Speaker A
I spent the last decade designing apps and products. And in the last year, everything changed. Apps all of a sudden became easier than ever to create, but harder than ever to differentiate. And what I realized is that the app space essentially got compressed to four distinct levels. Level one, where most founders start and get stuck. Level two, a level surprisingly few founders graduate to. Level three, where lots of founders end up and where everything kind of goes sideways. And level four, a place most founders never reach, but also a place where if you do reach it, your app is far more likely to be competitive. And in this video, I'm going to walk you through all four of these levels in detail. And I'm going to explain what it takes to reach them by building a wake-up alarm app four different ways with AI. To make sure we're comparing apples to apples when evaluating the apps, we're going to use a scoreboard with three different dimensions, including one, point of view. Does the app actually stand for something or is it just kind of there? Two, consistency. Does the app hold together like one system instead of five different apps stitched into one? And three, execution. Is there any interesting product craft or thoughtful execution in this app at all? Each app and level will get scored on these dimensions from one to five. Starting with level one. Again, this is where most founders begin. You type a prompt, you hit go, and the first draft looks something like this. And honestly, it's usually pretty decent. This one is clean. If you had shown me this just a couple years ago before the AI era, I would have told you a real designer made this. But since we're in the age of AI, it now looks like a dozen other apps you've come across in the last week. And the reason is this. There's no decision in here. Nobody sat down and deliberately chose what this should be. So when a founder tries to push it, saying make it premium, make it more modern or make it feel like Apple, it moves a little, but it never really blows you away. And this is clear if we bring out the scorecard point of view. Not much to talk about really. So a one. Consistency actually pretty solid. It's a clean, tidy app. This is what AI is decent at today. So a three. But product execution, it's still a very basic app pattern. Nothing revolutionary going on here. A one, which nets out to five out of 15. Pretty clean but completely empty. Nothing that really makes you stop. And again, this is the level of apps we see everywhere today. What we call AI slop. Right now, here's the interesting thing. Every time a new model drops, this level one gets a little better. The base looks a little nicer. The slop is a bit less sloppy. So, it's easy to think, give it a year or two, and this gets so good that nobody needs an eye for design at all to achieve great-looking apps, and I get why you'd think that, but there's a study going around right now made by Contra that shows something important. They put the best AI models head-to-head on design, and the ones with the best taste won easily, but on vague briefs. The second someone handed them a real specific creative direction, those same models lost to a model that just followed the direction really well. This tells us that the real ceiling is the direction. And it's something I see personally every day working with early stage startups. The better the models get, the more that direction is worth. While a floor raises for everyone as the models improve, the ceiling flies away at the same time. Because with new models, we can continue pushing what's creatively possible. So this tells us a clear story. The real move and the real moat isn't waiting for a better model. It's giving the AI better creative direction. And that is exactly where level two comes in. At level two, you stop leaning it to chance. You start handing the AI real design vocabulary. A few skill files, maybe a basic design system, a way for it to understand what you actually want and maybe even why. And that works. Watch this screen tighten up. The spacing gets more deliberate. The typography use is cleaner. The app just looks a bit more intentional overall. It's still a similar basic prompt, but now the AI has more honed-in direction to work with. Here's the catch, though. Unless you've got a great eye for design, it only really gets better by a couple of percentage points. And it's a bit of a gamble, too, because AI is unpredictable. Sometimes you nudge it and it really lands. Many times you just want to punch your freaking computer screen, right? But let's score it and see where we land. Point of view. It's leaning somewhere, but still not really differentiating. A two. Consistency. This is where it climbs. Even cleaner and more deliberate. A four. Surprisingly good actually. Product execution. Slightly better, but still no out-of-the-box thinking. Another two. Call it eight out of 15. A real step up from level one. But you can still feel it. Nobody art directed this. There is no taste. So let's not stop here. Let's give it even more. More reference images, world-class apps to pull from, a whole design system to build on. Let's pump it full of everything we've got and see how far that takes us. That's level three. Every scrap of context you can find all at once and you just let it cook. And at first glance, it looks like more is happening. It's flashier. To an untrained eye, you might genuinely think it got cooler. But let's break it down together. Now, all of a sudden, components don't really match each other anymore. The spacing goes haywire. Same with font sizes. The whole system starts coming apart at the seams because we're forcing the model in too many directions. And nobody is in charge of a single one of them. It's what I call a Frankenstein, which shows in the scoring point of view. Level two had it leaning somewhere and got two points, but crammed in this much at once and every direction cancels out. So, it washes right back to a one. Consistency is broken. This is the one thing previous levels actually did a very good job at and now it's gone back down to a one. And for execution, not really worse than level two, but not better either. Still no real product decision anywhere. So a two. That adds up to four out of 15, telling us that level three doesn't just stall. It drops below level one. And this is not a small point because it shows us that even if we give more direction to AI, that doesn't guarantee us a great result. Past a certain point, more context and direction can make it objectively worse. And it's funny because this is the exact opposite of how it works with a human. When you have a person with real judgment, more context, like more info about the business, the customer, the goal, then the result typically gets sharper, not messier. But as we see here, when we do the same with AI, everything can very easily break. So what do we do to overcome this mess? Well, meet level four. Now, for the first time, you put a person with a vision behind the wheel and not to add more. No, no, no, no. To decide what to cut, what this should feel like. This is where it stops being about just looking good and starts being about character, a personality, a voice, something that looks like it was actually handcrafted with intention. That is what true art direction really is. And it's the first thing AI genuinely cannot bring because it doesn't have a point of view. It just has an average. And if you don't steer it, it will always stay average. It will continue looking like all the other apps out there that are built without direction based on the same models with the same averages. And like we touched on before, even though the models raise the average all the time, it still stays the average, the ordinary, what everyone else has access to, no matter how good it gets. So this gives you as a founder a couple of choices if you want to level up. Option one, learn it you
