Beyond Vibe Coding - Open Office Hours - With Ray Deck (LIVE) - Episode 7

Hello, Ray Deck. Hello, Yasin.

How's it going? It's another week. It's another podcast and it's another exciting topic. Today we're going to go beyond live coding. Now you made that term up, so I'm going to let you start on that.

Yeah. I got the idea for that title from actually our first session together back before we were starting to livestream these things. Where we were talking about vibe coding as an idea and the way that it often gets conveyed as you just don't have to think. The hey, you're going to just vibe and let the computer do it. It's all about feelings, man. I think good software has always been about feelings. Do more. Yeah. It's more than that. There's a lot you can do to start. You can give it a little bit and it can run. But then there's like the, there's a, I mean, we talk about the hardest 5% over at StateChange, right? And like how do you actually finish this? How do you create something that's going to really create business value? And I think that the way we build with, use AI to build with software changes the way that we build, but it's not a replacement for our thinking, right? And the, and that's why I talk about like the whole beyond vibe coding thing. Yeah, sure. It lets us go faster, but how can you get speed and control, right? You're going to, you're putting on the rocket to go behind the car, but you're leaving steering wheel on so you can be figuring out, so you can be in control of where you're going to take it and how it's going to create real business value for you. I think when you combine those things, it can be really extraordinary. Like people who are cranky, you shouldn't use AI. That's not the right answer either, right? But like, but that sort of requires all to look at this and say, how does it help us? That means going, beyond vibe coding?

Yeah, I think when vibe coding was established, that was when the ex-no-coders found out that AI can write their code and they were going very wild about it. And most of the people who never built an application their whole entire life then were able to build, you know, all their bad ideas using AI. But most of the times what I've found is you need to build it yourself to be able to understand what it truly means to have a software company. To run a business. Just giving someone who has an idea, an AI or a developer back then on Fiverr, that's how we used to do things, who's able to build your idea does not mean that's going to be successful, or that's going to be a good idea, nor that it's going to be worth the time that you're going to put into it. I think most of the times when people, you know, back then before VibeCoding, we had entrepreneurs outsource their ideas. It's just that now we have a cheaper way with more errors in them, or probably as much, depends on how high quality the people are that you want to work with. It's just that bad execution will always have a bad product. Now you can just build it faster with live coding and way less in terms of costs. But the moment you actually start building products, that is when you give up on that, have other people do it, have AI do it. And it's actually when you try to learn some of the principles yourself. For example, for myself, I was first hiring people to build stuff for me, and I didn't know how to program. I didn't know how to do this. And I was going on Fiverr, I was finding people, and I had an idea, and then they were building the idea for me. I was doing the design and they were doing the development. And it was never quite the way I liked it, and it was never truly successful that anything that came out of it, because it wasn't my idea. It was my— I had to start, and then the developer continued on with the idea to realize it. The moment I sat down and started learning WYSIWYG, the moment I sat down and started learning Webflow and JavaScript. That's how I started out. Maybe I started out a little bit too high, directly to JavaScript. I wouldn't suggest that anymore, but I had some courage and I started out there. The moment I was able to build this stuff myself, my ideas were way better. I knew what is possible and I knew how it is possible. At the moment I had an idea, I was like, no, I can't do that yet, but I can do this. And that is a big thing. It's having this understanding, having this awareness of what's possible, what's not possible, and how to approach it. Of course, there are going to be roadblocks down the road. And that's when, you know, I joined StateChange.ai, which has completely transformed my life. Sign up today for as little as $39.99 or $39. You even cut out the 99. I forgot about that. Look at that. Great discount. Sign up today, send Ray your money. It's definitely worth it. I did $149 immediately because it was such a great value proposition. But you know, that's when you join those communities, that's when you get this, this extra help. But just being able to do it yourself and just turn movements on the mouse into working software, that is the moment when you actually get really good in having those ideas and executing your ideas. And this is actually when you see something that was like a miracle for me seeing the first time. The idea I had, one-to-one translated into a product. It wasn't like, oh, I just had this idea, today's the delivery on Fiverr, let's look at the project. You're already excited, you know, you get the Dropbox link, you're opening it up, and then your face goes from this to this, and you're like Oh gosh, I paid $2,000 for this.

Yep.

I, I think about that kind of like the same thing with Lovable plenty of times.

Oh yeah. See how that goes. And then it's like, oh, this, there, there's this, I think, repeating pattern in here. Uh, like it also applies to marketing, right? People say, oh, what I need is marketing. I'm gonna go outsource it to like a marketing pro who's going to go make marketing happen for me, you know? And like, I think in both situations what you really profit from is that software is like, I once had a really smart guy, a venture capitalist told me that the software solidified consulting, which I thought was a really neat idea because it's just about like your ideas, right? It's about the product you made. And so like the best software is when you have a pure translation between the ideas in your head and being able to convey it, right? You can convey it as content, you can convey it as consulting, you can convey it as code, right? These are 3 major ways you can be manifesting, right? Or memorializing your knowledge. And all of them, they reward more you, right? So like the idea like, I'm gonna have AI write my marketing copy for me is losing it. Having AI enhance or multiply your marketing copy, that can be really good. But what people want to hear from is more you. And like whenever I'm giving marketing advice, it really boils down to how can we have more you you in the market. And when it's about creating software, like, which is sort of the point you're making, right, is how to have more you in the software. If you delegate, if you're like, oh, I have an idea and someone else will go make it happen, that's a very little amount of you.

The thing I have observed, plenty of people building stuff— clients, friends, colleagues— and I can tell you that nobody was truly successful outsourcing the idea. They were never truly successful when they outsourced the idea. Unless you're in a room together and you have a big whiteboard and you can say, I want this to your developers. If you outsource it remotely, they were never happy with what came out of it. And they were just trying the best to make the best out of what they got.

And, you know, I, I think that part of that is that delegation is a skill.

Yeah.

And people don't know how to do it. And I think that, that because you're giving the job to someone else, the whole thing's gonna become easier. But really what you've done is you've made the whole thing contingent on not your idea.

Speaking against the wall.

Exactly. And of course it's really expensive too. Uh, and so like you're running out of money, you're doing it in the way that you have the least competence to do it because you don't have that kind of managerial expertise outside outsourcing the tasks that users can see.

And they don't need to, like, you know, like I do the UI, I do the backend, I do the server, I have other people at that iOS plugin that we need that is not something that directly affects the user. And if we have a mistake on that, and we don't just don't roll it out until it works, right? So those are things you can easily delegate to other people. And they're probably way— they are just probably way smarter than I am at that anyway. But just because you have the smartest in the room doesn't mean that it's going to be as you want it to be. Like, you want the core structure that will be the front end, the back end, the basic parts of that. You can do it nowadays. Like, it's not that complicated. Just throw away Supabase, start using Xano, and then use lovable in combination with that with the OpenAPI spec. So, you know, you're not exposing all your users' data and API keys because you fail to configure row-level security. Or, you know, use Nordcraft or, you know, use something else that's better. It doesn't matter. And I've been on a hot take recently. I have been saying it doesn't matter if you use Nordcraft for your frontend or Lovable or WeWeb or Wist I know I'm going against what I used to say, but you learn and it doesn't really matter where your frontend is on. If you're building a SaaS, you don't care about SEO most of the times. You could build in Wist if you like Webflow. Of course, if you want to pay $200 per million requests for just loading a CDN file, you can host it on Wist if you want to. If you don't want to, and you need server-side rendering, you can go to, you know, WeWeb Collections, and then you can— or you could use Prerender, or you can use Nordcraft, or you can use Next.js with Level, right? It doesn't truly matter where you are building your application, your frontend on. I learned that. I learned that many times. Of course, there's always a personal preference. My personal preference is anything that is close remotely to Figma that allows me to build simple frontend logic or even just write inline JavaScript to visualize design creation. I'm big on design. I do like that, and I'm not the best designer. I'm far from that, but I do want to be able to control the UI that I'm building and not just have AI make something that looks like every website nowadays.

Right, right. Except of course, you know, that I find that the look of the site is often less important than the content of it, but that depends.

I'm the only one bothering about UI.

Yeah, well, other people can't from the front end, the brand of it, the feel of it is very important. And that's all fine. You make an excellent point about like there are lots of tools out there. And one thing we talk about in StateChange, we have a bunch of mental models we use. And one of those is something we call the shopper's mindset. And the idea is that like, it's what are, you know, what is the right tool based on what's world-class in terms of tool setup at the time, which changes all the time as you well know. What is your understanding of the job to be done?, right? And where are you in your technical journey? Like what's gonna fit best with your work, right? For the thing you're trying to do based on where the tool is at that moment. And all those things keep on changing. And, but, but it also means it's not like there's one tool that's sort of the right way to do it and that endures. It keeps, keeps changing. And that's something I think a lot of people find scary about the world of software is that it's not that solid, it's not that stable. But the opportunity is that we can create software, meaningful software, so much faster and with such lower cost. That's not like the artifact we create needs to last for 10 years in order for us to create payoff.

It's not like that. I had a great conversation in our office hours today. I think someone asked Andreas how much, you know, how long Tartle can support on the money that they raised. And I think they told me that Andreas said they can run for the next 3 years on the money that they have raised. So it means they're burning like a million a year, $180,000 a month about. So, and then they ask, no, should we worry what happens in 3 years? Will Tuttle go or Nordcraft go bankrupt? And Tuttle is already far gone, right? And like, is Nordcraft going to go bankrupt or something? In 3 years it doesn't matter. Like, don't worry about something in 3 years, we'll worry about it again in 2 and a half.

I, I think that's totally right because the, uh, another mental model from StateChange, which is sort of a running gag inside there, is you will build this again. Yes. And, and because of that, like, it's just not that big of a concern.

That's right.

Times before it even hit the market, so many bigger sources of risk, uh, that really what you want to do is to sort of shift your, um, you know, shift the money-making part.

I made a little joke. I said, you know, they're open source. When they go bankrupt, I'll just make my version with a better name.

Well, they're open source, don't need to wait for them to go away, right?

Yeah, the arguments— they want me to pay a license in order to do that.

So, you know, okay, which also, I guess, and they're probably open core, not complete open source. The— and that, which by the way feels totally fair to me. I think a lot of people who like distribute this stuff in open source just sort of get reamed over it. Docker is good.

It's just I don't want to do it unless it's completely necessary because I don't like— I don't want— I don't like the idea of running a frontend web development tool. That sounds like a lot of sleepless nights with hosting.

There are some folks in SageChange who use open source, you know, front-end web development tools and they sort of tweak them a little bit to their own tastes. And which I think is sort of neat because then you could take that, you combine that with use of like these AI-assisted tools, right? And now you make the tool itself build process better, etc.

Which one? Builder? IO, not AI, they're bankrupt.

Oh goodness gracious, Builder AI was a fraud.

It was big. I remember when I was doing research for competitors for Despi, I found Builder AI and I was like, how is an AI no-code tool $30,000 to make a mobile app? It doesn't add up. Now we know there were Indians making them. That's why it was $30,000.

Yeah, sort of the— I mean, there are other sort of AI-assisted agency models. It appeared they were just not as AI-assisted as we maybe thought that they were, or potentially Builder AI, right? Uh, and then the, the, there's Builder.io, uh, which—

please don't send us a cease and desist. This is a great tool. We love Tempolabs. We used to love Builder AI. It's just sad that it went that way. Um, the, um Those are only our opinions expressed. We do not check validity or something.

That is a really good point. Let's not be sued for libel. I'm down with that. Down with that.

Then you get sued. That's the way it goes.

Yeah. I'm trying to remember the name of that open source tool. We can probably get that into the notes because there are others as well. I think there's a lot of really cool stuff out there., for adding, uh, for, uh, for being able to build that doesn't have nearly as large a market, right? They don't have the marketing dollars because it's open source projects, uh, that are, that are pretty cool. And I think far more accessible now because of, you know, great with these no-code build—

the, uh, open source is you can do what WP Engine did. You take something that's free and then you use it, you build your brand around it, and then you get sued by the inventor, and then you settle. Or you donate to the WordPress Foundation. There you go. It's easy. It's always easier to beg for forgiveness than to ask for permission.

Right. And I actually think when you think about like building for the ecosystem, WP Engine is— I mean, look, Jason Cohen and Matt Mullenweg got themselves into a whole thing. But the idea of like building support and ecosystem is I think a pretty good one. There are other hosts that are very— like Bluehost is also really large. Out of Utah is a large WordPress hosting system. And so like they, they're, there are many that are, that are in this game, right, that can work relatively well. And so you figure out, hey, wait, there's this like this big market, how can I be part of that or be an important part of that? And that's what WP Engine was doing. And then, and then I think that maybe WordPress is looking to get more money and that gets harder and whatever. It's always playing with others is always tricky. But the, but when I look at like the, the, the, the nature, right, of AI building, this sort of gets us back to the whole beyond vibe coding thing, right? It, I, I think when people looked at like no-code, low-code tools, they thought of themselves as a consumer of the tool to create whatever the experience is. They were gonna go create like a SaaS and go make a million dollars. And what, what, what, what, what the things didn't quite work that way. And, but the sort of the same promise is kind of running through in that term vibe code. I have something, it's going to be easy, it's going to be fast, I'm going to make $1 million. The big Bolt hackathon that's going on right now, the world's largest hackathon, I think is also very much in this space. People looking to quickly build something, make $1 million. But the real opportunity I think is how does this change software itself? How does this allow you to create the thing that's really special about about you and be able to express more you instead of expressing the what a SaaS is supposed to look like. What can— how can we express you in the form of software? And boy, is that an exciting, you know, opportunity because there's all sorts of weird and wonderful software, a lot of which has like significant front ends, some which has no front end, looks more like an automation. The way we can now build agent approaches are also really, really cool. Generative UI, like building with a chat or conversational frontend, which is never not the same for any two people, is also super exciting. These things are all possible because of working with AI, but only works if you use your brain.

You need to use your brain. That is very important. You're completely right with that. I love how good AI is with frontends. I want to share you something. Please. Let's see, can I, how can I share Chrome? Can you share my screen? Can you see my screen?

You can see my screen.

For example, right now for Despia, I'm just using an AI tool here where I just put my, put my design specifications in, right? Deploy using no-code Cloudflare Workers. Got my prompt. I can copy it Toddle because I'm not calling it Nordcraft. I can create a new page, new, and I can paste it and then I get my interface. Now I just need to add functionality. I built this, the only reason I built this is for myself. I built a tool for myself where I can put a prompt that I get from ChatGPT. It builds me the interface in Toddle Nordcraft and it will look one-to-one in there and it will use the native styling conventions and all that as you can see. It will even add hover, focus states, all that. Then I just need to plug in the logic and it's trained to follow my design taste in just knocking off ChatGPT and UI and Vercel stuff.

Cool.

I think also a big benefit in people just building their own systems, building their own tools to work more efficiently. That is simple because I built that whole thing with AI. I built it with AI, an AI to help me build faster with AI. It's almost like Nargrave being made in Nargrave.

Look at that. Well, and you can cut these things just back and forth. You know, earlier today we had a session in StateChange where one of our members had created the MCP server for being able to create function stacks, databases, etc. in Xano. Now, Xano doesn't support in their official stuff.

This is like using their beta API. XanoScript, I mean, I'm always afraid to do or say something about it because they made me sign NDAs.

But if you take a look at your— in fact, if anybody here is a Xano customer, if you go take a look inside your metadata API, you're going to find all these beta endpoints. The beta endpoints are actually shipped to everyone. They're not sitting behind whatever your NDA is. And actually, I probably—

I'll just let you talk about it. I'll just step out of the room for a moment.

Yeah. Like saying Skull and Bones. But now that it's public for everyone, and it practically has been on the quiet now for a few weeks as I've sort of discovered, you can make use of these APIs. This guy, Rob Boulos, had made something. If you put up, it's called mcp.snappy.ai. He has made an MCP server that enables one to communicate with your Xano instance, to, I mean, Vibe Code, again, it's a term I don't really like, but like you can use a chat interface like, you know, Cloud Code or anything that uses MCP, like even Cursor or whatever, to set directives to say, please go make this function stack, please go set up whatever it is in your— It was called VWeb. And yeah, that's where I was about to go with this. So he made the tool, which is super impressive by itself, The tool is the tool he used AI to help him build in Cloudflare, right? For the backend of it. Cause, cause MCP is a server-side technology. The frontend of it he built using WeWeb, frontend of the user experience for signup, for like, you know, management for administration, et cetera. And then the, and then the hosting for it he did through Cloudflare. So it's just about putting together the pieces of the puzzle that at the time you're doing the work, give you least friction to create things you really want to create. And he created something that was really remarkable, blowing, blowing away everybody who was in that room, and is allowing him to be a real 10x developer, you know, relative to Xano, because it has secret sauce. And he's giving it away, the MCP server. He's doing it because this so boosts his reputation, right? So this goes from being something that's just him, you know, him learning about MCP to him being able to 10x himself, to him being able to again 10x his reputation. And then, and that's like the promise of what we can do with the software. It's not just we're going to create, you know, a SaaS and customers and have that sort of everything trickling downhill. But he has that whole cycle going that's like building his reputation, building his ability to deliver, increases his ability to charge rates as a, as a technologist and opens up all sorts of cool doors. And I think people who look at this technology more, you know, again, turn on your brain and go beyond just vibe coding it, there is tremendous value to create and money to be made.

We love that. It's very exciting. I mean, I'm looking at it right now, it looks very promising, and it's very exciting. And to see what, what, you know, what the future just holds and, you know, how AI is changing all those kinds of things that we used to do all by hand. Um, I'm curious, you have insights in, in state change. How relevant is Wizd in the age of AI for people right now?

Um, we have a couple members who are using Wizd, uh, fewer than we, fewer than we used to. Um, the, um But actually the folks from FinSuite came over, I guess a few months ago, gave a presentation on the new generation of Wiz.

Yeah, I took a look at it. It's just they just named the actions events.

Well, whatever, but they're actually pretty thoughtful about how they can be leveraging people who want to build more with AI because Wiz is a neat— I really like FinSuite's approach to Wiz. Because they never had all the money that like, you know, Minecraft does or WeWeb does in terms of having fundraisers.

They're actually profitable. That's the difference with them.

Agency, they never had any access to capital, right? Yeah. And so the thing they do is with an eye towards how does this create value for our customers today? And that's very clarifying. And one of the tricks they've used, they decided, and this, they took a while for them to figure out this, what they were about, but they did figure it out for like, I think the third version. That their target are developers who want to go faster. They're going to provide the superstructure in the form of the Wiz runtime, but then you're largely putting in JavaScript. Helping people get that JavaScript better and faster is something they're onto. Being able to take things like examples of here's the variables you have in context, so you can copy them out and put them into ChatGPT and then copy and paste back. They recognize that was a big part of the workflow. Instead of them trying to make the tool do those things. They made it easy to get the data out of their tool to do those things in ChatGPT or Claude or whatever you want to use, and then bring it back in. Now they became independent of what the state of the art in those tools were. They weren't trying to chase the Joneses in a way that like WeWeb right now is chasing, trying to make a better experience. And they're fighting with Lovable and Bold. They're all trying to make the integrated experience.

AI, what I've seen in the forum, Yeah, people really don't like the WeWeb AI, what I've seen in the forum. Like, I think it was Roberto who commented and said something like, when is it finally going to actually work or something? I tried it out a few times, it just made up things. I said, I want a form that does this, and it says, yeah, we could just use the WWLib activate action method. And I was like, I never heard of that. And then I looked it up, it doesn't exist. And then it just wrote like a whole workflow with that method that was never there. Yeah, but that was like a month ago.

I don't know if that changed. Yeah, it's always improving. I'll tell you, workflows are something AIs do not do super well. It's a hard problem, and particularly they don't do well with front-end technologies doing workflows because most front-end technologies are not designed around doing workflows. There's a state machine for like responding to human interaction, and like we sort of have to force it a bit to have it do more complicated workflows. But the other aspect is I think it's very hard to change your stripes. You know, like the— if WeWeb— and I actually talked to a VP of marketing, I think a month ago, who commented that the company she was working with was community-led, and she did not mean that as a compliment, which I thought was really a neat point. Because what she was saying was that because their community was leading them around, they were, they were responding to what their existing customers wanted, right? And making all of these marginal improvements. Whereas there's a whole market of people who weren't their current customers, who weren't being heard from, they weren't in the community, and that meant they weren't getting growth, right? So they were kind of optimizing for this, you know, relatively small pool of people who had chosen to join them the first time around. That made their initial customers happy. They could have this very happy dance, but the market, but the customer wasn't growing. Growing as a result.

Like I said, as a marketing person, like, well, how do we get more marketing? Similar thing, they're now chasing ex-React developers. So like, I don't know if you saw that they were on CodeTV and they made like a challenge with real developers on how fast they can build an application, and they're now marketing it as Node-based web development. So if AAA games are built in Unity or whatever that thing is called, They did Unreal and Blueprints, right? Yeah. Yeah. Why don't you do a web application that way? And they're now targeting real developers with that. And then also the fact that they're open source. So they completely abandoned now the no-code market. But it was fun. They made a poll in the Discord. How many of the users here actively using Norcraft are, are no-coders versus developers? It was like 70% were no-code. And now they're just like completely going in.

No, we're for professional developers. Well, it's a tricky thing. Like how much money is there in no-code? And even if 70% of their audience is no-code, it also highly correlates with people who are joining the free tiers or the minimum tiers or what have you. As long as people describe themselves as no-coders, they tend not to be an audience with willingness to pay. The people who have gotten themselves to a point where they're shipping, tend to describe themselves as more technically adept. They don't describe themselves as no-coders.

Yeah, I think it's an interesting change because they're aiming for people who right now are building applications by hand, who are using React, and they want to have a more professional version of doing that. Like probably the people in SageMaker are using open source technology to build their applications more visually. I think it's an interesting change. I think it would be hard for people because like with AI, it's just so easy to spin up a React application. I remember like a year ago when you taught me React and you were like, oh, we just use Cursor and create React app and we made like some Node-based workflow builder. It was pretty cool. And I was like, with that big ecosystem, you got React Flow, you got React Charts, you got Vercel with all their libraries and everything. And like, what are the chances developers are going to go with an ecosystem that doesn't support that, that has packages made by no coders? You know what I mean?

I think that's a real— I think the issue ecosystem is a big one. And we talked about WP Engine and Mullet.

That Norcraft isn't running on React or Next.js because it would have such a big ecosystem just opened up to itself.

Yeah, yeah.

But you don't have it because now that you just, you just got to use vanilla JavaScript, you got to bring it in with the CDN. And most modern SDKs are npm only. And you know that when you work with WeWeb, you said, oh no, I can't use the npm plugin to bring that in because it just brings in the CDN. We got to do this as a Vue component, right?

Why not have the Vue component? This is what the AI can do relatively.

The downside in Nargraph is there are no Vue components or React components.

You don't have a code component to work with.

You don't have the component that can bring in the Vue or the React-based libraries. You always have to look for CDN-based libraries, which will bring you back into the age of jQuery.

There's this company that shut down a couple of months ago that I admired a great deal called FastGen. And the FastGen's, you know, FastGen had a few different parts to its pitch, but the thing I think really separated in the market was they had very good code steps. And so they were a backend builder. They kind of competed with Xano, they competed with Buildship. And the crux of their value prop, I think, was that they could do like, you know, boxes and arrows flows for their, you know, building out the logic for like a cron process or for an API endpoint or whatever. But then that you could at any point stick in a code component that could be either in Python or in Node.js. And that meant you could use all the SDKs, you could use all the, you know, pip packages, you could introduce whatever complicated logic you wanted to do. I had it install LangChain. That was one of my first tests with them, and it pulled it off. And the— and you could do really cool stuff inside those code blocks but not have to commit to having the whole thing be code, right? You could like have subcode pieces instead of like a Lambda, which is, you know, all the code for a given endpoint. You could have mostly no-code and then like this one, the one hard part that has it too.

And it's such a great thing. I built a facial recognition system in Xano with the npm package and it's just, it works amazing. It's very accurate. I got a whole facial recognition system in there.

So Xano has now caught up to Ray and it will say it's Ray.

It's pretty cool.

Xano's caught up to that. I think it substantially changes the way one can build with Xano now. When I'm consulting people at the higher ends of Xano, I'm often bringing in, they're like, oh man, but do I want to do this to no-code? I'm like, why are we doing no-code? Let's just use the Lambda for that part. Then, oh, but then I got to do code for the rest. No, you don't. Now you can use the best of both worlds.

I love that. It is a big thing with Xano. I would love an npm import because most npm packages that you just have to do in a build machine, like installing dependencies on Swift or something where you just have to import the npm packages. They do work with no JavaScript. It's just they don't have a CDN. I didn't want to give you a CDN because not compiled that way, right?

You mean on the front end?

You mean for the front end stuff? On the front end?

Yeah, because, because then it has that support for, for the—

but having something like, like Xano did it, where you can just import any npm package and then it will be part of your Nordcraft build and it will be there as a web component or something, that would be something that is— because right now ecosystem gap that I see in Nordcraft. You can't bring in a Vue component, you can't bring in a React component. You will have to use libraries that are CDN only or support ES6 module imports, which is not everything because nobody's using the CDN anymore or loading packages client-side. They're already including that in the whole application because most is React, which that is kind of like pretty sad because it kind of like cuts out half of the modern ecosystem out there.

Then you get to tools like, you've got on the one hand, Nordcraft, which is like we're stiff-arming the AI a little bit. I like the way you're basically using clipboard as a way of transporting information from your microapp that builds things with AI into Nordcraft. You have things that are more like—

They're working on an AI and they said it will be done in 3 months or so.

I guess I gotta do my own one. They'll work at whatever pace they work, but I know I'm gonna have done— I will have built and sunsetted like several things before those 3 months have passed, right?

So like, that means like I need, I need AI right now, and I built it I think like a year ago or so, and it's been working great. And you know, uh, that, that's why I'm so fast, and that is why I can make progress so fast. And I don't, I don't want to build stuff by hand. That is very boring.

Yep. I also think that like the, the fact that AI can build, I think that we, we should ask like, what does that mean for software? Right? Right now we're looking at how can we build the same app you would've built in 2021 faster, right? Instead, what is the app that you can and should build in 2025 that's different from what you would've built in 2021, right? As a result of having access to this technology. This gets back to the point you were making before about building tools. You built that tool that can put things into Norcraft that's allowing you to go faster. There's then a level above that of what can you make for customers that's taking advantage of the fact that AI knows how to create code on demand. That's a whole field that's called generative UI and that I find to be super exciting as the next phase of what front-end development can look like. And the, and the, I, I, I'm hopeful that, cause I like a lot of these low-code, no-code vendors, I like their teams. But the, the, the model they have, like putting you on Rails to build something that looks like a web app doesn't look like setting you up for what's next. And I'm, I'm really excited about like the, the companies like, you know, the Crayon package, Assistant UI is a company that's doing this. Vercel has a play in the space. That are all about creating, creating UIs that are dynamic, that are, they're just different for each person based on the question you're asking, based on the things that are useful to you. And that is totally, totally different. And that, that's also like a whole beyond vibe coding thing, right? It's not like, how do you use vibe coding to make the same crap faster? But rather, what is the new stuff that adds more value over the top?

Yeah, I can actually show you something that I'm working on. Lots of behind the scenes today. So I'm in Nordcraft here and I'm building a component for a widget builder.

We need a Viking helmet or something to go up each time we mention or like show Nordcraft or something.

We'll see if we can get that sorted out within this. But I got, I got like a widget builder here and I had my AI kind of like built, you know, the start for me. That was it. And I turned it into this, which I think looks a little bit nicer already. And then I have my widget deployed as a Xano worker in here, which this is how those widgets get deployed. They get deployed as an API endpoint that loads the dynamic SVG. And this is how we load the widgets into Despia so you can build them in Xano. And then what we're doing is, because I think most people will probably want this no-code, so we're going to have a template on a Xano API that will allow you to build those widgets with dynamic data coming from your backend to be updated on the home screen. You can deploy with no-code Xano or you can deploy with code using a Cloudflare worker. We give you both options and we have like two templates how you can deploy such a widget yourself. What I'm— what I don't want to do is I don't want to provide a service where you can pay $20 a month and we host the widget for you, because I— the last thing I want to do in my life is run a hosting company. You know, I don't need that stress. So I just outsource the hosting part for the stuff that is hosted with Despia. We just outsource hosting to other companies and say, here is a super simple template, you can host that part. The widget right now is the only thing you need to host yourself, and we just give you like a template so you can host it either for free on a Cloudflare worker with a very generous free tier, or you can host it for free on, uh, you know, Xano if you want to have it visual. And then I hope that, you know, now this July Xano will have the $29 plan out there, so it'll probably be cheaper anyway when it comes to requests than us providing you hosting for $50 per million requests, which I thought about doing, but I don't want to run a hosting company. It's too much hassle. I just rather have you pay for the Despia licenses and that's it. And we don't, we don't have to carry your infrastructure. It's too much responsibility. I rather— I do it like Zapier.

I just refer them to Make. Yeah, from that, the old, the old origin story there.

Yeah, because we would just write a Cloudflare anyway, so it will cost us 30 cents per million views. Why don't you go directly to Cloudflare. And if you don't want to use Cloudflare for whatever reason, we've got a template for you that's plug and play. And if that still doesn't work for you, use Zano. It's $29 and you probably already use it, right?

Or, or go use one of these, you know, freer guys like Netlify, Vercel, which have super, super generous, uh, free tiers.

You know, dynamic backend server that's building an SVG. So you need to get data from your database With API call, then put the data into the SVG and deliver the SVG. It's just easier to do that in a worker or Xano.

Yeah, we'll do it super, super cheap because the way that workers work, because they're not really Node runtime. It's also really interesting conversation to have how durable objects work, etc. But the, um, but, and Xano is all about making it super easy. And because they're instance-based, you're not worried about running up a huge bill because it's all sitting inside this one, you know, pre-features.

Oh, I have to ask you one thing because they've been like, they've been like shitting on me in the, in the TOTLC Discord. Is Xano a virtual machine or not? Oh yeah, well, the, the build is not. And I, they, so I said Xano is a virtual machine, uh, it's not like Subbase where it's server, where it's like a cloud instance base where you, you know, serverless and whatever you eat, that's how big your invoice gets. And someone in the Nordcraft community said that's fake news, that's not true. Xano is not a virtual machine, it's like Supabase.

No, uh, Xano is, uh, okay, well, I mean, Xano software can be deployed a few different ways, but like the, the, there's the build plan that's free, uh, where that is probably what he was referring to. It actually has sort of a serverless quality to it. But which is a huge pain for them because architecturally it's so different from what their paid plans are. All plans, all the paid plans, self-serve paid plans, Xano, basically they buy, they rent a virtual machine on your behalf, you know, give it the IP address, give it the URL, and then they load in a bunch of, you know, Kubernetes pods into there that are, that basically sum up to the Xano So your Xano workspaces are the containers and then your instance is the VM? Well, actually the workspace, I'm not sure the containers don't really correlate to the workspace. That's not the axis of it. There's one that's for the, like one that's for like the web server, probably running the NGINX web server. One that's for the database, one's for the Redis, one's for all the Lambdas actually run in another one of those containers as well.

Oh, so you have different containers in your V— so your Xano instance is a VM. That the virtual machine has many containers. One container runs your database, one container runs your Lambda system, one container runs your API and MCP or whatever.

Yeah, that, that's called application-centric containerization. And that is, that's very common actually when you go look at virtual machine built systems like things that run on-prem or what have you. Nowadays it's much more common to see that you have your VM. Your VM is basically running a bunch of Docker containers are running Kubernetes cluster, and the— and then the each application kind of gets their own piece to sort of protect, protect memory, security, all sorts of—

it doesn't happen with Xano when I'm purchasing for $2.49 my VM. I'm not sharing that with someone else. I get a VM and that is only mine, and that VM has containers. One container is for the Lambda, one container is for this, one container is for this, but it's all just for my workspace.

For my instance. Yeah, total separation between your stuff and somebody else's stuff.

Uh, camera to prove I was right, but that's—

thank you so much. There's like a centralized authentication system, and then there's the opportunity to make it more complicated because they, they offer something they call auto scaling, but it's— the technical term for it is horizontal scaling, so that if you have a box and your box is like running against the edge because you're utilizing it too much, particularly like peak usage situations, times a day or times a year or whatever, then they will offer horizontal scaling, which is like parallel machines that basically clone your code but all talk to the same database in order to get more power out of that.

But in general, yeah. I have one question to you. Let's say I'm doing an API call to my database. I'm getting a value. I'm putting it into a variable that's an SVG and I'm returning it on the Xano $99 plan, how many calls can I do a month of that? I need to have some calculation and ChatGPT is not helping me because it doesn't know what I'm really getting with Xano, but you know everything in and out.

I don't know how you could do that.

Just a rough estimate. I can refer people back to the video.

I'm expecting it has probably a gig of memory and Each of those processes are going to take like 30, 40 megabytes memory. You can do some division on that and say you can use roughly half the memory for doing your job. If you have half, let's call it 500 megs and over 30, that implies you can do maybe 15, 20 maximum simultaneous requests. Now the key is you don't need simultaneous requests because the faster each request goes, the more you're freeing up lanes on the highway to be able to move through. What I have found is that a lot of people respond to, I'm having troubles with Xano by paying for bigger instances. When I find that the better move first is you do a little bit of investigation, you can find out what your slow endpoints are.

Let's say I have an endpoint that takes about a millisecond or less to get data. It's very quick when it just gets like one step, get database, set HTTP header, return the data. Right.

So now math out from there. If you have something that takes a millisecond, right, to clear through, and it's going to take more than a millisecond, but let's just use that for math. The implication is if you can do 20 at once, right, and I'm just going to use that number for ease, if you do 20 at once and each one takes a millisecond, you just take the product of those and that means it's 1,000 per second, right?

You should do the product of those two things, you get 20,000 on Xano and how much I can actually do. I just can't reach the ceiling at that given moment?

If you hit the ceiling, it will queue behind. You can get more requests and like, well, you got to wait. This happens all the time with requests. If the waiting happens too long or if the waiting room itself fills up, now you can have a problem and this can lead to running out of memory, thing keels over, you have 503 errors and that's not a very good time. This becomes why you do things like rate limiting. The same user is asking you for a bunch of it, slow it down. This is where you might just throw more money at it eventually if you're really hitting high speeds. But the other thing I'm going to say is there's almost no request that takes like 1 millisecond. Most requests take like 50, a really fast one will take like 50 milliseconds in order to clear the system. Much more likely it's going to take a couple hundred milliseconds. When you consider that, that means your request rate looks much more like 100. Than does look like those couple thousand. If you're slower than that, well, they can go down and down and down. Your big opportunity is how to make that go faster.

Let's say I'm just doing one API call, I'm getting an auth key or something, and I will be running that endpoint every 5 minutes for 1,000 users on the $99 plan with Xano.

Is that feasible? 5 minutes, 1,000 users, Wait, you're running your file— you're running it as an endpoint? You're running as a task?

As an endpoint, like an AuthMe, you know, like when you go to a page to protect it, to see if the page is—

okay, I get you, I get you.

Use case in Despia where your widget runs an API call every 5 minutes.

I'm just trying to see if this will work. Whole base systems are very expensive. For the reasons we were describing. So what you'll find is that if you want something that's sort of trying to be aware, being up to date, this is where using something like a real-time or notification system can be—

can pay some dividends. Yeah, what, what's going on is, um, with Apple Smart Widgets, you can only pull data, but you can't pull it conditionally necessarily. So what we're doing is we have an endpoint where your widget updates in the background. So your application is every You can adjust the number, you can do every hour. But it's every, default is 5 minutes. Every 5 minutes it's calling an API, getting data, updating the data in your widget. Calories, your macros, whatever you build. The widget is always up to date like the weather widget in Apple. I just want to know, let's say you have 1,000 app downloads and you run this every 5 minutes. It's like 50-millisecond request, very simple. Would this work or would that kind of like be when Xano on the $99 plan starts sweating?

Well, it depends on how efficient you are in terms of spreading it around. Let's just doing quick math on it. Like 5 minutes, that's 300 seconds. So if you're talking about 1,000 users over 300 seconds, that means you're doing one request roughly every 300 milliseconds. Yeah, that sounds pretty doable on its own, right? That seems fine. So usually the— but it's not the averages that get you. It's where the—

it's where the peaks are., you know, so like the, the, the more this can sort of be spread out, pretty spread out because every app will run it every 5 minutes.

So it will, and if everybody runs it on the minute, just for the sake of argument, that can be weird peaks. And so you just sort of pay attention like how spread out that is, but that'd be the way of the worker. I'm not going to lie. Well, think of the workers are just so cheap to run. The big virtue of working with Xano, I think, is how quickly and easily you can create and express your intent. But another trick I've used with Xano is I've taken like an endpoint and used like— there's— we have the state change extension for Xano. So like state change, in addition to being in Qt, we've also got tools for our members, and one of them is like this extension for Xano, right? So right now, by my count, there are 3 in-production development tools for Zano. Eli Beechy's Bubble plugin, the state change extension, and Robert Boulos's MCB server. Those are the 3 big things out there. But the state change extension has, like, you can go to any one of your endpoints and click make documentation. That's nice. You can save your documentation. But the really cool thing is you can take that documentation as Markdown and give it to AI, and AI can go rebuild your logic say in JavaScript or something, and maybe you can go redeploy it in, in code or what have you.

And this is a discount for the Xano extension now for everybody using Despia with the widgets.

And that can be something that you might choose to use as a way to make everything go faster. It's like, well, how about instead of going through Xano, we go through Cloudflare and then have Cloudflare call this?

But I think I will have to make a virtual, like a visual builder for configuring those widgets outside of Xano that will just give you the code that you can dump just into a Cloudflare worker. Maybe we will have an integration to one-click deploy it.

And that could be cool. That's for, that's for things like being able to create the code, being able to help you out with deployment to the right place. Um, and then, you know, Cloudflare gets really excited about that, right? That gives them more stuff that is going to drive up demand.

I want to run the hosting as long as we're using your account, it's all fine. I said, right, it's a responsibility I don't want to do because it's just too much support for too less money.

Well, you know that like Windsurf has a relationship with Netlify. Um, they did this partnership like I think maybe a month or two ago, uh, where, uh, they're just to make it really easy that if you build something using Windsurf— Windsurf is an AI-assisted code editing tool for people who don't familiar with it. Um, and then they have like now, uh, easier like one-click or whatever support with Netlify. So now you can get your stuff up onto Netlify if you're doing it with Windsurf. They don't keep you from using anything else because you can use GitHub integration to go into any of these tools. But that idea of Windsurf doesn't want to be in the hosting business either. They want to be in the tool business, right? Business development to facilitate increasing the value for their customers and probably maybe creating some money for themselves, but also definitely marketing value, you sort of get to go that way. So I was thinking like what you were just describing, like helping them get over to the place that'll serve them better. This is really antithetical, I think, to what a lot of the low-code, no-code players do. 'Cause a lot of them are, we want to host your stuff and that's the way you're gonna be paying them. 'Cause I don't like them. But be, be paid for the value you create. I think there, there's a real lesson in there. Like if you're a great tool, be paid for being a great tool. Don't try to get paid for being a host if what you are is a great tool, cuz you'll be a crappy host.

Hosting from all the no-code tools completely sucks. It's always down and it's not reliable at all.

And they should just focus more time on having a good editor and, and put, and, and then they just also need to figure out how to get paid for that, you know? And all of them are trying to figure out how to get paid.

Partway with, with the GitHub integration that they just opened up.

What? And some of them are increasing their rates for those kinds of integrations. But yeah, like, it should be that this tool is so great that you never want to have to live without it. And if that's the case, you'll keep on paying for it month over month, year over year. That would be great. Having a tool vendor try to be in the infrastructure business, I think, is very tough. Yeah, I think And it was— Xano is an interesting play in that, and Buildship too, because they both realize they are in both businesses right now. And whether it's good for them to be in both businesses isn't totally clear to me. But the— but, but like, you know, Xano definitely is about providing infrastructure. That's because you pay for these scale plans or whatever for paying for more iron, right? But then we talk about, well, why do you use it? You use it because of the tool. The disparity between those things is a tension that I think is creates, you know, strategic challenges for them in the market.

Yeah, yeah, I think so too. And, you know, I just— I think I don't want people to pay like for an inflated amount of hosting. My job as the tool vendor should be make sure that the tool is better, that the tool itself provides value, and then charge for that, and not charge by selling them very shitty infrastructure for 50 times of of the price that it costs me to run.

Yeah, and that requires having conversation with the market and that they're willing to pay. I have found in state change there's a willingness, right? People are looking for, for help, they are looking for tools, they're looking for whatever. There's willingness to pay for those things. Um, now the question is like, would that support a venture business? I don't know. But that also maybe points to like what, how the whole market evolves. And, and that, you know, uh, evolutionary aspect of the market can be pretty interesting topic book for us for next time.

Yes, I mean, I think we definitely went beyond by vibe coding today. On beyond, it's wonderful. Yeah, but I, I like that vibe coding companies don't sell infrastructure. Yep, they don't.

Barely, but they barely do. Well, Vercel, I thought, is an interesting play by an infrastructure company to provide a vibe coding platform to help sell more, they're more buying on them.

They're getting filthy rich by all those tools integrating with Supabase. Yeah. But that is very cool. I mean, right now, what my goal is, is to have a REST API coming very soon for Despia. So that's— API, you can upload your web application code as a remote URL, and then it's going to deploy your application and you will have webhooks if there's an employment error and all that. So you can completely use the whole infrastructure. And for that, if you want to get access to that, there will be like a monthly subscription to that, which will be like, I don't know, $20 a month. So you can use the REST API and to get webhooks and to kind of like automate that, because I was like we can go into two directions. We can either start charging money for infrastructure and gatekeep some of your code, which I don't want to do, it doesn't make sense. Or we can start adding features to it that are independent from the core of the technology itself that some of our power users really want and they are willing to pay for them. And what's $20 a month to get API updates for your whole entire workspace? Space, right? It's not much, and if you really need it, it's a good deal, uh, because otherwise it's just, it's just not there for you. Would have to use an email scraper, and that would be too much, right? So we're trying now to add those things in there to have like a subscription we can sell, like Despia Pro, which gives you everything that is kind of like cloud-based, where we have an API for you to deploy things. You can get webhooks when an API was successful or not so that we don't have to become an infrastructure company because I hate doing that.

We should quickly put up the Despia QR code.

Don't get anything wrong, we're doing great on selling the templates alone. It's just that people ask for it. We don't need more money, we're doing fine. It's just nice having recurring revenue that you can count on so you can hire more people in engineering. It's not like you have this amount of sales this month and it's astronomical and then next month you have like 5, which happens when you sell a one-time purchase. We're trying to see how can we provide more value for people without having to gatekeep your code and abstract your ownership from your application. How can we add things to power users who really want that. So like every time there is an update, you can get a webhook and then you can automate that with your Slack to notify your client that there was a new update to their application, has just been submitted to the App Store and all those kind of things. But those are things we're exploring. We don't need to do them right now because we're doing great.

But I don't want to be a company. No, we will, we will, we will continue to not being hosted companies. I think that is a good Good ambition. Heavy Lambda with your hosting company. Oh yeah, well, the Heavy Lambda, Heavy Lambda as one of my initiatives has sort of faded a little bit as Google Cloud Run got better.

I don't blame you. Also, you don't want to be in the cloud infrastructure business.

Do not want to be in the infrastructure business. So how can I be more in the tool business? The closest we have to that right now is Scriptag is actually a pretty good place to be in the hosting business because people who are using it have a hosting problem, right? And this solves it through like micro hosting. Uh, but the, um, but, but things are like bigger, like the kind of problems Heavy Lambda solves increasingly sort of finding that this, the, the better tool for the job is to use something like GCR, uh, and plus AI dev, right? And then you put those two things together and like, man, there's so much you can create so fast and it can handle a lot of weight.

Yeah, that is absolutely— that is amazing.

Yeah, wonderful.

Perfect. It was a pleasure talking to you. I can't wait to do this again next time.

Absolutely.

And wonderful. Thank you so much for all your time. And thank you so much for