Mei-ling: What if an AI could build an app, write the whole thing, but then it could look at its own work and say, "Hmm, this isn't very good." Or, "This has bad taste." Can you imagine that?
Marcus: An AI that just... codes? Like, the whole thing? From scratch?
Mei-ling: Not just codes, Marcus. It builds an app, yes, but then it looks at its own work and says, "This isn't good." Or, "This has bad taste."
Marcus: Whoa, okay, wait. So, a critic? Like, a design critic? That's—that's next level.
Mei-ling: Exactly. Because Anthropic, they built this system for long-running app development, and the real secret, it's not just one super-smart AI model.
Marcus: No? What then?
Mei-ling: It's like... they built a whole team of AIs, working together.
Marcus: A team? What does that even... I don't know what that means.
Mei-ling: They have a 'planner' AI, right? That breaks down the big project.
Marcus: Okay. So, like, the project manager?
Mei-ling: Yes. Then a 'coder' AI, it actually writes the code.
Marcus: Right. The dev.
Mei-ling: But then, the crucial part: there's a 'critic' AI. This critic, it has... what they call 'taste.'
Marcus: Taste? You mean like, if it looks good? Not just if it works?
Mei-ling: It evaluates the design and the code against specific principles. Saying if it's 'good' or 'bad' or 'cheap-looking.'
Marcus: So it's not just checking for bugs, it's checking for aesthetics? That's kinda wild, Mei-ling.
Mei-ling: Yes, exactly! And the whole thing, it's inspired by deep learning, Generative Adversarial Networks. You know, GANs.
Marcus: Oh, yeah, I've heard of those. Like, for creating fake images, right?
Mei-ling: Precisely. One network tries to create something, the other tries to tell if it's fake. Here, it's a 'generator' agent creating code, and an 'evaluator' agent critiquing it.
Marcus: So it's like they're fighting each other?
Mei-ling: Yes. This adversarial loop, it forces the generator to make better, more... 'tasteful' output.
Marcus: That's... that's like what I heard about how high-end game studios work. Like Naughty Dog, when they were making The Last of Us. It wasn't just one big team. They had these tiny 'strike teams.' One group was just on the feel of the melee combat, you know, the punch. Another group, just the AI of the Clickers. All these little specialist groups, hammering on one specific problem.
Mei-ling: Hmm. Like little AI strike teams for software. I never thought of it that way, but it's very similar, isn't it? Mimicking how human teams collaborate.
Marcus: Yeah, it's like, you can't just throw one big brain at it. You need a bunch of little brains, each doing their own thing, but also checking each other's work.
Mei-ling: So for Anthropic, it's not just about making one AI super smart. The real breakthrough, I think—
Marcus: Yeah?
Mei-ling: It's how they design the system around it. Giving it structure. That's it. It's not the AI, it's the system.
Marcus: Structure... and maybe a little bit of attitude, if it's telling you your app looks cheap.
Mei-ling: Exactly. A little bit of... critical spirit.
Marcus: Hey, did you see this data from Ramp?
Mei-ling: Oh, the Eric Glyman piece? 'Getting on the right side of the ice'?
Marcus: Yeah, that's the one. He's basically saying this economic gap between companies using AI and those who aren't? It's not some future thing. It's happening right now. And it's widening like crazy.
Mei-ling: Mmm, I read that. It sounded... quite dramatic. Like Shackleton and the ice floe.
Marcus: Dramatic, sure, but it's real dramatic. Because he's got the data, Mei-ling. Ramp, their platform, it sees the spending and revenue data of fifty thousand plus companies. And what they're seeing is the top quartile of AI adopters have, get this, more than doubled their revenue. Doubled!
Mei-ling: beat
Mei-ling: Doubled? That's... significant.
Marcus: Yeah, Mei-ling. While the bottom quartile is just... flat. And it ain't just tech companies. He talks about, like, a roofing company in Texas, they're up twenty-something percent with AI for estimates. A window installer in Utah, up almost sixty percent.
Mei-ling: But is it the AI causing the growth, or is it that the companies already well-positioned to grow are also the ones adopting AI? You know, the good, well-managed companies naturally become the early adopters. It could be correlation, not causation.
Marcus: Nah, I don't think so. My friend's dad, he runs a small accounting firm in Dayton, Ohio. Super traditional. Forty years he's been doing it. Always said AI was just hype, trusts his people, right? But last year, they lost two of their biggest clients. Two! To a new firm that uses AI for automated bookkeeping. They can offer the same services for thirty percent less. Thirty percent! Now he's scrambling.
Mei-ling: That's... that's a tough story. But still, it's hard to definitively say the AI is the cause of the success for everyone. Maybe it's just a tool for efficient companies to become even more efficient, and leave the others behind. It's not creating new markets, Marcus. It's—
Marcus: But the gap is real, Mei-ling. He calls it a 'widening chasm.' And it's happening in these super traditional industries. Like, if you're not on the right side of this ice, you're just getting left behind. Fast.
Mei-ling: I... I agree the divide is real. But the mechanism... how do I say this? Whether it's truly additive, or just amplifying what's already there—that's still fuzzy to me. The roofing company, they were probably already looking for efficiencies, and AI was the next tool. It wasn't the magic wand.
Marcus: I mean, if it's an amplifier, that's still a problem for the ones who aren't using it, right? Amplifying the gap is the same as creating it, when you're on the wrong side. It's a zero-sum game for the market share. You get it, or you don't.
Mei-ling: Hmm. I need to think about that more. It's not as simple as 'AI makes you money.' It's... it makes somebody money. That's the part. It takes from... you know?
Marcus: I saw this little gem about why ninety percent of these 'vibe coded' apps look cheap.
Mei-ling: Oh! I think I've seen that one. It's like, you know something is off, but you can't quite put your finger on it.
Marcus: Exactly! The author says it's two main things: fonts and spacing. Pick a font that matches your app's purpose—serious for finance, friendly for consumer. And then be ruthlessly consistent with your spacing. Like, use a 4px base unit, and only multiples of that. No weird 13px or 22px spacing.
Mei-ling: That's... so simple.
Mei-ling: beat
Mei-ling: And yet, I can think of so many apps that just feel off because of exactly that.
Marcus: Right? These small details, they make or break the whole 'vibe.' It's like—
Mei-ling: Wait, Marcus. If this economic divide is really widening—
Marcus: Yeah?
Mei-ling: What happens in five years to all these small businesses? The local accountants, the contractors... do they just get competed out of existence if they don't jump onto the 'right side of the ice'?
Marcus: I don't know, Mei-ling. It's kinda terrifying, honestly. Like, you see it happening, and it's not even a slow fade. It's just... this crack, it forms, and then it's too wide to jump.
Mei-ling: Too wide. Yeah. And then what? You're just... gone? No choice. For many.
Marcus: Yeah. And that's the thing that really stays with me, from that whole piece. It's not about being the best at AI. It's about not getting caught on the wrong side.
Mei-ling: That's it.
Marcus: And that's me, Marcus.
Mei-ling: And I'm Mei-ling.