#59: How this AI duo found PMF by building 3 products in 9 months | Andy Keil & Kyle Ledbetter

So our initial plans were that I was just doing a prototype basically and we'd get real people to code to come in behind me and as we got further along and there were challenges frankly with getting this people but at the same time AI was getting better and the products were getting better. You got to remember this is before cursor came out, right? So it's like we were copying and pasting things in chat GBT and anthropic cloud Back and forth and like cobbling things together. And sure, we had VS Code, but it wasn't great at

the time. That's gotten better. But yeah, things matured quickly around us to where we said, you know what, maybe we don't have to give up that equity. Maybe we can actually get this to market. [Music] What if I told you that two nontechnical founders, a head of product and a head of design with zero CS background, built A complex technical product without a single engineer. And not only that, but they're operating at a level of AI mastery that makes even seasoned AI users look like beginners. Today's episode with Kyle and Andy is absolutely gamechanging. These

guys aren't just using AI tools. They're literally reinventing the playbook for product development in real time. We're talking oneshot megaprons, bakeoffs between Prototyping tools and seamless handoff process from design to code that will blow your mind. If you're a product leader wondering whether you need a technical co-founder or if you're building a side project and think that you need to learn how to code first, this episode will completely shift your perspective. Kyle and Andy prove that with the right AI workflow, nontechnical founders can move faster and build more sophisticated products than traditional Engineering teams. We

go deep into their process from high fidelity to low fidelity, back again. And trust me, the insights are incredible. This episode has tons of screen sharing. So, if you're not already, I highly recommend watching on YouTube or Spotify video for the full experience. This one is fire. Let's dive in. If you enjoy this conversation, don't forget to subscribe to the podcast on your favorite podcasting app. Follow Us on Substack. Give us a rating. It'll help us a ton and help future people discover us. while also making sure you never miss an episode. Thank you, Kyle.

Andy, great to see you guys. Tim, Mark, good to see you, too, Mark. I'm so excited for this. And I've been thinking about what's the best way to kick it off. And maybe I'll drop two like insights that I think is just really interesting about, you know, the way you work and maybe what I've heard from Talking with Andy and just some context. I've met Andy probably over a year ago through Supra and yeah just really talk enjoyed talking to him. He was the head of product at Quotapath which was a series B startup. Now

he's a co-founder with Kyle. And yeah I think what's really interesting is two things. One is that Andy and Kyle are building a pretty technical app but none of them are actually like trained engineers. Like they've done some coding on the side but I think that's just like super interesting. And then the other one is I was talking to Andy was like even though he had 20 engineers or so more or less in quapa like he he was like telling me hey I feel like I'm moving faster now with two people than I was in

my previous team. I think that's just so interesting. So maybe let's start very quickly by you guys I think it'll be helpful for describing the context of what is the app that you guys are Building and maybe like high level what do you think allows you to move so quickly in this new world. Yeah, this is fun. I just want to say super glad to be here and Andy's worked with teams that of that size and I've worked in some enterprises of 10,000 30,000 with teams of hundreds or thousands of engineers and neither one of

us like you said are trained engineers and we are where we are in a matter of months and sometimes weeks. So if we can't tell the story of Why AI augmentation is amazing, I don't think anybody could. But I'll let Andy take the serious question you just asked. Yeah, I mean I can I can start uh at the top dream base at large what we have laid out we started in a much different place but where we are now is we're building tools on top of superbase which superbase has become the AI native database backend

as a service off is rolled in there's edge functions lots of really cool stuff and as we looked at Where we were bu building where the industry was building we spotted a gap where it's really hard to get started and then start operating your little MVP or proof of concept as a real business. And there's a lot of people like that's now possible. You can go zero to startup in a few a matter of weeks. And I'll let Kyle share kind of the story of how we got to here. But we really zeroed in on

how do we provide product analytics out of the box monitoring and the ability to Build some internal tools on top of these cool new abilities to bring an MVP to life. Yeah. And so amazing Andy being head head of product in his career. I've been head of design, head of UX roles like that in my career. And yeah, for us to be able to go from an idea, which is not even a year ago. I live in an island, Galveston, Texas, and Andy was down here in a coffee shop, and I showed him a new

feature at the time from Anthropic Claude called Artifacts. He Was wanting to build something, and I said, "Why don't you just give this a shot?" It was going to be like a three-month contract and in 30 minutes he turned his computer around and he's like, "Did I do it? Did I build it?" And it was an interact a series of interactive apps. And I said, "Pardon my French." I said, "Holy crap, if you can do that, no offense with your skills, then anybody can do this stuff now." I said, "This is the product." So, we

felt Like we were on the bleeding edge with building artifacts, as the world was calling them back then. It kind of just turned into apps. And way before Vibe coding, it was like jazz coding. I don't know what the precursor to vibe coding is. Never heard that before, but that's a good one. Yeah. But we were like, this is the product. And we grabbed some open source tools. One of them is a really cool sandbox as a service called E2B. E2B.dev. Longtime collaborators with us Now on this project. And we basically grabbed their service and

we were able to build our version of that with Nex.js JS apps, executing code and building full-fledged apps. And that was months ago, right? This is actually before VO and Lovable and Bold all came out and we knew we were on to something, but it's just two of us. So, while it's a great thing, we also didn't have a team of people to release this and a bunch of money in the bank to burn on LLMs. So, We kept iterating on this and AI helped us all along. Andy was doing gorilla research with people, funneling

back to me directly, and I was doing design and development. And we can get into talking about the process about how we did that so fast. But basically, we were able to quickly start really broad and then get more and more narrow over the last nine months to now settling on this AI ecosystem, if you will, of Superbase. Every app now integrates with Superbase. If you want a database, every front end is Shad CNN UI. So, we use those kinds of tools and we're just focusing on that niche and we're saying there's a bunch of

tools in the market. Dreambase fills this big gap of cool I vibe I vibe coded an app how do I turn that into a real product that's actually functional and secure and how do I ship that so we focus on things like monitoring analytics and management inside of that so that was our journey yeah and I think If I remember correctly from talking to Andy like you've basically built like three distinct products like end to end that are functional with front end backend deployed in nine months which honestly is pretty unheard of like before this

era and like it's only two of you. So a none of you have been like formal engineers. So how let's let's get to the process of why do you think you're able to move so quickly and it sounds like Andy has been doing a great Job with the user research but it sounds like you know you have a problem you identify you do some research you've done the research then once you have an intuition of okay here's what people want like yeah what can be yeah like steps through step of like how you do that.

Yeah, before I want Kyle to answer this, before he jumps in, I really want to clarify. It's more like one of us because I do little to no coding at all. We handle a lot of things Conversationally. I spent a lot of times with customers and bring them back into the mix. But it's mostly just Kyle who's coding, which I think makes it even more impressive, especially given his background as VP of UX. Yeah, actually. Yeah. Before we get to that, I think there might be something interesting here. So, can you guys share maybe like

your team of two and Andy? Yeah. Product background, Kyle. Like UX background in leadership, but also like uh you know, We can call you like a weekend hacker if you will, that did some coding. Like how do you split your responsibilities? Who's doing what? And how intentional were you about that in the beginning? And how has that shifted over time? Yeah, I think we we spent a lot of time on that before we decided to actually jump in and do this together and at the early stage there's a lot of moving pieces. There's a lot

of people that want to be involved. There's a lot of Roles and responsibilities and delegation and there's there can also be a lot of overlap. We're both really what I mean to toot our own horns a little bit. I think we're both pretty excellent product people and so it it wasn't the easiest decision just to sit down and decide where we were going to spend time especially given that it's only two of us. I think the kind of short answer is that I raised my hand to do a lot of the customerf facing stuff. Spend

a bunch of Time with people cuz that can be a lot more sporadic and like my calendar is owned by other people more so than myself. And that allows Kyle to say a little bit more heads down on the product and really get larger chunks of coding time in because these and he'll tell you about it and I've experienced just a minor portion of this. But it can feel like you're wrestling. It can feel like you're in a bare nickel boxing match or you're like held hostage by the LLM. You're like almost there but it's

not quite and you're like one more prompt. One more prompt. And so I think we just to be more specific, we did run through like a a form a little bit roles, responsibilities of what did we want to do, what do we think we're good at, even if we maybe didn't want to. And then also just like the needs of the business early on and we spent a lot of time talking about it. A lot of trips back and forth from Galveston and Kyle Here to Austin and I think we landed in a pretty natural

spot. But Kyle fill in if I missed anything there. And the cool thing about it is we didn't intend to not have a technical person. Actually, we had actually ced a couple of uh founding CTOs. We thought we might get we thought we were going to be at least a team of three if not four when we started. So, our initial plans were that I was just doing a prototype basically and we'd get real people to code to come In behind me. And as we got further along and there were challenges, frankly, with getting those

people, but at the same time, AI was getting better and the products were getting better. You got to remember this is before cursor came out, right? So it's like we were copying and pasting things in chat GBT and anthropic cloud back and forth and like cobbling things together. And sure we had VS Code, but it wasn't great at the time. It's gotten better. But Yeah, things matured quickly around us to where we said, you know what, maybe we don't have to give up that equity. Maybe we can actually get this to market and then we'll

bring on engineers when we're either more funded or making more money from customers. Makes sense. I think the really interesting observation that I think it's unique from the pre-AII versus postAI is that this like very intentional distinction of there's one person in the team that like you They get a lot of deep focus work and protecting their time is super important versus the other one that's maybe like kind of context switching and maybe doing a lot of the customerf facing and maybe like they have more flexibility of when they can take meetings and I think

that's a different paradigm than when you have multiple co-founders. is that they're all wearing multiple hats. So, I think that's like an interesting part. And I think having played with tools Like cursor or replet, I totally see why that's the case. Like I feel like I've been like in rabbit holes. I'm like, "Oo, I'm so close." And I if I feel like if I lose all the context, I might waste like 45 minutes getting back into it. So, that's like a really cool way of thinking about responsibilities that I had never thought about. Yeah, I

would I would say and that one more man that one more test is the bane of my existence sometimes for sure. You guys are saying I want to go to the gym or I want to get lunch or I want to go spend time with my wife upstairs and he's I just got one more test to do. I know I know this is going to be it. And I tested ah damn we almost got it. One more. One more. Amazing. And so let's get into it. Like so and I correct me if I'm wrong, but

I feel like the whole like customer discovery part of it like maybe has changed a little bit through AI. like maybe you can run the analysis through And it can help you do maybe connect some of the dots but I think you still have to do the hard work of meeting customers talking to them asking the hard questions. So I think maybe we can skip that unless there's you think there's something really important there but I think what what is really interesting is once you have a sense okay here's I think what a good MVP

should be like how do you go from that kind of like highle idea to then Actually something that's workable and that you can share with customers. So so yeah let's go through that process. Yeah, I think you nailed it. And just to tee Kyle up, the there's nothing novel about the the customer research and discovery, but the thing that is really cool is we can get much further into it much faster. So, we can actually put a real seemingly workable prototype to the point where we had several people, what you'll see here shortly, who walk

Through it and like actually thought they were going to be able to like connect Superbase in the prototype that we put together. And I think customer discovery has elevated with the ability to build these MVPs so fast with a lot of this tooling. Yeah. And I'm going to illustrate, we put together just a quick board of a lot of the fun things that happened. So this is day zero as we call this backwards tool how a lot of people start. Notice I'm on the whiteboard Next. But we actually first went to database.build which is built

by superbase. It's on top of pglight. And one quick one quick side note. So, if you're listening to this right now, highly encourage you to go into YouTube or connect to Wi-Fi and check us out in Spotify where we also upload videos. But we also try to explain it and narrate it as much as possible for those that are audio only. Good point. Okay. So, imagine a dark mode UI with a database In front of you. And I will say really quick, if you're watching this or listening to this and you want to look closer

at any of this, Kyle and I are very easy to find. So, reach out. We'll walk you through this as well. Yeah, this is definitely high level just representational and the zoom in is pretty crazy too. But so database.build is just a place with natural language to talk to an AI and it starts actually building a schema for a real database And from there you can actually take it into superbase or take it out as Postgress or wherever you want to go. We did this because we did already, as you guys have talked about, we

had two versions kind of under our belt where we had already started and gotten feedback and we knew we were too broad and we needed to focus. So this was a little bit of a rearchitecture of what we had in terms of our database, but it helped us focus on what's the architecture Moving forward. And I just wanted to show and we could talk about this a bit more about a database.build that's like a tool that anyone can use basically. And yeah, absolutely. It's a it's a free tool here yet and I even have it

open. It's database.build and you just talk to it on the left. It builds the schema on the right and then you can deploy it and things like that in the top right. This is going to be fun because I even just this first step is quite unique, right? Of like I think usually people start with wireframes. You guys are starting with a database. So it's crazy. Can you see the conversation on the left? There's lots of going back and forth with AI about that's not the relationship I want with this database. So it's very technical

way to start. Yeah. Yeah. So, I'm curious like I mean if someone is listening they're like okay that sounds like a pretty daunting way to start right like start to just Picture the database before you even can picture the product. So what are some of the maybe the key questions that you guys are asking yourselves right when you're designing this database? Yeah. And this actually the first part of the prompt that if you're watching that I'm showing now. It's it really does start with a basic outline, right? What do we think we're building and we

just talk about app profiles, workspaces, projects, chat sessions, Integrations with APIs, just the bare minimum features that we wanted for Dreambase. Of course, then we get a little bit more Superbase specific in the conversation because at this point we we knew we wanted Superbase and we know a little bit about Superbase, but it just builds from there. then it's just more talking about those same features and then all the columns you want in your database will in your tables and it takes it from there and Runs by the way. Yeah, Kyle, I would I think

it's also worth pointing out we did do a little bit of like UI like whiteboarding, wireframing exploration, playing around with even lovable just to get some ideas out and that also helped fuel what we wanted the database or how we wanted to model the data. Got it. Exactly. So it sounds like yeah you had some high level of okay we're going to want some projects some tasks and that's maybe was what was in that that level Prototype then when you have a high level okay I think we've got like the high level what we want

then that's where you click that and then translate into the prompt. Awesome. Yeah. And and some of that can just happen just like this. And this actually we're showing this before and after because this is actually our process that we do over and over. It's iterative. So, even after this one, we did get on a whiteboard just lean against Andy's wall in his House at 11:00 p.m. And the next morning, we went to a coffee spot in Austin called Flat Track Coffee, which is great coffee. And just in a notebook, four pictures. And this was

I don't this is dangerous, so people don't do this if you're driving, but I actually had to get chat GBT to recreate the image of me driving and talking to my phone. But as I was driving home from Austin, I was able to get these basically the first mockup images of what I thought the Onboarding process should be like for this product that that jumped into wireframes that I created in this tool called whimsical and then it got to cursor and I think this is something that I want to call out the product people that

everybody can do. One yeah one one text because yeah I definitely there's a lot of questions I want to get into. So they basically for the record I was texting him back and told him to wait until he gets home. Yeah, I want to Get into why he was, you know, talking to Chachi's car, but but we'll get to that in one sec. But the So it's interesting. So day zero basically the first day and maybe you're playing with Lovable. You get a sense of okay, here's high level what it would look like. You get

then you create that database. Then you maybe take that database design that you got after you prompted database.build. Then you maybe made some changes with The whiteboard in person, right? You're like, "Hey, like actually why don't we do this or that?" And then it sounds like after you felt good about that database design, then that's when you're like, "Okay, cool. I think we let's go back into maybe thinking more intentional about the user experience and how we want to on board people." And that's when you moved basically that's day two basically when you start moving

um into notebook sketches to just kind Of get those ideas out there. And then the day of the crime, so day two, that's when Kyle was driving back home and you're basically you're what are you doing? You're basically talking to Chajb PT to be like, "Hey, here's what I want to create and based on the notebook sketches and then Chaj just created those images." Is that more or less what happened? Yeah, this is as soon as the images feature just came out in Chad GBT. So, this wasn't long ago and it was just working through

it. I knew it. We had really high level sketches and I knew what I wanted the flow to be, but when I got home, I wanted to immediately get to work and build prototypes. I had it run through it for me. I didn't think it would be that good, so I got distracted by it and tried to prototype the whole app on my ride home. Yeah, that's what I wanted to get to. Like I've tried Chajip images and they Definitely don't look like closely as nice as the ones you have here. I think one is

like the theme and the colors like look really nice. I'm curious like what kind of prompting were you giving Chachd to get to those images? And for those like just listening like we're looking at some onboarding flow that looks very fresh like a really cool like green color dark background some really cool like placeholder charts. So I'm curious yeah what was your how were you Sweet talking JBT to come up with these cool and on boarding close. I think I'm guilty sometimes of brushing over prompting assuming that everybody just prompts the same way. I when

I prompt since I'm so used to using tools that have been it feels like been around forever now like midjourney and we use replicate.com and we can use 20 different open- source image models. We spend a lot of time creating images via our app and therefore I like type out Prompts as if they were created by an app. And I I do things like I list a color palette and I say reference superbase because we knew we wanted to be to feel similar to superbase and I say words like isomorphic because I knew I wanted

isorphic icons. So this is where this is further proof by the way that AI is not coming for our jobs is people that are already good at these jobs are just better with AI. A non-designer wouldn't know how to Describe a polymorphic server or an isomorphic server or charts that are based off of recharts or charts. That way the AI is actually going out and looking at these charts and it's already looking at the chart libraries that I want to use to prototype this thing. So if you just dump that into the prompt in a

certain way, you know, you're going to get something way better. It's not like you were just talking to for one minute like a quick, hey, create like an Onboarding flow that has these features. It almost sounds like maybe that was more of a 10-minute or like maybe voice note where you're getting pretty specific and not just like the color palette and the design guidelines, but also the type of grabs you want, maybe even like screen by screen type of experience. 100%. screen by screen and yeah, it's like a more like a therapy session than a

quick prompt. I'm talking to it for a while and I've got a Cajun Accent, so who knows if it's understanding me when I'm talking. Yeah, surprisingly the whisper as someone who has an accent, the whisper models are pretty good at catching and understanding and getting most of it right. Sweet. Okay, so then you're driving home, you're talking to Tajibd for a while, being very specific. And I think the other part that's interesting is so the wireframes, how do you said whimsical is that is that what you used? Yeah, that's the tool that we're actually looking

at right now, Whimsical. And it does, it's a lot like Fig Jam and those kind of tools. It's a precursor, but I really like it because it's only built for wireframes really. It's got all the built-in tools for this. And it forces me to do low fidelity cuz since I go straight to code instead, I don't I barely touch something like Figma ever. So, it's really important for me sometimes to Just get the bare essentials onto the screen and not make it look pretty. That that's also like a counterintuitive step that I'm glad we're touching

on because like another thing that's interesting that I hear a lot is okay with things like Bolt or Vzero or Lovable, right? Like why would I do like lower resolution or lower fidelity mockups when I can just get beautiful things a lot faster? And it sounds like with Whimsical like my understanding of the Tool, you still have to add all the components yourself, right? It's not like by coding the thing. So yeah, maybe walk me through like yeah, why is that still important to you and what are decisions that you're making that that feel very

critical to those flows? Yeah, and Andy can attest to this and this is why it's a we have a great partnership and you need two different kinds of personalities working together on this because I will go straight to the high Polish sometimes or I'll put in a lot of work and because it's really easy to go in a hole by yourself, especially when you're remote. This is great because everything's intentional, right? So it's not just me popping it out in VO or lovable real quick or even going straight to cursor as I do a lot.

It's no no I want these exact buttons. I start out also thinking about the kind of text and language I want on the screen. Right? And the big benefit of Doing it like this is now you just copy and paste this wireframe into another tool. Right? So everything from the left carry forward carries forward to the right. It's not just it's not just text. It's not just images. It's the culmination of everything. So, a lot of these this wireframe and these little prototypes that I did, I actually copied and pasted them over to cursor because

cursor can accept images. And it feels reverse. It's almost like I we went Completely upside down. We did database architecture, we did low fidelity wireframes, we did markdown text, and then we actually get finally start building our product. We go all the way back down to plain text here before we start building. Yeah. Yeah. And I think I was just going to say before we get into to cursor, I think the like for me what's most bene beneficial from like zeroing in on the wireframes is we do a little bit of ideation. We get our

heads Around what's possible. We do some playing around in different places, but then it's it's Kyle and I aligning on the copy. It's not and it's not just what's going to end up in the app and on the page. It's the bigger picture like who are we targeting? What's our ICP going to look like? I can now start to take this and translate it to user research discovery. Who do I need to go find on LinkedIn? And so we take this pass. And after we get it like pretty Dialed in, then it's mostly like Kyle,

and correct me if I'm wrong here, but it's easier for Kyle to go in a hole and start coding and creating the thing. And I also have my kind of path to run down to go find customers and go start having these conversations so that when we have the kind of final prototype ready to go, we can get it immediately in front of people because the thing that we're optimizing for is speed to learning as fast as possible. How do we like he and I can build and iterate all day long, but until we get

in front of a customer, I don't think it's quite real for either of us. Yeah. Yeah. What's really interesting about this is that I wouldn't even say it's backwards. It's almost like you feel it feels like you guys are bouncing from one world to the other back and forth a couple of times until you start building like you to like recap you started with lovable then you build the database then you jam on The database um with AI then you get into whiteboarding in person then notebook sketches that again are like kind of low fidelity

low definition then you go back into higher fidelity with TAGBT to visualize things and then you go back to lower fidelity to wireframes and that's where you're actually being really intentional which is super interesting and then it sounds like then the next step is then you jump right into building with cursor and before you Share like what you do with cursor I'm curious why jump directly to cursor as opposed to like maybe using something like replet or vzero or lovable yeah what's the thinking there yeah and you're actually right about the path and I like

how you did call out that we go back and forth it's almost like fidelity pogo you hop from one to the other back and forth real quick. It's good for the brain. It's good for creativity. So, we don't go 100% straight to code here. We Both use cursor for autocompleting of markdown and to talk to an AI agent for markdown before we even get to code anyway. So, we always in our apps, you can, it's kind of hard to see probably, there's a hidden folder called documentation with a dot so it doesn't get tracked. So,

we do markdown files in there. That way we can collaborate on all these thinking about like product requirements and product architecture and stuff like that in text. And I think That tools like windsurf and cursor and vs code are way better than any kind of like autocompleting AI tool out there because you can actually just have an agent session that'll write markdown for you because agents like markdown anyway, right? So this now turns into our real like source of truth now for everything. So we don't just code from this. We actually take that and we

created one prompt and then we went to VO lovable and bolt and we did what we call a bake Off like a vibe prototyping bake off and we use the same prompt in all three and see what we get. Yeah. Yeah. Before we get to the bake off, I wanted that's okay. I just didn't see that going that way. But it sounds like what you're using cursor at that point for after the wireframes is actually to create like almost like a PRD. If a PRD had a baby with a very long prompt basically. Is that

a good way to think about it? Yep. Absolutely. What we It's a PRD technical design doc and a UX design doc. And you'll actually see I'm going to spoil and say we actually baked this into our product. We think our big hypothesis is that th the combination of those things is not only how you get a good a good result in an image, but it's how you get good AI results in general with AI. that kind of thinking, product thinking because AI is designed to think like humans. And what I find like really fascinating about

this is that cursor Would be the last place I would have gone to do this work and collaboration. Like I I feel like I would have opened a notion doc or maybe I would have opened a Google doc. And I know you just touched on that a little bit before, but like why do you think what's your pitch for why more people should be doing this markdown? one this I think the way this super long prompt I don't think a lot of people are doing it and then two like what is the case for starting

to use Corser for writing more documentation for writing more PRDS because I think honestly you guys are the first people that I meet that do this it's a basic need what I was saying and can agree with this first of all everybody knows how to write markdown at this point every product supports markdown and AI is designed to ingest markdown an actual heading in markdown is more significant to an LLM prompt is designed to look for those things. So since it's I ever use The word ubiquitous, but since it is so ubiquitous, why not just

embrace it more fully? And also again, sure, Notion and some of these products have better autocompleting with AI, but they're not going to beat Cursor or Windsurf. Those are like the best tab autocompletion tools, I think, on the planet right now. So why not just embrace it and actually put it where the code lives anyway? So we built up all this code in the left nav of cursor around that one folder. It Literally started with documentation and then we added cursor rules and then we just bolt from there basically. So how would I get let's

say I have and I don't know if you feel comfortable like doing like a live demo or if not you can just explain it but like how would I get started with this? Is it just like as simple as I create a folder with a kind of txt file and I just upload images to to cursor or is there more to it? It's just the folder. And I'll see if I can Pull this up properly without too much craziness in my cursor window. I think it would be cool because I I think I've used Corser.

I know what you mean by autocomplete, but some people might not be able to visualize the magic of autocomplete, which I think is what you're trying to get at here. Yeah, absolutely. So, all right, let's see. Where are you? Cursor. So, this is the actual repo that I was showing there. And if you look in The left nav, there's documentation and there's product requirements and this is highle product requirements and then it gets it it's it doesn't start with anything more than one first bit of text, right? like we I literally type that first statement

at the top and then when you start typing and if if you see it's already trying to autocomplete the next thing when I let's see if I just do pound it does product overview and then if I let it do product overview then I Can say the app will have bang it's already trying to tab complete you know what I'm saying and the more context you add the more context it feeds it so the better the autocomp completion gets is this cycle and this is one file And the crazy thing is as I add files

then I go and add cursor rules and I add like a custom cursor rule and I tell you can just add another line here that says now reference my product requirements. So everything in cursor moving forward will will reference this file. So then I can jump to the next file which is this one and I get into the color palette the look and feel. This was actually one that went into our overview for our landing page which we'll show you shortly. If you go to dreambase.ai I this was what that turned into. So it's auto

completion here, but then it's jumping over to the right and then you can actually ask an agent to say, "Hey, Take over. I'm going to go take a break and go get a bite to eat. Can you fill out the rest of this PRD based on what I've typed so far?" And you can come back and review the PRD just as you would review if you're building a new UI or a new feature. Amazing. And the way and the reason you do this in cursor as opposed to like in cloud or and I know probably

using the same kind of models under the hood is the reason for that is that it already like it's typing it in Context in the right file in the right folder and then also that and this probably lives in does this live in basically GitHub so then what that means is that Andy also has access to it and it's also a great way to collaborate is yeah we can push it up and collaborate it and yeah and then it's again it's context with all the tools like cursor and wind surf you can Now actually like

at mention documentation. So over on the right if I if I start a new chat and I Say at and I and I do docs you can actually add docs. So you see all the autocompleting docs in the top right. I can add docs now just for dreambase for us. So it's cyclical. It feeds itself. And like you say you could use and we have used cloud. He and I started using claude right around when they released artifacts. they release projects and projects can have like project files and project context and that's great but it's

almost like a training wheels Version of this so why not just go for the full version it's not that more complex really and it's actually more convenient and yeah let's maybe talk a little bit too about the architecture of this so it sounds like you have like almost like product requirements and then you have yeah what other kind of what are the key kind of like building blocks that you have be that you have some prompts to yeah so the requirements is it starts high level but it gets very Detailed. This is this is a

text version of every feature in dreambase. This us this turns into an actual SQL documentation. So I can feed this now back into cursor and it generates migration files for superbase. Literally our database didn't exist at first until I fed this into cursor got out SQL files that we copied and pasted into superbase. that built our database and now that's all even better because this is before the days of MCP. So now with MCP you don't even really have to leave cursor you could just say use this architecture and pre-populate my database just talking to

superbasis MCP. So again it just always boils down to either the document at hand or the other documents that you're consuming to build the next step and it just keeps moving forward. Yeah, this is really cool. And but it sounds like the first thing that you do is always like that product requirement, right? That's maybe Like what we're getting to that then will connect to the prompt, the one prompt that you guys have basically. Yeah, the mega prompts. What do you call it, Andy? Stuffed prompt is what I tend to call it, but I don't

I don't know. We're workshopping. If anybody listening has a better name for a very long oneot prompt. Yeah, oneshot stuff. Mega prompt. Mega prompt might mega prompt might be like that rolls off the tongue a little better. I like mega prompt. I'm Going to go ahead and edit that right here. Nice. Boom. Look at that iteration. Uber pro, but you might get sued by our friends at Uber. And yeah, so this is what it turns into though. And then it goes bit to this next step. And this is where Andy brings these prototypes to our

users. As you can see, our very first validations. And again, this was day five by the way. It feels I think we it feels like we've been talking longer in this conversation than It took to actually do this stuff. Um, but we got here and and we get immediate validation. I looked it up. We started on the evening of March 25th. I think it was like 11 p.m. just before midnight when we were doing some of the wireframing. Okay, I think we should focus here. And then it was on, ironically enough, April Fool's Day when

we got in front of three different customers for the first time or potential customers for the first time. Yeah, let's talk a little bit about the I want to make sure we don't glaze over the by prototyping bake off and so basically just like a recap. We you guys did the mega prompt using cursor and that was kind of like using the autocomplete to create that PRD. It sounds also like you pasted some images from the wireframes to help with that process, making it quicker. And then once you have this mega prompt and just to

describe it for people that are Listening, I would say it's probably like seven or eight if you were to copy and paste it into Google Doc like that would be like seven or eight pages basically. It's really long and detailed. And so then the next step is basically they copied and paste it into three different tools. Vzero, Lovable, and Bolt. and they just let it run probably from the spectrum of five minutes to 10 minutes or maybe more. And then yeah, why do you let's talk a Little bit about why do you do this bake

off, right? Like why like people are like, you know, why why are you using three licenses? Why don't you pick the one that you like? What is the benefit of of using these three tools at the same time? Yeah. And I can Yeah, that's a great question. I think one of the things that's really interesting about that that make a prompt is that we're not doing just one page. It's actually like six different pages in one prompt. So, it's it's kind of saying, "Hey, basically this is the full outline of the app that we're looking

for. We're kind of expecting there's transitions in there. There's a lot of really cool stuff which I think is relatively unique. We don't see a lot of people who are doing that. I think the and even some of the best practices that have come out from like lovable or bolt or any of these prompting guides have been focus on one page or a component at a Time." And so to answer your question mark, I think a lot of it is the AI can tend to be smarter than us or surprise us and come up with

stuff that even we didn't think about. So I think there's some level of leaving room for spontaneity or oh wow, I didn't think about that. That's really cool. And it just gives us a lot more kind of creativity to play with while we're trying to figure out exactly what we want the thing to do. Even though we got Pretty dialed with the wireframes and like had a strong vision for what we wanted, I think we're still open for what else could be there or what else could be introduced. And that's hallucinations within reason can be

exciting and when they come back and I'll let Kyle speak to it cuz he he spends a little more time thrashing across all the different tools. But that that's the cool part for me is that when we review them and we sit down, we're Like, "Oh, we really like how Vzero did the charts and built out this sample report and handled the interactivity and we like this graduated experience that Lovable did or put together." And so that's and Kyle will talk about it in the combination phase, but we mix and match or think about it

a little bit like Legos, I guess, is what's in my brain. So it sounds and by the way is that so it sounds like okay you let them all cook then once they're like fully Baked are you guys doing this review together kind of like jumping in a call and like discussing back and forth and I like this part of this part sounds like maybe you blog 30 to 45 minutes and you probably have a granola or similar taking notes and then basically like you you have all all the learnings there like all the comments

of like here are things that I like here are things I don't like and then what happens after you you have okay if I could have my Perfect remix would be combining all these different factors. So yeah, what happens next? Yeah, it's fun. Just exactly the point that you just said about remixing things together and the way that Andy described it's perfect. The other side of it I'll say is that we embrace the difference in models and products, right? So even within our product, we use 10 different LLMs for different things, different agents, different task,

and there's always Different strengths and weaknesses. I'll give you a good for example. VO is probably the best at Shad Cien because they created Shaden. It's the same team, right? And then they created Nex.js. It's the Verscell team. And for those that are very ignorant like what does that mean? Touch it. Yeah. So probably easier just to show visually and imagine this from vocals, but Shad Cen is currently I'll use the word again the latest ubiquitous library on the planet. I think we used to use material UI which is MUI but this one has become

the favorite for AI users and the favorite for apps that build with AI. It's basically built on top of a headless framework called Radics and then this adds Tailwind CSS and add styling. And what it does is it is basically food for AI. Think about it that way. Each one of these components accordians alerts that's nothing new, right? That's been around since jQuery UI and Bootstrap CSS. But they've made it to where this code is easy to consume and easy to propagate with AI. So not only did they give you the code and is it

easy to remix, now they've even got the registry which with one line a person or a computer can install this thing. So AI can go and just pick what it wants and install these things on the fly. So it's become by far the most popular UI framework on the planet. And it doesn't just work with one one one other Library. It's got vanilla react v next.js JS, whatever you want to use it in, it works in it. Yeah, for those listening, the library that Kyle just showed looks a lot like Notion in a way. It

has that vibe to it. I don't know if that's a good way to to explain it. It's just utilitarian, right? It's built for you to build upon. Notion is designed to look like a utilitarian tool. And at this point, if they're not using Shaden, that would be a huge myth. I would have To go check. But yeah, it's just so you can customize it. And you can see ours are just customized to look a little bit more like Superbase since we're directly integrating in Superbase. But something I want to call out too about the different

products and the different models is that we also do bake offs the same way as we do bake offs with these vibe coding tools. In fact, we did one this morning. If I click right here, this was a bake off with four LLMs with Perplexity and Gemini and Anthropic and Chat GBT because we were literally debating this this exact point that we're talking about right now of what's the most ubiquitous framework stack for AI to build with and it it agrees that it's React, Nex.js, Shad CNN. So all those things we were just talking about,

we were actually researching just to make sure we're not lying when we say that these are like the tools in the modern AI web stack. I can't believe Yeah. And and for for this bake off for example, you were just what you were looking for is do all converge into the same answer and then if not which one do I agree do I think that has the best reasoning basically is that yeah exactly they're all different kind of researching strengths and weaknesses for these tools and if they can come to the same conclusion then I

I kind of believe it right in the age of AI this is better Than going to Yahoo and Google and just searching this is all the LLMs in the world agree with me so I'm right. Yeah, that's exactly how he said it in Slack, too. If you have three out of the four, you're good. If you have four out of the four, then you deserve a monument. And yeah, so this is the pro the process all the way up to here. And then this gets into this gets into actually building, right? So like now of

there's a customer there's a customer validation part of It, too. So it sounds like is the customer validation. So basically like you bake off, you pick the one that you like. The customer validation seems like maybe you talk to a couple of customers, get feedback. Are you doing the customer validation with your favorite prototype? Basically, at that point, let's say you liked Vzero best, you just go with that one. Yep. Exactly. That's the easiest part is just pick one that is close enough and talk through anything that is Not necessarily there that you need need

to get the customers to kind of understand or just to get a signal. Most of the time they're close enough that or they don't deviate enough for it to really matter. But yeah, it's like we I think we did on that first day we did three conversations and we kept that pace up till we got to about 10. Yeah. And just to show you this is the prototype, right? And it like makes it look like you're really connecting the Superbase and it makes it looks like it re it's really scanning superbase to understand the database

and this is the kind of prototype we put in front of people. So it's not perfect like Andyy's saying, but it's good enough because you can go through all this and you end up with what looks like a reasonable product in the end. So that's way more than we need. Honestly, I think this is even more than we need to get enough signal from users that we're going in The right direction. And a small question there. Are you just sending like the link of the prototype or are you like having them tell you what to

do looking at your screen? No, I'm a really big fan of letting users drive and if they're not the actual ones touching and feeling it and I tend to trust it a little bit less. So we always send the link ahead. We have them share their screen and then we generally record the call. Everything kind of funnels back Into dovetail which we transcribe and clip if we need to and that makes it easier for us to share with investors, advisors and other folks that need to revisit as well. Nice. Yeah. So that's that phase but

we don't The reason I was going back to it is we don't also just stop there. That's just the big wide first version but then we do it per feature in the same exact way that we're describing. So this was going back and doing the same kind of bake off, but we Were building a Superbase report card. Basically, it's like an assessment of the health of your Superbase install. And this was just to show you the length of that prompt. So they do get smaller. It's still a mega prompt. Maybe this is like a pages

micro mega. But if you give it a good prompt like this, and notice what I'm saying, too, in my language. If I zoom in a little bit, it's like for our next award-winning beautiful page. So, I foster the sense of encouragement That they're the best designer in the planet. And I would love to be proven wrong, but I swear that always gives me better results. And because it says, I'll create a beautiful superb base report card. And it talks about the design elements. And this is pretty crazy. This is what the report card prototype looked

like on the left whenever we got the result back. And the functional version that's in our product right now is the one on the right. Wow. So, yeah. And for those listening, yeah, there's two screenshots side by side and spoiler alert, they're not that different from each other. Maybe the one that Vzero created is a little bit less colorful. It's a little bit more Yeah. But the other one is more minimalistic, but the key components are basically there. And again, this is the reason why we like these tools that everything uses these days because it's

using shaden to do the UI. So while we don't copy and Paste the functional part because we don't need that of course we can do that in cursor we actually give that code to cursor to say here's a rough layout that you can be inspired by and it's shad CN so it just pops right in and just adjust things where it needs to. So yeah maybe to recap so it sounds so we started with first creating like all the basically the huge mega requirements then that kind of becomes like a prompt and you do the

bake off with with different tools You pick the favorite prototype you do some customer validation you're probably getting both like high level but also like detailed feedback about different features and then at that point you start to feel comfortable okay let's spend more energy and break this down into smaller pieces and that's when you start doing building at the feature level, but you're it's almost kind of like a fractile or recursive in a way, but then you're doing like the same Process with the feature too. And I'm guessing you also then if needed, we'll do

customer validation on that exact feature as well, depending on who gave the feedback. Exactly. Right. And those each one of those feature prompts, I would call them, is I this is a good representation of the detail and size. And of course, we've got all of our cursor rules and everything set up. So anybody that that isn't familiar with tools like VS Code And Cursor and Windsurf and actually almost every AI tool now you can set context for the AI which is kind of like rules and guard rails. And so even though my prompt is just

text here behind the scenes there's a ton of technical prompts that are going on behind the scenes that are telling it this is how you use Shad CNN. This is how you connect to Superbase. This is how I want things to come out that I wrote myself like a custom rules. So it It's not as easy as it seems just by glancing at a prompt, but if you set it up right, then moving forward, it's really just talking about feature by feature. And Andy and I with our history, it's really easy for a head of

design and head of product to talk about this in Slack and then go and immediately build a new feature in a couple of minutes. And it's a good way to just so people can understand uh rules and Cursor. And that's my understanding, feel free to correct me, is that you can almost think of rules as a way to if you've used CHP projects or cloud projects, right? There's two key components in a project, right? There's the knowledge base of the project and then there's like the system prompt. So rules are basically like explaining which files

are maybe like system prompts, which files are knowledge based and then how to reference them across Your repository. Basically, that's exactly right. Yeah. And I'll give you a couple of good examples that are actually good resources. So if you use cursor, there's cursor.directory and there's a whole rules area. So depending on the language you're using, you can go drill down and you can find a detailed set of markdown rules for your thing, whatever that is, the programming language, the package you're using, the node module. I always Have some of these. But now people are going

further with that in an even more cool way. Superbase. And this is another reason why we love Superbase because they're so AI forward. And I'm going to connect some dots for you. Superbase is using shad cn's registry now. So within one command I can do npx shad cn and this this editor rules. And in the in your cursor rules folder it installs all these files which are custom rules for how cursor is going to use Superbase. And you can use this paradigm for anything right? Everybody can have their rules for their thing and then you

can make it easy for people to install these rules on the fly. So this is and it's self-perpetuating. Yeah. And I think that the cool part about this is if someone if an engineer was building this from scratch or having to set this up, this probably would have taken like weeks of work, right? Or maybe at least like days, right now all Of a sudden you're getting all this kind of power and functionality for free basically. And yeah, and it's standard. That's the biggest part. So if I click on the passwordbased off, this is probably

the most valuable one they have. And of course for Superbase they want people to do this. This is so you can log into Superbase. And when you want you run this one command you get all these files that are set up now to authenticate into Superbase with a Password and a login button and a sign up form. And it's using Shaden for the UI. So everybody that uses this now is using the same thing. We can rest assured if AI is building it now, it's not going to hallucinate as we used to see whenever lovable

and bolt first added Superbase. If you looked under the hood, it was terrifying. It was the mo the worst authentication code you've ever seen in your life. But now those developers can just point it to this and It's going to be direct from the source of truth from Superbase how they recommend to authent. And so again, the entire industry can embrace this kind of a standard. So everything getting injected by AI is standardized. Nice. And just to be clear because I think maybe while one I'm trying to put my self in the head of the

audience be like okay but if everyone's using the same libraries and the same standards Then aren't we just going to end up in a world where every app looks the same and from my understanding of how these things work correct me if I'm wrong like it's more I think it's more at the abstract level but I still like at the final touch you still have a lot of power on how you customize the color paddle like maybe how rounded the corners are like there's still you have a still a good amount of customization that you

can do on top of it. Is that a Good way to think about it? Yeah, you nailed it and it's funny. I just had to smile when you made the comment because people do say that but the funny thing is people have said that since the advent of the internet every time something came out and I feel like it really started for me when Bootstrap CSS came out but everybody was like everybody's going to look like Bootstrap now. We're all using the same things. we all look identical and they were Ignoring how convenient it was.

The fact that it was open source, the fact that it was secure because the world was testing the code and it was accessible and all this stuff and in the end it's just a starting point like you said go customize it and I think that's why people love Shaden because it's not like you can't touch these things. This is your login form code and this is the HTML for that login form. So just go and change what it looks like here or change Your Tailwind CSS. But you probably don't want to mess too much with

things like the Superbase client, like the middleware and how it connects to Superbase with the client code. This is the stuff that's hard to set up if you don't know what you're doing. And this breaks your entire site and and it makes it insecure if you do it the wrong way. So this is the most powerful part, customizing HTML and CSS. Sorry friends, if you can't be unique with that, then You're never going to be unique. Yeah. Like I think the connection point to me is like yeah if I was trying to describe a UI

in the car like you did like it would have looked very different to the thing you had. And I think that's going back to you do have to have that taste. You have to have that previous experience. You need to know what's possible and what's not possible. You need to have that almost like roex of like patterns that you that you're aware Of. Otherwise you're just letting AI choose for you. And then at that point, sure, you might end up in a place that is more similar to what other apps are doing. And so it's

almost like an aggregation of all the different UI components on the internet, which makes a lot of sense. Sweet. Should we go back to the process? Yeah. And we're near the end of it. And it never ends really. But uh this this is a fun one. This is if you looked at what any of those other Prototypes did for the landing page before you log in, whether it be bold, vo, lovable, it was all very vanilla like you see on the left here. And basically, I created this mega prompt that even has code and all

kinds of HTML and basically just said, "All right, this is what I want for my landing page." page and then I described like the entire landing page with a live product demo area with the 50/50 split where I have a UI that I can interact With and then I even had some prototype code that I said you don't have to use it but here's some starter code if you want to get inspiration so I gave it code to look at and then look how I this is fun how I talk to it sometimes but wait

and why do you a couple questions here one is like I know you for example like did a wireframe when you were building like the original app why did you choose not to wireframe frame here in the landing page is that because you Already have a vision for it and then the second one is like how did you I guess how did you generate this code and yeah like why and why share it I'm curious like what's what's the rational there yeah I typed the first part and then I autocomp completed with tab like the

second part of how I wanted like the actual layout to be and then I did have some other quick prototypes I did with VO and lovable to to get a little bit of HTML and it was just because honestly it Was a combination of laziness and shortcomings in wireframes. I couldn't quite do in a wireframe what I wanted it to do. It was a little bit more interactive. And so I skipped it and just went to it and said, "We already have a good base here. We already it had already done the report card, right?

And it had already done the other parts of my app." So it understood what my app should look like and feel like. So at this point, I didn't feel like the need To start all over. I could just say, "Look, these are the things I want. This is some code I'm thinking about." And then I this is where set it and forget it, right? This is the most fun part about AI. And with the tools that just came out yesterday and this week, NI codeex and Google jewels, I think Jules is the name of it.

Jules, you can now literally set it and forget it and it'll open a pull request to your GitHub with this. But so, same kind of thinking. I Said, "Okay, I trust you. Go. Let's go win some awards together and I'm going to go to lunch and work out and I'm going to come back later and it's going to be great." And it didn't get far off from the next thing which is our actual landing page. If you go to dreambase.ai, it built that pretty and that was in in cursor or the prompt. Okay. And because

our landing page actually lives in our app. It's not a separate website. So this is just a homepage of our app. But It built it was hard to describe clicking on these tabs and having interactive UIs and things like that. and then a bento grid that was actually animating with these things. So that's why it was easier at that point just to do it in code. Wait, can you go to the all the way back to the top? And you have cool animation too, right? That's is that it that top animation is just a movie

that was generated with a midjourney image and runway for the Animation. So you you did that somewhere else. I mean basically you imported it into Yeah. Yeah. And then everything else is just actual UI code in React and Chaten. So but that's why it's a different process for everything I guess is the way to say or a different way to do everything. And as you we just mentioned two more tools, right? So this is where I encourage everybody to keep trying every new tool that comes out if they Can. It's it can be exhausting, but

it can also be like Christmas every morning, right? There's a new tool to go test out and then maybe it's great. Maybe it's better than what you were using previously. Yeah. Yeah. What's interesting here too is that a lot of people start with the landing page and you guys finished with the landing page. You're not done with the product, but almost like it seems where you it sounds like you almost didn't really spend any Time thinking about the landing page until you had all the features fleshed out. And then once you had all the fees

fleshed out, okay, now I think we can do a good job at building that landing page. Yeah, I'm glad you highlighted this because I think as Kyle and I were really zeroing in and again we talked about at the top we basically built three full-fledged products in the course of 9 months which is crazy. This exercise is really fun for us because we Realized oh wow we didn't actually start on this until March 25th. So we're actually not even two months into this yet. Okay, I did my math right. But we focused on the third

product or dream base and really wanted to nail that onboarding experience. We the first two products were like really great fullfeatured people loved them but I think a lot of the struggle early on or where people were trying to like gro what it could do was just that in that Onboarding phase and we're really aiming to be more productled and get something that people can jump into they can get value out of really quickly. And so this time we really focused on that onboarding experience assuming that they landed on the marketing page, they're excited, they're

going to jump in and then we handled kind of everything downstream, which is also why we waited. Now that we know exactly what that experience looks like, we can cater that Top offunnel experience a little bit more. And so I would say the marketing site is like pretty close to where we'll officially launch with, but I think we'll also probably tweak some dials and knobs based on everything that we've learned through this early customer onboarding and discovery process. Yeah. I think what's cool about that too is that you also then you know what at that

point you've done probably like four, five, six rounds of customer validation, Customer discovery. So you know what messaging resonates, what doesn't, right? And I think that also really helps to with the copy and maybe even the order of the features that you show, right? Like you be okay, like this seems to be like the killer use case. So maybe we'll put it on the first row or the second row. Maybe to wrap it off. I'm curious then like we've almost like now covered like the 0ero to one a very unique zero to one, right? From

the Lovable database all the way to this beautiful marketing copy that's ready to go. What about when you need to make a change in one of the existing features? How is that process? Is it still you do a diff between the original like feature requests and the changes? Like how does that work? Yeah, it can be fun and it's definitely nonlinear. But if I pull up like let's see this is VO and you can see the very first mega prompt here in the left long one. And it wasn't just That prompt by the way. I

also put context into the pro the project knowledge base like you mentioned earlier but this is still one one big thing and then it thought and then it came back. So I'll just come back and pick up either in this conversation or a thing that I want to highlight with our product with this product and most AI products is the beauty of setting the product knowledge and having the history is that you don't have to stay in you're Not locked in this chat. So sometimes I will jump in a new chat and just start clean

and get a new feature. Other times I'll iterate on an existing UI by diving back into a previous chat. So like a good example is this report card. I came back to this dashboard that didn't have the report card and I described this part of it and it built out this feature kind of standalone. So that's a new feature that I just prototype real quick. I show it to Andy. We argue for a Couple hours. No, I'm joking. It's a fun co-founder relationship. No, we have a healthy debate back and forth about is this valuable

for users? Is this feasible with our database and our features and really quickly again Andy is out there talking to people anyway every day. So he can put it in front of people. He'll pull me into those calls or send me the recordings. We'll make a couple of adjustments and go right back into cursor. So it gets the pogo stick That we mentioned earlier starts like the drunken walk of startups. Like it starts out really wide really and it goes back and forth, but then it's narrow now, but now we're just king ponging back and

forth between our bake offs and diving in the cursor and getting feedback. Yeah. And it, correct me if I'm wrong here, but it sounds like some people really obsess about prototyping with existing codebase and making sure that You're using like the like latest up-to-date like version of whatever you have in production. And it sounds like you're not obsessing as much over that as long as like the prototype has all the key components because it sounds like the thing that you built in cursor looks slightly different to what you have here, but you still think, hey,

like it's just better to show them the vzero because we already we can add the UI faster and then we can almost Translate it into the actual feel and use once we have confidence that we're in the right direction. Is that more or less how you think about it? Yeah, we don't need pixel perfect. Yeah, we we're good with honestly any any fidelity or sometimes no visual at all, right? Like sometimes it doesn't you actually bias everybody knows this that's in the product world you can bias people by showing them too much. So sometimes you

just want to say would it be useful at All to have an assessment of your superbase. Is that even helpful? Because I fell in love with this report card because who wouldn't? It's beautiful. and I started pushing it pretty hard with Andy and the users and actually we changed it to what I just showed you when you and Andy were chatting. The latest onboarding now isn't the report card. It's actually more of this scanning the superbase and understanding superbase and building up to a mini Report card if you will and then give them a jumping

off point for our product which is to build one of these reports. So that report card that I showed you is still exists in a product, but it's a thing you do later. And we got that from really quick customer validation and feedback. So I wanted this thing to be the face of the product, but it didn't actually make sense. It's actually the end of the product experience, not the beginning. And this is something to be Fair, Andy and I, you know, I'm pretty old now. I've been doing this since the early 2000s and Andy's

a little bit younger. So it's not like we're inexperienced new people that haven't worked in organizations building products before. So we just zoom past that part sometimes and it feels like we we're really doing a ton of validation. Yeah. Yeah. It sounds like you guys are building stuff are very quick at getting with customers on the call which it also Sounds like that's a good thing when you have people that are willing to get on on a call. Like I think that's always like a good sign that you're solving a problem worth solving. And maybe

I know we're coming out of a time, but maybe one one last question for you guys is I think the way I think about you too is that you're almost kind of like both like a black belt when it comes to using these tools and not only just when it comes to using these tools, But also like how you work together and how you stitch them together and how you bake it into a product process that keeps the customer front and center. And I'm sure some people are like, "Wow, that sounds awesome, but I feel

a bit overwhelmed. There's so much to it and so much depth. So what would be your advice of what is maybe one thing that maybe like a founding team or a PM that's wants to start building like a side hustle they can start doing right Now that they will start to see the matrix the way you maybe you two did when you got started. I'm happy to share mine because it really did when it when I talk about it there's two very pivotal moments. The first one was when cloud artifacts came out. It just melted

my brain cuz for the first time as a non-technical PM I could talk to a computer and it could generate code and then visualize the code and I could say yep that worked. It didn't hear the Things to change and it was just me in the computer and that was awesome. I like that was the moment in the coffee shop where I turned the laptop to Kyle and I was like holy crap did I do it and this is how we're this is what we're focusing on as we start the company. the second time and

I think is the phase that we're all in and excited about. This is what's fun about the super slack. We have like the AI workflows, everybody's vibe coding, but was really Lovable like they they nailed it. They made it really easy to jump in, talk about a thing that you want to bring to life and it shows up on the screen. And then they iterated, they added Superbase, you can have a full backend, you can add off, you can like you can really go zero to microsass in a few hours or days. And so I

think all of that to say get started, jump in, start using it. We we do a lot of product meetups in Austin. I run the Austin Lenny's meetup and every often people are like, "How do I get started in vibe coding?" I'm like, "That's the beautiful part. You just start. Like all the tools are free to get started. You're going to get hooked. You're going to spend at least $20 a month, if not more. But start working at it. It is a deluge of information and there's an onslaught of new tools and everybody's going to

be talking about all these things that you need to focus on and I would really just Start carving out time even within your regular workday to work on AI tools and just start getting used to the process because it takes a few weeks to get over that learning curve but it's the fastest learning curve that I've been ever been exposed to in 15 years of working in tech. Yeah, that part is so key of if you don't get something going in the first day or second day, give it at least a week and don't give

up because like you know that the difference Between like you know 20% to 80% is so small and once you get there like that's when you unlock most of their gains. Yeah. What about you K? What would you say? It's funny. I love hearing Andy's story just because honestly it inspired it inspired all the products that we built, all the versions so far because that's the kind of user we want to target. What we describe to people is if you can articulate the problem, you can take over the world now. It's no longer And actually

we've seen a lot of industry engineers are having a harder time using these tools because they're thinking about it the old way. So a lot of it's trying to rethink things in a new way. And I was telling Andy yesterday, we have to do it ourselves. We've got such a good process that I just showed you now that we're proud of it, but we should throw all that out tomorrow and try something new maybe because things like codeex came out and Now I could be on my phone opening a pull request by talking, you know,

telling it what code I want. So, you got to just I think we have to embrace the fact that it's always going to be moving forward now very quickly and there's no set patterns. On the other side, it's try to break the thing. Like I always, again, I always toot his horn on this. Andy's like the best quality engineer on the planet. I get it bulletproof and I ship it to the dev and then he it breaks In two seconds. But using tools like chat GPT for prototyping or one of the things I I did

with Chad GPT recently was take a high fidelity image and get it to convert it back into a wireframe which is something that I've historically had a hard time doing with software and sometimes you want the low fidelity version but if you work in code it's hard to get the low fidelity version. So think about new ways to use things. I would encourage everybody Listening to this to go get in uh cursor or windsurf or VS Code and try to design things in a markdown doc and design it with an AI agent, but also don't

get stuck in a silo. A big feature that we worked on forever was making Dreambased multiplayer. And you're going to hear multiplayer AI a lot more these days. Lovable just added teams that can jump into the same chat and talk to AI at the same time. Everybody still has the same skills they had before AI. At least in This generation. I don't know. I can't speak for the next generation. But for now, there's good product people, there's good design people, there's good technical people. We need to be working on the thing together or we're just

going to show our worst part of ourselves better and faster with AI, right? So, I think the collaboration factor is something that it's harder and we have to make an effort to do to collaborate. Amazing. This is so much fun. I feel like I personally learned a ton and I'm so inspired by the way you guys work and how you push these tools and I'm just excited to go back and do the same. I feel like I've been using cursor in a very different way and I'm excited to start using it more for markdown and

for prompting. I feel like there's just so much potential there after seeing how you guys work. And one final question for you guys is where can people learn More about what you guys are building and about you too? Also, is there anything we can do to be helpful? Love that. I think both Kyle and I are very easy to find both on LinkedIn and Twitter. Kyle is Kyle Led Better and I'm always Sunday. Always Sunday but 1s. And then I think we're I mean we really just love the community aspect of this. It's been a

ton of fun to see what people are vibe coding and shipping. So if you're vibe coding, we want to hear from you. We want to look at what you're building. Bonus points if you're a founder and or you're building on Subabase. that's perfectly in our wheelhouse. But I think we're also really interested in more mature product teams with kind of extens extensive data, extensive data sets, and excited about AI, but maybe don't know how to put it to work yet. We're happy to show and help identify some easy starting points. Amazing. Yeah. I've worked my

entire Career in analytics and big data. And saying, yeah, of course, we're embracing the vibe coders, right? that and we can educate them and help them a ton. But we're actually really interested in in in established products because we can do instantaneous product analytics without having to ship off to some click stream and do all kinds of third party songs and dances. If you just want quick answers, that's what we're focusing on. And also, we're getting people to think About data a different way. Maybe you can use a product like Suabase as your data warehouse.

We're getting we're getting people to do that and move more data into places like that to where they can experiment with it. So, we're trying to think a little bit downstream of just get the vibe coders. Again, I'm trying to think of a fun term. What comes after the vibes, right? It's if jazz was before, what's after? I don't know. I don't think we know yet. I think vibe Was like it's at the extreme, but awesome. And dreambase.ai. I don't think I said that. Dreambase.ai. And you guys are already launched or you're about to launch.

We are launched in the sense that the product is fully functioning, fully usable. I would call it a little more early access. We'll we're teeing up a bigger splashier announcement, but if you want to get in, you know where to find us. We're happy to get your take and feedback. Amazing. That's a wrap. I appreciate you both. This is a ton of fun. Thank you both.