The New Age of NoCode

Hello, Ray. How's it going?

Hello, doing well, man. Glad to see you again.

Today we're talking about the New Age at the future of no-code? Does it even have a future? Because everything I'm seeing, if I'm on X, if I'm on the internet, I'm seeing vibe coding, AI. I saw some random, I think it's AI-made ad of Kevin O'Leary advertising an AI tool, but his voice was kind of funky. So I don't know if it was really Kevin O'Leary, but it was someone using Kevin O'Leary to advertise a vibe coding tool. It sounded like, if you can say a sentence, you can build an application. It had this AI voice vibe to it. I don't know if he just talks like this or if it was AI, but he was telling people to vibe code, or his avatar was. Vibe coding is everywhere. It's being sold on sketchy landing pages for $200 a month, now $29.99, and all those kinds of things like, you know, those ClickFunnels type of pages. Yeah, like have some sketchy vibe to it. Vibe coding is everywhere. It's being sold to everybody and their grandma. So is no-code dead, or is that the new no-code? What are you seeing from someone who has been doing this for, uh, way longer than I'm old.

That, that, that, that was, that was definitely done to sort of call me old right in there. Yeah. Um, yeah, I don't know, it all just sort of keeps on coming around. Like, the, the new stuff is always full of grift, kind of like we were talking about. Like, Brunson literally calls those ClickFunnels pages squeeze pages. Um, and the, um, But the question like, what's, you know, what does dead really mean? You know, like I think there's, when I think about like, you know, the world of, you know, low-code, I think about like two forces coming together, right? We've talked about this a little before. Like one is the changing quality of the tools, right? And then the tools that allow people without formal training to be able to create meaningful software, right? And the tools have evolved so that a bigger and bigger, bigger set of people are able to use them. But that is helped by the second trend,, which is demographic trend. Uh, like if we were having this conversation about like low-code tools when Mendix first came out 20 years ago, uh, the, uh, we would have, you know, they were trying to make it possible for people who were at the peak of their careers, call it in their late 30s, um, people who had gone to school, uh, in the late 1980s, right? Who were in like in college and like the late 1980s, um, you know, they want to make it accessible to them. And that's a very different level of technical understanding. Like people who went to school then, their experience with like big clunky IBM PCs, green screens, what have you, starting to get into, you know, people who might have been early adopters of Macintoshes at that point, you know, back when they were small, you know, sort of friendly but had tiny screens. But the thing that's happened is that as computers, as, you know, as software has eaten the world, right? More people have learned how to code early in their development. It becomes part of school curriculums, it becomes part of college courses. And like a pretty common thing I hear now is somebody who says, who's similarly at the peak of their career in their 30s, whatever, and they say that coding isn't what they do for their job, but it's something they did back in school, right? They have familiarity with it. So the technical understanding and comfort people have has been going up. While the, uh, while the quality of the tools have been going up and the requirements, technical requirements, being able to make good software has been coming down. That's the force I look at when I think about no-code. And this has evolved through, uh, just high-level languages, uh, to more visual control planes, um, like, you know, probably starting with like Visual Studio back in the 1990s, um, and where I think a lot of the WeWebs or Wizz or whatever is the world, they owe a lot to that history of how to build these component-based systems. Then you have this next generation of, wait, all of a sudden we can be describing using language and the computer can start to figure out what we meant to be coding. That becomes a major change. This is Software 3.0 stuff that we were just talking about. It becomes really powerful and it's easier to work with that on a pure play basis. Right. And I think that's one of the reasons why it's so grifty is because anybody can throw up a thin wrapper over like an OpenAI or Anthropic model and create something that can create some code and that kind of works. But the question is like, what comes after that? Right. And like, how do you really create good stuff? And the claim right now is that if you use those tools but you don't know what you're doing, you're not going to make anything that's very good. And that's true as far as it goes. But the big opportunity is like, how are these tools going to combine? Yeah.

It doesn't sound like the AI fake ad from Kevin O'Leary who I saw today. He said, you can now build an application with a prompt. You don't need developers anymore who will take all your money and you have to mortgage your house. One line of text, thousands of lines of code.

Well, I think that's actually a really important point. About the promise of the so-called vibe coding. One of the reasons it's taken off so much is because you can, with a sentence, make something looks like something.

Then there's a lot more work you got to do to make things really easy. It's like those houses at Disneyland. If you look from this specific angle, you have a palace. If you look from the side, you have a shipping container.

That's trouble. When you were working with low-code, you know, it wasn't quite— you couldn't quite get the Pallas thing so fast, right? So the shipping container side and the Pallas side weren't quite as out of sync as you can get them to be with the VIVE Codec. And the VIVE Codec is really good for that, you know? So I think there's another generation to come of people who are figuring out how to be putting together like the best of both worlds and really of three worlds. Right, which is natural language plus the visual plus the, plus code, but being able to describe it the way that we sort of do in Software 1.0, like WeWeb is trying to do. I, WeWeb, you know, I talked with the CEO of WeWeb, um, a few weeks ago, and the, um, and what he said was that the, the market was really responding to the promise, right? This promise of AI plus visual, that was a really strong idea. And I think people have an intuitive sense that's really what, what we, what we need to have is that the, the tooling, that one thing isn't enough. You put, but, but we put together multiple things and it can work very well. The question I have is which is the center and which is the periphery, uh, and the, and, and I mean, eventually it'll always be more cowbell, less cowbell, but the, um, but I think like his theory, right, uh, is that WeWeb being no-code native can add on some AI and that will fulfill the promise. Then we've seen Lovable, and we can make fun of Lovable's nods towards their low-code stuff where they are primarily about the conversational interface and then have a bit of visual augmentation for it. I am looking forward to the next generation of tools that aren't locked into the patterns that either of these tools have had., right? And that are thinking about how we can put these things together natively, right? Like being able to sketch plus, uh, talk about it plus send a piece of paper plus, you know, like the, the, the— what we want to do is lower the friction between the brain and the expression of what you're trying to do.

I always think it'll build the application for us.

Well, yeah, and then I think there, there's a There's a thing after that, which is all these things are about how to build an app. And, you know, the, you know, Apple did this wonderful job of saying there's an app for that, right? That was part of their marketing for like the iPhone.

There is an app for that later on and nobody's using them? The application that has a useless app clip, you can use Despia today.

Um, but I, but I think it's worth asking, like, what— I mean, not only like, what's future software development, which is like the no-code question, but like, what's future software itself? Like, I think to understand the way things are going to get built, you want to ask what things are going to get built, uh, what things are going to get built at all, uh, and the, um, and the, the— I, I think that's really looking to change in the course of the next, you know, 6 months to 6 years. Um, so like, is no-code dead for making current today's apps? Probably not. Um, but what's going to be useful for like what, what we're going to get the most value from next?

Agents.

And, and how, and like how we, how we think about software at all.

Everybody's building agents. It's, you go on YouTube, it's like copy this strategy to build a $50,000 AI agent. How I Made $5 Million with MCPs. I would say, if you made $5 million with MCPs, I mean, if I would make $5 million with MCPs, I wouldn't make a video on how I did it. I would just keep making another $5 million.

But, um, there's a little bit of like, what's the value of secrets? It's absolutely true. Um, but the, the, but actually, I look at agents, I think agents might be thinking about it wrong. Um, because agents are like trying to bridge between, hey, it's AI and it's an app that does things for you more autonomously. I'm actually a little bit less convinced the autonomous part is as important rather than being able to think in terms of delegation versus automation. Like, hey, I want you to go take care of this problem for me. Oh, you took care of that problem. My life is so much better. Right.

And the—

and previously there was so much coding required to go get to do a single task that it wouldn't have good payoff. You want to just get it done, just do it.

Like, you know, package to do something specific in formatting. Versus calling the chat completions API.

Yeah, right. Yeah, yeah. And I think people do that.

That's right. Like AI as a function. Instead of running a piece of Lambda, you can just send it and put it into a chat completion endpoint and have that AI with a very good prompt do what you need to do.

Right, or instead of setting up a complete automation that's gonna take care of, you know, every email going forward, say, hey, go check my inbox and see if there's anything interesting in there, right? That's like the immediate action versus the automation is going to be for the future. And, um, the, the, the, the first of those is what I'm seeing a lot of people do, uh, particularly like MCPs and what have you using ChatGPT. And it sort of changes things up. It's like, instead of, I don't want to make an app, I just want something that does stuff for me. And that's pretty useful for thinking about like what software is really going to be driving value.

Yeah, yeah, definitely. I do think we'll still, as long as humans will be using technology, we'll still have an interface. I think people saying the future is an MCP, if you take a look, the majority, 99.9999999% of the population will not know what an MCP stands for. I think it's Model Context Protocol. I could be wrong, I could be right about that. I would count myself halfway into those people that barely know what it is. I know how it works more or less and what it does, but I couldn't give a TED Talk about it. Most people, it's irrelevant to people what an MCP is. How does having this word MCP and the little JSON for the tool make their life easier? They are looking for an application that does their job, solves their problem, gets them to the goal that they want to go. And I think as long as there are people who are doing, trying to do stuff with software, there will be an interface. I think most of the times we're looking at things that look interesting for developers and they look interesting to other developers, and we are ignoring 99% of the market out there who has the majority of money to spend in our economy, and we're, you know, disregarding what would make it easier for them. I think if you have the ability to bring AI into an application, give the output of an AI, use some generative UI to, like Claude does it sometimes, turn the AI response into an interface with actionable steps. Let's say you have a meal planner application. You say, hey, I'm looking to cook a meal and I'm allergic to this and I can't eat this because I'm on this diet. You figure out what I can do because I think I could only eat air now. Then the AI comes up with some recipes. And you can pick the one that you like. I think taking those ordinary applications, enhancing the way people work with those engines, of reducing the amount of friction that they have in going to a specific goal with the help of AI without them even knowing it's AI, I think that'll be the future moving forward, especially for content generation on content-driven apps. You don't need to pay humans, creators anymore to populate that content. You can automatically on a cron job basis populate those apps with content, which is a big benefit. I think we're looking at how we bring AI into our applications a little bit wrong. We're thinking that everybody of our users is a developer, but in reality, most of them are not and they are fascinated already by ChatGPT. There are so many millions of people who do not know how to use ChatGPT and can't even spell it, if that makes sense. I think we need to take a look at how can we bring AI into an application that is not disruptive to the existing way people would use interfaces but still get the same benefits out of it.

Maybe. But the introduction of GUI interfaces was quite disruptive compared to what came before. The introduction of web interfaces was very disruptive compared to what came before. I think we're going to find disruption happens. When we talk about MTP, it's not so much that there's no interface, it's that the chat is the interface. Right? This is the— where you're sort of handing it over to Gemini, to Google, or to OpenAI, or to Anthropic, or maybe to Elon Musk, uh, and— or maybe we'll, we'll see who else comes out as like a major vendor of this. Um, but the, the promise of like just having it inside there— people used to say the same thing about web pages. Wait, I would only— I would never want to have this thing running inside a browser because I want to control that experience, have my app. And now of course most, you know, Excel experiences are run inside the browser. You know, and I think thinking about software that way as being a component of a bigger thing is a helpful mental model for it, right? It's not that the interface goes away, is that it's being integrated into something else, right? It's being subordinated. And I think that can be wonderful for like increasing the amount of software that's out there because it reduces, it could actually decrease the the visual component, right? And the cost of visual. And I find that a lot of the money that I see go into software over the course of the last decade or so has been heavily on the visual side. And if what you can do is you can be serving up the data and that data can then be used and presented to the user in a more useful way, like the whole generative AI part or whatever, being taken care of more by, you know, the, you know, the, again, the ChatGPTs or Claude or whatever of the world. Um, that's a really interesting opportunity to create more value at substantially less cost. And I think that the cloud, for example, is very smart. They don't call it MCP, they call it integrations. Cursor is very smart because even though they're a development tool, they created support for one-click integration. Oh, you have a subscription over at X? You push this button and boom, now it shows up in your cursor because it just sort of takes care of adding that JavaScript, JSON stuff, and whatever. Like, makes it not your problem. It just says I'm going to make that go away. Cloud's doing that now too with their Cloud Desktop extensions. I think that when we look at what kinds of things people are going to build that will create more value, it's not just how do we build the same kinds of apps we've been building for the last 10 years, but what are the apps that make more sense in this Software 3.0 type domain where we have that interface? An MCP, even as most people won't know what it is, just like they don't know what HTTP is. But people who are developing, if you look at that—

I just know that you can use it in XAML to make an API call and they call it. You don't need to know how it works, you just need to know how to make it work.

Look at that. That's a good point. Hey, we have some comments coming in. One word, maintenance. I think that was in response to the vibe code conversation we were having a little bit ago. But this gentleman is making— or this individual rather, I don't mean to make assumptions— is making some interesting comments about both WeWeb and Xano that I thought play into our conversation.

I'm wondering if you had some thoughts on this.

Interesting.

WeWeb's domain, what they are capable of fails with the AI. It builds things that are incredibly dependent and fail unless you buy more and more tokens. I mean, that's pretty much the idea of an AI company. They want to sell you more tokens.

Yep.

You know, usually those no-code tools companies, they gave you templates, they gave you everything because they—

we were—

keep in mind, we were until half a year ago only used to make money once you were shipping your application. Now they make money as you're building, and they make more money while you're building your application compared to you actually shipping it. So, you know, you have to follow the money a little bit and you will see their motivation. Yeah, also true. Xano, on the other hand, has a limited domain. And does a better job recognizing and advertising that it provides a good start and not an embodied product, which I think is true. I mean, most of the time it's about getting it set up. I think one of the most common questions we have in Office Hours is, how should I structure my database? And I usually say it doesn't matter at all. Like you're going to rebuild this again. You're going to just make all the mistakes right now so you don't have to do them, you know, in a few more iterations. It's easier to do all the mistakes right now. We have another comment from David. Privacy is becoming a more important issue to users and hate the intrusion into their lives. Intrusion. Just turned off to technology in general, is AI even capable of policing itself?

I mean, you could get—

what do you think? You could get an AI to watch another AI.

It's an interesting point too. We have one—

we had one in our community, he was building an, uh, let's call it an OnlyFans alternative. And, and I think he used AI and upload care to make sure that people are not uploading certain content on his website that would be against laws. Yep, I don't get demonetized.

Um, but, but David's— I, I think it might be worth like just, you know, taking David's question at face value here, right? Like he's making a point about AI and privacy, right? That everything runs— AI is sort of sending it off to some other company to go run through their servers to go crunch, crunch, crunch and providing answers. Um, and then, you know, sort of talks— there's a little bit of a second issue he's bringing up around like, you know, AI potentially going rogue or replacing itself. But sticking with the privacy thing for a second, is this something you've been hearing much about as you've been working more in the world of AI-enhanced apps?

Right now, my main thing that I've been seeing is people using AI to moderate their user-driven applications. You have people post something, you use AI to look at it because you don't want people looking at that stuff that they may post. That's what they have been using. And then AI would flag it and send it to a human who will then do the control. So AI for moderation and OpenAI has a moderation API actually. They, they have this prebuilt system that will moderate user content and that can be handy. Now what it does with the moderated content, it's a completely different, different question. You know, you're sending your data for free to those companies and paying them to use it. There has been a lawsuit. Now OpenAI is required, I think, to store all your conversation history even if you delete it, and even if it's in a private chat or in their incognito mode. Everything will be stored and will be used to train the model. It could be that when someone asks a question at some point, that certain things that you told the AI may be appearing on another user about yourself. And the bigger AI gets and the more you ask a question, the more likely it is for that to happen.

Yeah, I hear that. The, um, the, the, certainly that kind of thing happened. The, I, I, the, the, the, I, I, I wonder how much privacy is going to become a thing. I keep on hearing how privacy is important And the people who are true believers in crypto talk about privacy being really important, right? The anonymity of it. People who are cybersecurity, people who are concerned about that stuff talk about privacy being important. And I generally agree that privacy is an important thing, but I got to say the market just doesn't agree. People keep on using Facebook and they throw all their data into all those systems. And they use things like ChatGPT and they don't seem to be that concerned about training on it. I'll hear a little bit of it from the business side, but not usually from individuals. I mean, what I find is that people are willing to throw their privacy considerations to the wind if it means being able to save a dollar or get a cat picture.

Have you seen the big scandal with Meta? AI where when users have been asking it some wild stuff. But what Meta did is they linked all the questions that people ask it to their public Facebook account and username and showed all the chats with it on a feed that their friends could see. People have been asking about it to generate some very interesting pictures, to have it generate, to have it help them. I think one asked like, how do I hide $100,000? And I was like, oh my God. And they just did it overnight, saying, hey, you have like 3 days to respond to this email, otherwise we will put up all your AI chats publicly or something like this. And I was like, oh my God, I hope this doesn't happen to everybody because, you know, like, I would see some wild stuff on the internet. Goodness. And they've been doing that because, I mean, Meta is, I think, a little bit failing at AI besides people using their API and open source hosting their models. But when it comes to consumer-based AI, I think they're falling behind. That's why they're like, I think, cloning MrBeast and putting him on your WhatsApp or some shit like that.

Yeah, they're integrating AI into their products and doing so relatively well, but what they're not doing is— they're— I think they feel like they're missing out, right, on what's next for software by not having their own play that's on par with ChatGPT or Claude or Gemini.

And then like, um, xAI has been doing like this wild anime AI thing which I saw some videos about and it's been crazy. It's completely unmoderated.

Unmoderated?

Unmoderated. You can ask it stuff, but it will tell you stuff that no other AI would tell you.

Okey-dokey. The unmoderated aspect is more useful when you have a much smaller database to work with. Um, that's actually where like Llama works really well with like, um, local, local AI, right? And creating fine-tuned versions of it, uh, because that way it's just sort of focused on whatever that data is. But the, um, but yeah, the, the cloud-based models, ones that have lots and lots of data, the opportunity for them to get corrupt is really high. We— what was the name of the Microsoft one, like Tate or something, Tata, um, the, um, which was like, it was basically a chatbot that was trained to act like a Nazi, and then that, that did not go well.

I can assume that isn't— that, that, that does not go well. I mean, Grok was also pretty wild, I think, a few days ago, so, uh, the Mecha Hitler thing, yes, that was not—

that was not okay.

Oh my God, I don't know what's happening to AI. I think it's just looking how stupid humans are and it just copies them. And I, I think we're not going to have this end. I don't think this world will be ending by AI advancing in intelligence. I think it will be ending by AI trying to copy our intelligence.

Oh, and that's what's going to protect us because we're not actually all that bright? I think that's a, that's a reasonable point.

I think if AI will be as dumb as we are, I think we're screwed.

Yeah, I hope. Yeah, and, you know, David actually put up a post just now, one talking about, you know, the absence of guardrails, where that creates legal culpability, right? And the And that's, I think that's a really reasonable, it's a reasonable question for which I'm completely unqualified to answer, right? Not a lawyer. But the, it certainly, it can be concerning when it's used poorly, but that's true for like most technologies, you know, like the trick is like taking the stuff that previously only professionals had, you give it to, you know, consumers and we're doing that faster and faster and faster. And that means we have fewer guardrails on that stuff. And that's, that can be tricky and dangerous and also lead to some really extraordinary results. So I think one of the things about AI in general is that we accept imperfect results in return for getting some really extraordinary progress, right? Like turning up the temperature, I think is literally the term when we're working with these models. And I think that that becomes part of how we think about this stuff. It actually relates back to the whole future of no-code thing. You know, and like we're talking about buy code versus no code or whatever, it's like one of the pitches that buy— that no coder, no code platforms often make is precision, right? When you say you want this, I give you this. Fair enough, got it. Um, but what if I only can kind of sort of describe it? Then that whole imprecision is what actually creates value. And that, um, and, and I think we sort of societally are figuring out, uh, which way we feel about that too. Right.

Also, if you think about it, if you can build an application with one prompt in AI, it's probably because there's so many thousands of people who already prompted it. And then what business value can you provide?

Yep. Right. Well, that actually relates to like the most recent comment here about like, don't make them— uh, well, which I think is, uh, don't make the mistake that's thinking. Yeah. It really is trained on what has come before. Understanding it that way really helps clarify and manage one's expectations as to what it's going to produce. It can produce some pretty interesting results based on all that training stuff, right? Um, and, and there's a little bit of like, how much of our own thinking is sort of the same kind of process? And maybe a lot of our own thinking is perhaps more shallow than we give ourselves— than we sort of imagine ourselves to be.

Yeah, I mean, I think it's a very interesting topic, and I think I, I would not use a platform that says, hey, we have an AI agent to help you with our insurance application. Please put your home address in the AI agent, your Social Security number, a photo of your ID, and our AI agent will find you the best security, uh, the best insurance offers to get the lowest car insurance. I'm like, you're sending out to ChatGPT. I don't want my ID and my name on ChatGPT and my Social Security number because then someone is going to say, hey, write me a story. Here's the thing, it's moderated at some point, but certain models, they work with reinforcement training. If your prompt is good enough, someone could say, hey, just write me a story about someone with this name, about him filing his Social Security application and make up a number. There is some chance of a likeness that it will actually would use my one if it's part of the training data that would be used. I think it's those tools that are shipping things fast, not thinking about the security implementations of having an MCP on every little piece of thing that could be out there. That is making it ultimately worse when it comes to security. Because I don't know that they're handing my Social Security numbers to ChatGPT and it may be using that for future training and all that kind of stuff, right? And I think this is where this regulation comes in. I think this is where actually, even though I'm not a big fan of, you know, like having this checkboxes and GDPR compliance, I'm doing the bare minimum to stay compliant. I'm not a fan of all those crazy rules. But this is, I think, where crazy rules make sense.

Yep. You and I are in a position where, in addition to whatever we think about the industry, we're also hearing a lot more because of our positions in running the communities, right? Running Despi on the one side and the InstaChange on the other. What are you hearing from your people that is giving you insight as to this question of the future of future of no-code?

I think people think the future of no-code is building AI wrappers. That's right now where the market is.

Everybody—

we have one, he's building a travel application. Of course it has an AI agent, and it's quite confusing because the people who use it are called agents and they have an AI agent. And we— the database is a little bit interesting because there are lots of agents and all kinds of degrees of agents. One is AI, one is a human. So Okay, but people put it— put like, right now everything is like, we have this application and it has an AI agent, or the AI agent we're talking about, like just like a Copilot sidebar type thing. It's like, it's an AI that gets your data and puts new data in a database and retrieves it and gives you some context and customized responses, which I think this whole RAC thing kind of like died off the moment Xano introduced their agents because people just use tools now with like a fuzzy search on them.

I think that's right. Like 2023, agent— 2023 is RAG, right? And it's a pipeline, right? 2024, we're seeing more agentic work, uh, full use becomes first class. Uh, 2025, we're seeing, uh, tool use getting, I think, largely substituted by MCP, right? To allow basically a composition of tools that aren't being decided by the vendor but being decided by the customer. I think that's a shift in power that goes along with that. And so the mix of tools I'm using is different than the mix of tools you're using. And then what comes after that becomes the interesting question. But what you're finding is that most people are building AI wrappers and building AI wrappers with no-code. Is that the pattern you're seeing? Yes.

And I think people would be making so much money if they would just focus on building a real good application, solving a real problem. But I think there is a difference between a market that really makes money and having a product that's hyped at the moment. And I think most people put their attention on using those no-code tools to build a product that's very hype and makes money on a short term and dives off because AI will have a plateau at some point. Everything that goes up goes down and goes up again, goes down. It's like the stock market. Or crypto. Coming back to new keywords so we get better indexed by Google. No, but everything that goes up goes down. It's like my views on YouTube as well. And, um, you know, there will be a time where people will not want AI if certain things come out of, you know, if, if people would know that every AI CLI that they use is essentially stealing their idea and giving it to all their competitors. Imagine how much more people would consider no-code, right? I think that would be actually something for Nordkraft to consider advertising, but they're working on their own AI, which probably is not something they want to advertise now. But if they would advertise, we don't steal our users' business ideas, AI-free development, privacy-driven, keep your IP, don't get your IP exposed before you get a patent lawyer, something like this. I think that's the way to advertise to builders actually, because people are so afraid of someone having their idea before they have it out there. And if you're afraid of that, AI is probably not the right way to build it, because every prompt, every time you, you, you scream at the AI, it's being put back into the training set.

Right. Yeah, you make a really good point about the feedback loop that's built with that, um, as well. So the, um, but what do you think this means for people who are like in your group? Like, uh, if right now they're building these AI wrappers and like, hey, what goes up must come down, right? Or sort of the thing about that, like, what's the What's the term that comes next, sort of, in your view, based on what you've seen?

I think what comes next is— I think there will be a time where— right now there is an opportunity of white labeling AI, and there is always— like, 5 years ago, people have been building their SaaS on GoHighLevel and all that kind of stuff. There's always this trend right now is for enterprise tools, agent, Copilot. For, um, for silly apps that actually— I found that those silly apps actually work pretty well, like make a photo of meal and get like an inaccurate measurement of the calorie content, or things that you can advertise on TikTok with AI-generated UGC creators, right? They actually do very well. And, um, you know, you can just grab those models. Like, let's say you have the, uh, V03, which is $6 a minute. There are certain places— because here's the thing, if you have a Google account, you can get unlimited generations for free. So right now there are companies that have like bots create thousands of Google accounts and give you VO3 for $1 a minute. And if you can make an app and then sell it for $3 a minute, which is half of the price of what the others would be charging, and you can put some of them together and build something like Sylvan is doing and sell that to direct-to-consumer and sell it for like $19 a week. And you know, you know, they'll only stay 2 weeks subscribe to it, but you made $40 of them, you get a good return on investment. You can fuel this by App Store advertisements and cheap TikTok ads. And, you know, your audience is not necessarily— like, you sell essentially to Gen Z, which is a good market because they buy pretty much anything. Um, that is, I think, the next thing that may just pop up and people start building. Because I think all the people right now that make videos about how to build agents, I think none of them is making money by actually selling AI agents. They're making money by selling new courses, how to build AI agents.

There's a lot of that world of low-code, no-code as well.

I think if you take a look at most of the people, All those 16-year-old millionaires who build applications and all that, which most of those stories are fake, but the ones that there were some truth to the story, most of them built direct-to-consumer mobile applications for silly AI wrappers. I think making an AI wrapper as a mobile application makes more sense than making an AI wrapper as a web application because you get discovery through the App Store. Don't expect this to be a long-term business. This is a grift., but you can ride the wave if you can develop fast. And here's the thing, if you actually develop with a low-code, no-code tool and you actually have a quality application that takes the payment with RevenueCat and actually gives the user access and not takes the money with RevenueCat and then the webhook fails on your Supabase, which was configured by AI and the app dies off and you get a 1-star rating on the App Store, you actually have a quality application. You will shine compared to all your other competitors who have I've coded without experienced leadership in building it. I think we'll be going back to templates to move faster. Having the ability to use AI, I think the future will be what we've been doing since 2 years in office hours together, writing a little bit of custom code with the help of AI, in a low-code environment and using some templates and some UI elements to make, make it move faster. I think it's how can we get this critical code that really makes the difference written fast? And if I don't know how to write code, how can I have AI help me with that and condense it into a platform that is visual so I can move quick, but I can also work with precision? Because starting a business is easy, not getting chargebacks is hard. You actually need to provide a quality product. And chargebacks ruin people, you know, that's $50 every time you get a chargeback. So if you get enough on a Vibe-coded application, you're very easy in the negative.

I think that's very right, that the, um, particularly when you're doing a lot of grift, you have a lot of people who don't actually ship. And because they're not shipping, the— where's credibility, the whole thing coming from?

If you take a look, vibe coding, I think the way it's being advertised right now is a new dropshipping. You have a low quality product that you're selling for top dollar.

Yep. And you're— and it's not— yeah, and just like dropping, I was gonna say, is it's not intrinsic to the fact that you're dropshipping. It's just that, you know, when you lie down with dogs, you get up with fleas.

It's how you do it, because dropshipping essentially, it's a form of fulfillment, but it had to be as a business model. Vibe coding as a form of fulfillment to deliver your application is completely fine. Vibe coding as a business model, saying I'm vibing, I'm not thinking, I'm doing it, I built my application on the store, I'm making money, you know, what comes up goes down. And you know, if you're making lots of money with a bad application, you're going to get lots of refunds, lots of bad reviews, and it can put you more down than it gave you, what made you go up. Okay. And the App Store is very generous now in doing refunds for any 5-coded applications. If— here's the thing, App Store usually doesn't do, you know, refunds easily, but if you look at their policy, they give refunds if the app doesn't work. So if the user just makes a screenshot saying, hey, I click on this button, it's not generating anything, they can get a full refund for even previous month that it used to work. Yep. Um, once said, amen, what's old is new again.

I like that. There's nothing new under the sun. Let's skip to his most recent point though, because he wrote that while you were doing your most recent riff, and I thought that was an interesting convers—

different conversation Long-term, this will lead to language coding consolidation and essentially homogenization. Boredom will then lead to spikes of creativity until AI actually starts thinking. I mean, if it starts thinking, it just means that it resembles thinking. It can't think. It's not a human. It can't. It can think it's thinking. It's hard to put it in words. It can put some thinking tokens in there and reason and look in the internet and do research. It's like me looking in the internet about a coding question and then coming up with a solution that makes sense, that seems to work, and then I show it to Ray and he says, completely wrong.

Yeah, I thought that the, his point about consolidation, homogenization, uh, was an interesting one. Um, I, that is not the direction I'm seeing things go in the short to medium term. I'm seeing a whole lot more diversity, um, and I'm seeing it because the, the same kind of core apps people have been building that becomes a much redder ocean. We're like, hey, lots of people can make a replacement level web app or whatever. Those are going to be lots and lots and lots of web apps get made in the course of the next couple of years. But where I think the money is to be made is the fact that tools are making it easier to go build the next level up of complexity. You were alluding to this on the mobile app side., right? And then I look at mobile apps, I look at things that are designed to integrate directly with AI, I look at extensions, right? And this, I think, the, the fact that you get that diversity, uh, it becomes the, the challenge for at least today's generation, right, of, um, of low-code apps. Because today's generation of low-code apps are really focused on helping you build, uh, a couple of very specific kinds of tools, right? Like WeWeb, Wiz, you know, whatever, Bubble. They're focused on helping you make a, um, a web app, right? An app that you can deploy on the web. And yes, Bubble has like a new thing for making native apps. I haven't really investigated that much. Um, the— but if the real money is going to be in making things that run inside of, you know, these, uh, chat environments or making a mobile app like the kind of thing you're doing with Despia, the, uh, that tooling, right, that allows you to be able to go into those, you know, higher value environments is, is where there's more money to be made. And like, as you discovered with Despia, there, it's a surprisingly empty field, right? Like, Despia is good software. I think it's great. I love the way you've done it. It's great. But the, um, but the, the number of competitors for being able to make that work and have that kind of simple path is relatively small. I think on the no-code side we have like FlutterFlow, maybe Draftbit. Um, but otherwise it's mostly just web apps and you're sort of leveraging the fact that you can make a web app and then get over into the App Store because that's a bluer ocean. I think there's no competitor to Despair right now. Yeah, not a non-direct competitor, not one-to-one.

Um, I think turns your shitty web app into an even shittier mobile app. They'll get rejected by Apple. But yeah. And there's Median who charges you $40,000 for essentially the same thing, you know, and that's about it.

Yeah, and I, I, I, I think that, that, that increasing diversity of the potential software that creates value means there's lot— makes lots more, uh, niches, right, where people can be building businesses that, that can have some real value to them. But the temptation is always going to be, well, let's let's point ourselves towards the biggest market. And the biggest market is web apps, which it probably will be for the next 5 years at least. Um, but it's going to be where the, where the puck is today, but not where it's going. And I am far more excited about being where, you know, where is the industry going? And like, hey, where's, where's no-code or whatever going to be in 5 years? And let's go there, right? So we can be ready to catch it. Um, or catch it as it goes by, as it continues to—

like, what's going to be after that? I think it'll be people. I think you will see more code editors in no-code tools and an AI assistant to help you write that code, the little pieces of code that matter in your application. Think of like Wist. You have lots of code widgets that will extend the functionality, but you'll have an agent helping you write that code and helping you understand and learn it. As you're vibe coding, you're learning it because you see the code intercept with visual elements in that builder. I think actually someone is considering learning JavaScript, I would suggest to start with Wisp. It's a great way to learn JavaScript. I think if you get an AI assistant inside of there that can teach and walk you step-by-step through something that works and you can see what changes get what results, I think that's maybe where the future is going. It won't be an IDE, it'll be an editor with many IDEs inside of it. Remember I told you if I could just have something that's like React in an IDE and I can click on buttons and then it says pick your event, click, double-click, and I get this field where I can write code and set variables that are visually on the side, kind of like looking like WYSIWYG, I would be happy.

Happy.

Yeah.

And I was like, we need an IDE like that. I mean, you use Framer, right? I mean, does Framer have some of that going on?

No, I don't use Framer anymore. I've built— actually, I don't have landing pages in Norcraft. Okay. I built an AI building interfaces for me in Norcraft so I can just drive it, and then I got full control over it in Norcraft. So yeah, solves that for me. I just have the AI build the landing page and I put the soul into it.

Okay. Yeah, there's a lot of value that is sitting in that and sitting in these different kinds of apps that can create, that can deliver where people actually do work. But I think that actually, that probably is the key idea I want to land on with that, which is software needs to go where people are going, right? Rather than expecting that people are going to come to the software. Like, that's the, the 'I'm making a web app, come to me, you will buy my stuff and use my app the way I decided to.' Like, no, I don't want to. That's, that's one of the reasons why MCP— sort of coming back around to that— is that it's so interesting, is that I can use your service but not have to deal with like the interfaces you built for your service. It's kind of like being a high-end customer and not really having to deal with the usual channels. You can just Go say, I want to speak with the manager, and you get that experience. I think that's the apps that can be delivering on that will create a lot more value for themselves.

I think there is a business opportunity for MCPs and for tool usage. I think it will be combining that with a little bit more interface. I think getting a little bit more interface into the chat where I can have the chat generate an interface based on the JSON config file from an MCP saying, hey, What do you want your post to be? You can either tell me those and I'll fill them out and you see the progress of the interface, or you can just type it in yourself if you don't want to write 50 messages back and forth with me. Because one thing AI is really good at is, perfect, thank you for saying this prompt.

Are you sure you want me to do this?

That's a great idea. You sure? We had that in office hours and we put the prompt in it, no chitchat in caps with 3 exclamation points.

Okay, no chitchat. I think that's a— yeah, that chitchat thing also sort of makes me think of like what I was doing earlier today with Warp. Right? And how to get more, um, how to get more value out of just having the computer do a thing for me. Like, I was having it upload a bunch of PDF books. I want to convert them into Markdown, get them uploaded to an endpoint I had set up to receive them. And then, uh, I said, go take care of that. And it figured out how to take care of that, right? That, that kind of, I want to get something done, creates value. And it's like, it's very different, uh, from what, uh, what we think of when we think of like building an app. It created a script, and that script will be reusable. But that was a, that was, that was a side effect, right, of the actual, uh, job to be done, which was just, I got a bunch of stuff, go take care of it for me.

Yeah, I think there are many layers to the future of no-code. It's how will we deal with privacy, GDPR compliant is Xano still if it sends all your user data to OpenAI?

If that's what it's using for its model, yeah. Of course, there are different now multiple OpenAI servers.

This is complicated, but it's turning the wheel on it. How do we manage compliance? How do we set that up? How do we get the data processing agreements which you need for certain things? How do we get that? Then if there's a middle layer who's doing that, who's making sure it goes to the right pull zone and all those things. There will be more Governance, I think. More, you know, more— oh, this is a good one.

Chit-chat increases token usage.

It sure does. But did you see that ChatGPT is, I think, spending $1 million a day or so on the AI saying, you're welcome, or hello? Manners have a price. No, but I think it will be compliance. It will be governance. And then also it's, are we overdoing it? Are we putting AI in stuff that doesn't need AI? Because now everything has AI. I was buying car insurance, you know, like you have AI everywhere.

Like, right, we have things that didn't have AI and now AI is getting slapped onto stuff. And then there's going to be the next generation. I mean, we've already seen some of these tools, right? But, but, but I think there's a generation of tools coming out that are going to be just AI as a starting point. Right, they're gonna be native to it, and that's where like really exciting values— we talked a little bit about Dia last time, which probably got you that ticket. Um, we got, uh, we talked about— I just talked about Warp, which falls in that category as being an AI-centric, um, you know, terminal program. Uh, Cursor, we've talked about many times before, um, and Windsurf, which we're assuming it's still around next week. Um, what is— those are sort of examples of like AI-centric development tools. In contrast, say to the way Visual Studio Code slapped Copilot on the side, increasingly becoming more central, whatever, but that idea of what are you native to, I think really matters. You know the Hello Fellow Kids meme? No. Okay. Steve Buscemi, who is an actor who's looked old his whole life. Near as I can tell. He's dressed like a backward baseball cap and a t-shirt says Rock Band. He's carrying a skateboard and he says, you know, how's it going, fellow kids? And he's sort of entering high school and like everyone's looking at like, clearly he's not, you know, a fellow teenager. And I think the AI tools run the risk of being that guy, right? Of like trying to like jam on some of that AI youth to their craggy face. And there's also an opportunity to become AI native because it's a new space. Yeah, agree. AI as a starting point is a winner. Yeah.

And then I think that brings it back to our— we all got to build live coding tools now and get some AI-generated advertisements from Mr.

Wonderful. That, that too. But, but also I think AI as a starting point becomes a pretty useful way of thinking about like what is the future of no-code. Because if all software is AI-centric for the next 60 months, which is 5 years, um, then I think it's reasonable to say that, okay, that means future of no-code is AI-centric. And what that looks like isn't totally obvious to me.

I think, but the future of no-code is something we'll see once we're there because it's, it's no-code has been always been Wild West.

Yeah, it's been Wild West.

It's been new.

And yet we may call it, we may call it no, no vibes or I don't know, maybe, but it's also worth, I think it's worth sort of keeping in the back of our heads. What the most popular no-code tool in history is and like the, uh, and, and so like why and, and, and maybe why, you know, cause it's not Mendix, it's not OutSystems.

What would you say it is? I want, I wanted to say OutSystems cause they've been around forever.

They've been around about 20 years. Yeah. Like Mendix, right? They, they, they're sort of that same vintage. I call them the same class of, of no-code. But I, I would, I would posit that, that neither of them really are, are that. I think it's Microsoft Excel, because Microsoft Excel is a tool through which you can produce all sorts of stuff that's immediately useful to your business without, uh, without having to be formally trained in code. And the fact that it uses text-based formulas doesn't really stop it, right? It has this big field. And in fact, when you look at like some of the more popular of the no-code tools like, uh, you know, Glide or Airtable, uh, they are kind of— they start with Excel because they understand Excel is already where people are building things that create business value, um, today. And, and, and I bring it up just to say, because it makes us think a little different about like what constitutes a build tool, right? What constitutes software building? Um, because like when people make an Excel spreadsheet or Excel workbook that has like all the flows they've made for their business Plenty of, you know, people who are like maybe formally trained software developers look at that with horror, but there are millions of dollars running over these and lots of businesses. It just needs to work. Yeah, first make it pretty, then, um, well, actually, just first make it work, then make it fast, then make it pretty.

Someone said HyperCard. I don't know what that is, but I probably, I probably think you know.

Oh, that was from back in the day. That's when Apple, Apple thought everything was going to run on HyperCard. And the guy who invented HyperCard went to, um, he had a second run at General Magic, uh, which is itself a famous story about hardware being a little before its time.

Like the old version of FileMaker or something like that.

Yeah, yeah, basically they thought they'd have a deck of cards be the way that people store things in databases, we had interfaces be based on the cards. So it was lots of really interesting ideas about that. And, and I think that's perfectly fair point to say that would be well considered like a low-code tool, and that these have been sort of coming in and out throughout all of, you know, all the history of computer science, going all the way back to 1967 and Robot Draftsman by Lightman.

But also, it's— I think the first no-code tool would be the modern-day computer interface where you don't have to write in the terminal anymore or do a punch card.

We can keep on going back, right? Like things keep on getting better. And like the— and I think by freeing our minds just a little bit, we can be seeing the future a little more clearly.

And then we can also be participating in future. Do you have visual building slash development before you had no-code? Because if you think like, if you, you know, write, write something on a whiteboard, that is, I think, somewhat considered visual development, or you put like some cards on the table together and put them together. So it doesn't have to be on a computer, it doesn't have to be code to be a visual, you know, interface.

All that is true. Um, and I think the other thing that gets really— gets me excited about like the world of, um, you know, AI in terms of being able to facilitate is being able to take that whiteboard, take a picture of it, And have that be the foundation of like software that's getting written on your behalf.

I like that. Yes, that I think that is great. I would not want to give the control up. I want to have— I wanted to organize the project. I wanted to give me some ideas how to start. And then I want to put it together so I know where everything is. It's like if you have someone else clean your room, you don't know where they all put everything. And you got to be searching for it for for quite some while.

I think AI as a teacher is really good. Yeah.

I think that is— I think we should focus less about doing it for you and more about teaching you how to do it and being there as a guide.

Think about like Ray, as an AI in WeWeb. And everybody's saying, help, I'm trapped inside WeWeb. The, um, yeah, I, I, I, I think that's, that's right. We, we, in, in StateChange, we talk about two models for, for AI. We talk about the lackey and the teacher. And that, um, I think we, we use it as a lackey a lot, um, but if we, if we alternate using as lackey and as teacher, we can actually make a lot more forward progress. I agree.

And speaking of forward progress, let's do this again. Yes, I can't wait for our next topic.

Wonderful.

Thank you for your time, and thank you all for watching. I think this was like— we never had that many views as we have right now that are actually watching live. I think this is a very hot topic. AI option.

Our favorite developer advocate at Xano. Yeah, we, we, we, yeah, uh, Daniel, Daniel, we need to get him on the, get him on the vidcast.

Let's do a podcast. And he says he's busy and unbelievable.

All right, well, let's see if we can, let's see if we can lock him down.

Maybe, maybe we'll use AI. If you email him, he's going to respond and he's going to be on here. If I do it, it will take another 2 weeks. It's just like the pull of getting an email from Ray.

All right, well, I will, I will, I will do what we can. I will look forward to seeing you next week. And Daniel, you're going to be on deck to be showing up with us sometime pretty soon. Wonderful.

Thank you all for watching, and we'll do this again next week.