Ohneis - The Pattern
Tech, AI, design, and the hidden logic behind it all.
Ohneis - The Pattern
Why Typing to AI Is a Design Failure | with Lauren
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Technology moves fast. Design makes it matter. AI changes everything. This is oh nice. Try something for me right now. Try to describe a spiral staircase out loud. But you cannot use your hands. No gesturing, no pointing, no drawing in the air. Just words. Now imagine typing that description so precisely, so perfectly, that a blindfolded person could draw it on their first try. Every curve, every step, the exact proportions. Brutal, right? And yet, that is the interaction model we've been handed for the most powerful technology ever built. A tiny blinking text box. We've been told this is the future of computing. We've been told to get better at prompting, that if we just learn the right words, the machine will finally cooperate. But here's what almost nobody in the industry wants to say out loud. The chat box isn't a feature. It's a regression. It is, functionally, the 1970s command line with better marketing. Today I've got Lauren here, and we're going to break down why the text prompt is dying, what's replacing it, and why none of this was ever your fault. Lauren, welcome.
SPEAKER_01Thanks for having me. And honestly, that spiral staircase thing, I felt that in my chest. Because I've been there. I think most people have been there. You're staring at a blank text box, you know exactly what you want, and somehow the harder you try to describe it, the worse the output gets. And what gets me is that nobody really questions it. Like the blank box showed up and we all just accepted it. We started buying courses on how to talk to a machine correctly, which when you say it out loud is kind of wild.
SPEAKER_00And that's the thing that I keep coming back to. We didn't accept bad menus at restaurants. We didn't accept bad door handles. There's actually a whole field of design built around the idea that objects should visually communicate how to use them. It's called affordance. A door handle shaped like a bar tells you to push. A coffee mug handle tells you to grab it. A blank white text box tells you absolutely nothing. It gives you zero information about what the system can do, what it needs from you, or whether you're even asking the right question. Don Norman, the guy who basically wrote the book on human-centered design, called this a design failure. And AI just rolled it out at global scale and called it a product. The thing I find fascinating though is this wasn't malicious. It wasn't a conspiracy. It was something much more mundane. These tools were built by software engineers, people who for decades have communicated with machines through precise written commands. For them, a text interface is completely natural. It's their native language. So when they built the first AI products, they built them for themselves and then handed them to the rest of the world. And actually, if you want to go deeper on what happens when you give these AI systems more autonomy than just answering a text box, we did a full episode on that with Nigel called The AI Intern Unchained. Worth a listen after this one.
SPEAKER_01Right, and I think the engineer bias thing is so important because it explains so much of the friction. I mean, I work with a lot of creative people, designers, writers, marketers, and they all describe the same feeling when they use these tools. It's not this is hard, it's more like, I feel stupid. And that is such a red flag for any interface. If your users feel stupid, that's on you, not them. And the dual coding thing really explains why. Our brains literally have two separate tracks for processing information, one for language, one for images, and they run in parallel. When I glance at a rough sketch of a layout, I understand it in milliseconds. My spatial brain just gets it. But when I have to convert that spatial understanding into words, I'm doing a translation job in real time, and something always gets lost.
SPEAKER_0013 milliseconds. That's how fast the human visual system can process an image. 13. You want to know how long it takes to read and parse a 200-word prompt describing the same image? Seconds. And even then you've lost half the information. The exact shade of color, the spacing, the feeling of the thing. There's a reason architects use models and not essays. There's a reason surgeons use scans and not verbal reports. And yet we looked at creative work, design, layout, photography, video, and said, no, let's handle that with a paragraph. Lauren, you mentioned the feeling stupid thing, which I want to push on a little. Because I've heard the counter-argument and it's worth addressing. Some people say, and smart people say this, that text is actually more precise. That the well-written prompt gives you more control than clicking and dragging. Is there any version of that that's true?
SPEAKER_01Hmm. I mean, yeah, in some narrow contexts, if you're asking an AI to rewrite a sentence or summarize a document, text makes total sense. Words in, words out. Totally fine. But the moment you're working on anything visual or spatial, a layout, a photo edit, a UI design, text just collapses as a medium. You're trying to play Pictionary over a walkie-talkie. You lose so much in translation. And the precision argument gets used a lot by people who are already good at writing prompts, which is a self-selecting group. If you've spent months learning how to prompt well, of course you're going to defend it. But the average person who just wants to make a decent-looking slide deck, they shouldn't need to learn a new dialect to do that. That's the thing that bothers me most. Prompt engineering got rebranded as this cool futuristic skill. It's on job descriptions now. And in reality, it's a workaround. It's like being proud that you've memorized the bus schedule because the app is broken.
SPEAKER_00The locked-out director. That's the frame I keep using for this. Imagine you're directing a film. You know exactly what the scene should look like. You can see it, but you're not allowed on set. You're not allowed to point at an actor and say, move here, do this. Instead, you have to slide written notes under a locked door, and then hope the actors interpret your tone correctly. That is prompt engineering. And people spent the last three years perfecting their note sliding technique and calling it a career. The shift that's starting to happen now, and this is what makes 2025 and 2026 genuinely interesting, is that tools are starting to put you back on the set. You can point, you can draw, you can show the AI directly what you mean instead of describing it in a language it was never designed to process visual ideas through.
SPEAKER_01And the tools are real, like this isn't vaporware anymore. TLDraw, which started as this simple whiteboard app, has basically become a canvas where you sketch something rough and the AI fills it in. Figma Make is letting designers describe interactions visually and have them generated. Adobe's generative fill lets you circle a region of a photo and say, change this without writing a novel about it. And then there's the node graph stuff, where instead of writing a prompt, you literally draw a flowchart of what you want to happen. Connect idea A to idea B, add a condition here, drop in an image there. It looks like a map, not a text document. And suddenly, non-technical people can build complex AI workflows because they're thinking visually, which is how most people think naturally.
SPEAKER_00The node grafting is interesting to me because it's actually quite old as a concept. Musicians and video editors have used node-based tools for years. It's just that nobody connected it to AI until recently. But I want to ask you something that I think is the harder question here, because there's a skeptical reading of all this that says, okay, visual interfaces are more approachable, but approachable doesn't always mean better. The 1984 Macintosh was easier to use than MS-DOS, but it was also less powerful for experts. Is there a version of this visual AI canvas future where we trade depth for ease? Where we make things feel more natural, but actually get less out of the technology?
SPEAKER_01That's a really fair challenge, and I don't think the answer is simple. I think there's a real risk of that, yeah. If visual tools become so abstracted that you have no idea what's happening underneath, that can create its own problems. You lose the ability to debug, you lose understanding. But I think the historical comparison actually cuts the other way. When the GUI replaced the command line, experts didn't lose power. Photoshop isn't less powerful than the command line tools it replaced. It's more powerful and more people can use it. The underlying capability expanded because more people could access it. And I think that's what's about to happen here. The people who could already prompt well aren't going to lose anything. They're going to get visual tools on top of their existing knowledge. But the millions of people who gave up on AI because the text box made them feel inadequate, they're going to show up. And when more people can actually use these tools, the quality of what gets made goes up across the board.
SPEAKER_00The menu analogy keeps coming back to me for this. A restaurant menu doesn't limit you, it shows you the range of what's possible. A blank piece of paper with a pen isn't freedom, it's anxiety. And I think that's the core psychological insight that the first wave of AI companies missed. They confused open-ended with powerful. They thought giving users a blank box was giving them infinite possibility. What they actually gave them was infinite uncertainty. And uncertainty is exhausting. There is a reason supermarkets spend millions on product placement and shelf design. There is a reason every app you use on your phone shows you buttons and icons instead of a command prompt. Humans navigate by seeing options, not by inventing them from nothing. The tragedy of the last four years is that the most capable technology ever built was handed to people through the least intuitive interface ever designed. And then we blamed the people.
SPEAKER_01Yeah. And I think that's what I want people to take away from this. If you've ever felt frustrated with AI, if you've ever spent 20 minutes rewriting a prompt and still gotten something wrong, that wasn't a skills gap. That was a design gap. The technology was genuinely hard to use. Not because AI is hard, because the box they put it in was hard. And that's changing. Slowly, messily, not all at once, but the course correction is real. The industry is finally asking the question it should have asked in 2022. What does it feel like to be the person on the other side of the screen?
SPEAKER_00We had the most capable tool in human history, and we put a blinking cursor in front of it and called it an interface. Lauren, this was a real conversation. Thank you for being here, for pushing back where it mattered, and for making the node graph thing finally make sense to me. Genuinely appreciate it. If you're listening and this landed, subscribe, leave a review, share it with someone who's ever given up on AI because the text box broke their brain. The chat box isn't the future. It never was. It was just the fastest thing they could share.