doesn’t change that. It just makes it cheaper to get the answer right.

‍