The Problem with Vibe Coding No One Talks About

There's a lot of videos you can see of people doing vibe coding, which is basically, you say, like, build me this landing page. You hit enter, you pull out your phone, you start scrolling for two minutes while it does its thing. You look, doesn't work. You hit please fix, please fix, please fix. AI engineering, or AI augmented code generation is much more like changing with how we interact with the computer, but your mind is still what's dictating the logic of what's happening.

I think I've heard you say that, you know, no code tools can be erased to the bottom. That's a kind of a bold statement. You know, why do you think that traditional no COD code tools won't scale for real businesses? You guys have become, you know, experts in this field that you're in, and so. But yet there's still a lot of confusion. So I'd love for you guys to talk about some of the trends and terms that we see. You know, there's. There's obviously agentic, which is the name of this podcast, but yet we hear people using AI agents even still to mean, like, a chatbot. Somebody builds a chatbot and they call that an agent. But people are confusing agent and agentic. Then. Now we have this new trend with vibe coding. And I wasn't sure if what you were talking with was if you would consider that vibe coding and how. What is vive coding and how that differs from, like, AI engineering.

So there's a lot of videos you can see of people doing vibe coding, which is basically you go to one of the no code tools you use, cursor, wind surf, whatever it is, and you basically say, like, build me this landing page. You hit enter, you pull out your phone, you start scrolling for two minutes while it does its thing. You look, doesn't work. You hit please fix, please fix, please fix, right? And you kind of like trying to understand what happens a little bit, saying, like, oh, this button doesn't work. Make it happen. But it gets in these loops and it just generates a ton of, like, spaghetti code. Eventually you might get something that works or you might have to kind of restart. So I'd say that's what I would define as vibe coding. I'd say AI engineering or AI augmented code generation is much more like changing with how we interact with the computer, but your mind is still what's dictating the logic of what's happening. So one of our engineers told me this the other day. His best analogy for it was, like, back in the day, you know, you had to Use punch cards to code with. Right. So like that was the modality with what you interacted with the machines. And then we switched to typing and IDs and all of that stuff. And now it's like, all right, we can use the AI to be what writes the actual code, but you as a human still need to be building that understanding. Like your understanding is what coding is like being the details of what is happening behind the scenes of exactly what you're instructing the computer to do. So I think that is AI augmented engineering and that is what we do. All of our engineers, they use all these tools super extensively, but every single line of code that is written by the AI is directed by the human and then reviewed by not only themselves, but someone else. So as the engineer, you are still responsible for making sure that the code you write is high quality, maintainable, understandable, all of those things.

Brad, how would you describe that process of like a user interacting with laws and like, and you know where that falls between, you know, a no code tool, engineering, like traditional engineering by coding, what would you describe that it is emulating?

The experience of talking to a technical co founder, right. Like as someone who's non technical, the details of how it's implemented, which database they use, you know, what the data model looks like, you know, what technologies are used under the hood, what languages is like. Frankly you don't care, right? You just want it to work and you want it to be something you feel confident putting in front of a customer and you know that if something breaks, someone is there who knows what, what the system is to fix it. Right. And so like that, that I think is the experience we're trying to give our, our founders on our platform is using it as if they have an engineering team that they are directing what to do.

That makes sense. And then I guess continuing this discussion around, you know, no code tools and I, I think I've heard you say that, you know, no code tools are, can be erased to the bottom. That's a kind of a bold statement, I guess. You know, why do you think that traditional no code tools aren't won't scale for real businesses? And, and what are the, the true moats, the moats and defensibility for AI kind of code gen tools?

My high level thinking on this is that like if you're letting the LLM do basically the entire coding of a project right from scratch, it chooses what to do it. You know, you're just messaging with it back and forth. You put a little bit of sort of Scaffolding to do web hosting or mobile app hosting, whatever around it. The differentiation you see between other companies who are doing basically the same thing just kind of goes down and down. As we've seen, there's like tons and tons and tons of companies similar to the Lovable and Bolt that are being spun up kind of every day. And the way I see it is like, people right now go and test all of them, right? They spin off, they go to five different sites, they type in the same prompt, they hit go, and then they see which ones they like the most, and they take that one, and then they kind of repeat the process over and over again for each step that they do. So it's a super interesting phenomenon. It's the same way with kind of like the model providers right now. We switch to the best model provider every day, whichever one is the best at that given moment. It's kind of the same with these tools. There's really no moat in sticking to them as you can, you know, you can take your code and move it to the next one, or you can spin up a new project that's a copy of your old one and maybe slightly better because the other team has figured something out. So it's a really interesting, like, how do they create a moat? How do they build a real business that's sustainable? As you know, the LLMs get better, and it just becomes easier and easier to spin up a full copy of what they're doing. So that's kind of my thought on why I've said that. I think it is a race to the bottom, and there'll just be thousands of people doing the same things, maybe slightly customized to tiny, different niches. But beyond that, unless they build some serious tools to allow people to build real businesses, I feel like it is a race to the bottom on those tools.