Founder’s Guide to AI: Building an MVP with No Coding Skills

Hello everybody. Appreciate you all making time to come out and talk today. We're going to be talking specifically about building an MVP with no coding skills. I'm going to hop right into it because I think the best way to begin is by showing you what, what I'd recommend you do. Then we'll talk a bit about the reasoning and what's going on behind it. So I was thinking about what a great product would be for us to build and kind of came up with I have an idea for something called The Monkey's Paw. Now, if you are a— if you went to school in the United States and learned about, you know, early 20th century fiction, that might mean something to you. Otherwise, don't worry about it. But I have pulled together in a little notepad some ideas for what this product is. The basic idea I had was, uh, when you ask AI to do something, very frequently it does something kind of like what you asked it to do, but in the end, not really. And this is going to analyze the prompt that you give AI and look for the ways that your prompt might go awry so that you have a feedback loop before you spend money and credits on having an agentic AI do something. So I put together very simple text here. I decided, because The Monkey's Paw is in the vein of Edgar Allan Poe and like authors from the early 1900s who were kind of horror and scary stuff, make this a kind of a dark Victorian-looking web design, which is definitely different. And then for the purposes of security and just making my infrastructure easier, we are going to store the analysis that we have created locally in your browser's cache. So I don't have to worry about cloud architecture. So like, there's an idea for a product and I can hit and run this. Now, when you run Lovable Year, which is the tool I'm using, it takes it a while to parse through what you're asking for and begin to tackle it. I want to leave it running in this half of the screen and you'll be able to see The AI is building some stuff out. I'm going to talk to you a bit about what it is we're going to be building over the next 10 and 80 minutes, and also what our overall agenda is going to be. So the 10-minute landing page, we're going to go into all of this and we're going to watch as this builds and talk about how this works. We're going to discuss how to do product planning, how to make sure that you've got some way to revert when your AI, as it certainly will, ends up going down some crazy side path and ends up either breaking your features or writing a bunch of features that end up being not useful for your app. Then how you deal with the AI itself, the feedback it gives you, and how you understand what your product should develop into. Uh, and finally, I'll give you two case studies of tools I have built recently and what I— how they have developed in Lovable or in other tools. So you can see some actual use cases of somebody who's not a technical founder. I'm a terrible developer, but I built a number of technical products over the last couple of months, enough to validate ideas, see if it's worth investing more in, and begin to scale the startup. So before I go into that, why, why trust the word I say? If you've seen me in FI before, I have been a startup founder and mentor for, I think this is my 8th year. I've mentored something like 72 cohorts of founders over all these years. I have built and launched 15 different startups to more than $3 million a year in revenue. I built and sold a tiny AI startup 2 and a half years ago. Uh, I have a lot of experience at this early stage of I'm beginning to build something. What do I do to make it make money? And I am right now in the midst of, uh, as co-founder launching two different startups as well, because why only have an agency, uh, to kind of keep my hands in, make sure I'm, I'm doing the work as well as kind of in the space. So all of that said, what I put together here, this prompt I gave, right? And you can see the AI jumps right in. It immediately starts to work. There's one problem with this, which is that I just did everything wrong. In terms of how I told the AI to do this, in terms of what expectations I have been setting, none of this is what you should be doing for AI. What you should be doing instead is actually giving your AI enough information to build the correct thing as opposed to whatever the heck it thought it was building. Now, for example, here, hey, it's built this thing, it's done its first little version, and I could say, uh, hey, you know, this, for example, nothing's clickable. So I told it what I wanted and it didn't build any of it. And it is really easy as a first-time person doing development work to go, oh, The AI is stupid. I told it clearly what I wanted. It didn't deliver it to me, but you didn't really tell it clearly. The very first thing you need to do is come in with more than just a, ah, you know, here it's gonna, it's gonna prompt to, to make your AI outputs better, right? You need to come here with a relatively fully formed idea for what it is you want your product to be. Now, the good news is you don't have to develop that idea entirely by yourself. And in just a few minutes, I'm going to talk about how you can actually use AI to help you build out specifications here. You had no design documentation, so you can't really have AI build an entire application with one prompt. I know there's softwares online that promise you that they can do this. They are basically all lying because so much of how this actually should work is not like it. I haven't expressed it at all and the AI can make guesses, but there's no way the AI is going to know what I really meant. So you need to have a design doc that lays out everything from start to finish. And I'll talk about how you make that. Then, as we're about to see when I go begin to make some changes in here, you need version control. You need a way that after I've gotten a functioning app, if I do things to it and they break, I can undo, right? Everybody here on this call has used Microsoft Word. You probably used PowerPoint. I mean, heck, this is Google Presentation here, right? If I go, oh, I deleted that, oh no, I want to be able to undo it. Having that version control to go back to where things were in the previous version is really helpful when AI goes crazy and takes you places you didn't want to go. And then finally, I don't have an end goal for this app to actually, like, survive. And anybody who's in this call, you're not just building a technological thing because you want to build a cool piece of tech. You should be in this call because you want to build a minimum viable product for a business that's going to succeed and scale. And so talking about how you make sure you've got all of these pieces put in there is what we are going to be covering over the next 80-odd minutes. But first, before I jump into the specifics here and all of these things, to ease all your minds and to make you not have to worry too much about note-taking while we're talking, I'll just do this real quick. Um, you can go to this URL, crowdtamer.com/ai4founders, or scan the QR code here. It'll take you to a page where every prompt that I am going to talk through over the next 70-odd minutes is already saved and available for you, so you don't have to stress about taking notes in an all-fired hurry. Also, if you were in the last workshop, or not in the last workshop, all the prompts from that workshop as well are going to be in this email. And next week, When you go to the same place, it'll be all 3 weeks worth of workshops and so on. So with this information kind of handed to you up front, let's talk about what you need to do. And the first thing you really need to do is you want to use AI to plan your AI, which sounds very recursive. But if you were here, the last workshop, one of the things I talked about was human in the loop. Is crucial to have AI provide a result that you actually want to see. And so one way to make yourself the human in the loop is to have the AI ask you questions and you deliver what it is that it needs to get some answers here. So here is, and as I mentioned, you're going to get this PDF if you go to the URL I already gave you. Here's a prompt, and I can say I want to ask this to any of the chat models out there. I am going to, throughout the entirety of this presentation, going to be focusing primarily on Claude for two reasons. One is, with a little bit of work, which I'm going to tell you how to do today, a little bit of work Here on Claude desktop app, it becomes your technical co-founder. It writes code for you when Lovable can't. It works on backend code for you. It can create all kinds of crazy integrations and automations for you. So Claude desktop is a really, really powerful tool that helps you step through the technological parts of your business without having to worry about having a tech co-founder upfront. Eventually you will still need a real developer, a real technological mind to help you, but you can get surprisingly far where you are. So I've taken the prompt, I've pasted it in here. The first thing the AI is going to do is come back and ask me some questions because I've told it to do so. And it wants to know about the fart idea. So I'm just going to answer this in order here. I'm going to grab the prompt actually for this. There we go. That's just the first question. I've got an idea for something called the monkey's paw. Now, from here, who experiences this problem most acutely? Now, if we're in Cloud Desktop, do hit Shift+Enter and not just Enter, or it automatically executes. Offer number 2, it is non-technical founders. Man, I'm great at typing. Founders who are trying to build their MVP. And see, what are people currently doing to solve this problem? They take a lot of time and burn A lot of AI credits on bad prompts. I would note my typing doesn't have to be perfect here. AI will figure out what you meant. For what makes this existing solution inadequate, if you want to be either energy or cash efficient, you can't just hope that your prompts do the job. Do the job. And then see, what's the fifth one here? How urgent is this for a non-tech founder? Uh, this is mission critical, blocking them from creating the product that they are dreaming of. So there's our first set. Now, there's a lot of questions for this, for back and forth, because a product design doc, which is the first thing we're working on here, needs to have a few pieces to it. The first part, what I just laid out, is the vision, right? What is it that you are intending to do with this product? And now you notice in the background, while you and I are talking, or I guess while I'm monologuing, The AI is doing work in here, and this is rather deliberate for how I'm conducting this entire workshop because the AI takes some time to get back to you. And when you're working on— pick any product you can imagine— you're going to tell the AI to do something, and then it will go at a certain number of words per second and reply to you. So you can very easily multitask either on the same product or on different things if you can move your attention from one thing to the other. Right now there's a bit of overwhelm. I have Lovable, I have Claude Desktop, and I have my presentation, but we're going to try and not make it much more complex than that for right now. We're going to step through and finish making our design doc here. Okay, so one of the most common ways prompts go wrong, um, AI will usually do what you tell it to not what you meant. It often makes assumptions about what the user wanted that are not true. There's one. Most AI products issue a number of credits for use. Every day or month, and using them on bad outputs is a waste. At the age where, say, as tech products get more complex, more, more complex. There is a need to be more precise in the, in the prompts to work on that. Okay. And then can you walk me through a specific example? Now, how technical are these founders? Typically, 5 is, I think we can know if we're here, right? 5, not very technical because they can't write their code themselves. And then this is 6, is the current process is keep trying until it works or until All of your credits are used up. There we go. Okay. So we're moving now from vision into creating some execution here, right? And going into the product vision and goals. Now, I don't think you want to watch me slowly type back and forth for the next 5 or 10 minutes. Going through this, but what this is going to do is continue to interrogate me until it is able to create something that looks like a fairly detailed product requirements document. And from that, we will get a— give me a second to open this up. I have in my services here somewhere, my design doc. There we go. In docs, I have my PRD. So as an example here, this is a fully built out PRD. You've got my product name, and then we start to talk about Right? Vision, execution, and success. So vision is going to talk about who you're solving this for, what their problems are, and then like what a solution looks like. This is all expressed in plain language, right? I'm not using any technical language yet. I'm just, this is the output of a back and forth with Claude AI in a previous discussion. To lay out what these pieces are. Now, this is fundamentally a pretty silly product, right? I don't think in truth this is something I could monetize. That's not the goal here. This is rather a small product that we can talk through as an example and go from beginning to end during a live call right now. But we lay out features and then also future features, which is kind of a combination of vision and execution. And then we're going to talk a little bit about what we want to see from the user on the technology side. And then roadmap here, I would note that the one, this is clearly the context window for this AI is last year, that's saying 2024, but also this is something we can reliably like We could easily bust the whole thing out from start to finish in probably 2 days of work. So the estimation here is wildly off and don't worry about it. Okay. Um, with this laid out, I've now given the AI a source document that tells it my intent. And you don't have to paste this in to whatever app you're using to build things. So like, you don't have to paste this into, uh, Lovable. What you can do is either— and we'll talk about once we have this in a more directly editable format in a few moments— if you have this directly editable, you could just upload this into the folder, synchronize this with Lovable, and then Lovable will find the file. The other way is you can say, hey, here is the text of my PRD, please save it. Somewhere, which I will actually do here now. So I go back to Lovable in my browser window here, and I'm going to say, I have created this PRD, please save it in the docs folder in this project. And then paste it in. There we go. Now it has the PRD so that it knows what to be building towards. Now the PRD by itself lays out kind of your vision. It's what you want to see happen. There's a second thing you need the AI to help you make, and that is how it's going to work. You need technical design doc. Now that can be really tricky if you are not yourself technical. But once again here, you're going to use AI to tell you how to use AI. When you're building in these loops where AI is giving you the information it needs to tell it more, the benefit is I don't have to be an expert in this technology to be able to give the AI the information it needs. The downside is Every AI does its own little wonky thing. So something else you can do if you want to try to improve upon the output of one AI is go over to another one. You can pick almost any of the other ones out there. I would probably avoid like Grok and or Deepseek as they are, I think, the least advanced reasoning models right now. ChatGPT, Gemini. If you're using Gemini, try Claude, there's a number of other AI models you can go to. And you can just say, I have made this PRD for a product I am developing. Can you please improve it so that my AI developer will do a better job of delivering a successful product. Then paste this in and let it go. Now here, it's going to always say, hey, I've got some improvements. Build this out some. And so now I'm using a different model's different weights to improve this further. You can go back and forth a number of times and just ask, is there anything to add or is it good to go? And you'll reach a point where the AI eventually says, there's nothing more to add. And then you'll go, awesome, this has everything I need. But the next thing is, Having generated this PRD, and now this AI has in its context the PRD that was just generated, I can also say, I need you to make me a technical document and generate out of this how to make the existential bits of this thing work from a technology point of view. And again, this prompt, along with the other ones, will be available if you go to crowdtamer.com/ai-for-founders. And based on this PRD, please, um, help me create a TDD via the following steps. And this now has laid out for the AI the questions it is going to ask me to make an effective technical design document. Now, what you need to have there, and while it's thinking, we will review these, is a capacity to lay out the system design. Now, system design is talking about how all the big pieces fit together. Code organization is going to lay out the architecture of where each type of code sits. The data flow, it might be a flowchart, it might just be language to describe how stuff moves around. API and infrastructure, any of these projects are going to have to have sitting on a server somewhere, software installed. API that you connect to, information that you're using to make your data flow, and then finally security, how we make sure that you haven't vibecoded something that someone's going to hack and immediately spend $6,000 a day of your API credits because they happen to find it, right? Now again, all of these things aren't things you need to be expert in because the AI is going to be able to tell you what the outcome should be. I already mentioned— there we go— you're going to get a simple version of flowcharts and how all this works. And in this case, the AI hasn't asked any questions about the product. Now, I think it probably could. And you could even ask it, hey, are you sure you know everything? Are you sure? Are you sure you know everything? Um, are there any questions you need to clarify? AR, on to make a better TDD. Just to see if we can't provoke it to ask some questions. But again, I've told it explicitly in this prompt, you're non-technical. It's not going to be asking you things like, do you want to use Node or do you want to use React? It has to have opinions. But the features I've had this talk about, there's some questions about how the app itself will work. A implementation question that changes the technology, not asking you how to secure the API key, not asking you, you know, what encryption methods you want to use. But if you're telling people you need to give me an API key to make this app work, then presumably you should know what that means and you should be able to answer these questions. Likewise, what do you want to happen? When they click this button is a great question to analyze what the technical responses are going to be. Uh, and then finally, do we want to add any more features here? Right? So these sorts of questions, you answer it, it will inform the technical specs. And I could go through and answer all of these, but that would make for a pretty boring presentation. I'm sure you can see how if you were the founder, this is kind of challenging you, right? This is asking you some questions that you maybe aren't fully able to answer. And one of the benefits to using an AI like this is if you go, hey, I know the answers to 1, 2, 4, and 5. But I don't know the answers to 3, right? 3 is, is a question I haven't considered yet. You can either say, get back to me, like, I can answer these, make your progress, I'll get back to you tomorrow. You can just wait. There's, there's nobody here to lose their temper. Right? The pressure around if you're working with a technical co-founder who goes, I need to know this before I can build, it's not there. So with what you've got here, you've pulled together the foundational documents, right? Your product requirements, your technical design doc. Now we go into Claude Desktop and we start to work on how we use Claude with Lovable. I actually see Frederick asked a question specifically about this. I think that's, I'll be getting to it in a moment and explaining why. But as we're building out any of our responses, right, I like Claude and Gemini for the back and forth where you answer questions in one, you answer questions in the other. And then you can start to meld the two together yourself and say, hey, here's what Gemini asked me over to Claude. I've answered them these, these ways. Please update the PRD or the TDD. With this more information. With this, what you're getting to is after 3 or 4 passes around, an idea that was in your head that was kind of vague, by being asked all these specific questions, you've been forced to refine it into a more concrete product. This is the same thing any developer does when they start to tackle the problem, but most non-technical people don't really grapple with this because they go, I want an e-commerce website to sell something through. It's someone else's problem to fix it. The biggest unlock that AI makes for you as you are a founder building your MVP kind of by yourself is it forces you to start to think in that product-minded way and create a better overall vision of the product for yourself. And a better overall vision of your product as you build it out. So, but before we go into why you use Claude and why you use Lovable, I have to introduce something that's kind of scary. And the scary thing is GitHub. Most anybody who knows anything about GitHub just knows it's the place where developers do their work. And it uses a bunch of terminology that doesn't make any freaking sense. And why would I, the non-tech founder, ever use it? Well, two things here. The first thing is Claude Desktop is your tech co-founder right now. Claude Desktop is going to, once you install the right software for it, going to be able to take care of a lot of the scarier parts of Git management for you. I would note, and I will preface any of my discussions about how I use Claude Desktop, I've given Claude Desktop access to read and write files on my computer. I've given Claude Desktop access to read and write and push and pull files to and from GitHub. I do recommend that you do this, but I am also telling you, do this at your own risk. Because you could always accidentally tell Claude desktop with read/write permissions on your hard drive to delete everything and destroy your computer. If you're using Claude desktop to manage files and manage software development on your computer, be careful. Make sure you understand where Claude is reading and writing files and what it is doing before you enable any of these features. If you run this and you unfortunately brick your computer, I am sorry for you. I warned you up front. That's my legal liability notice. Okay, as we talk about GitHub, there's really 5 things you need to know, and that's it. And if you know these 5 things, you never have to worry about GitHub again because you can just tell Claude what to do. The 5 things you need to know are initiate or init a project. You only use this once for any given project, probably. You tell it where to start a project and it's going to basically take a folder and begin to track the changes here. All of us have used documents with track changes before, and you know, if you do a ton of track changes, eventually there are so many red lines everywhere, it's really, really hard to see. What the original text is, what the text you're working in on now. All of this can get overwhelming. And if you look at the raw code outputs in GitHub, you can easily be overwhelmed too. So I tend to recommend for a non-tech founder, don't worry about what the raw code changes are, any of that. Let Claude manage this for you. There's a pull, which means I'm going to go from the cloud and bring it down to my computer, the latest version of the work I'm working on. Then there is a branch, which is when you— I've got code that's working, I want to try something over here. It's basically like making a copy and working on this copy. And then if you completely destroy the copy, you can just delete the branch and the main part still works. You never touch that part. This is a really good idea once you've reached a point where your minimum viable product has at least one feature that someone's using. If you work on a branch, develop things, and then merge the branch back in, this means everything you're doing doesn't touch your working code, which means you decrease the risk of blowing everything up for people. And then finally, you have a push, which is you save all your changes back in the cloud. That's the, the main set of features you're going to go through, and, um, when you are developing, I'm just going to go here real quick. When you are developing a product, you can either use Claude directly to manage all your GitHub Functions. You can work in the command line if you want to manage all your functions. As a non-tech founder, I imagine you won't. There's even tools like desktop browser tools for GitHub where you can manage everything with a graphics user interface. Any way you wanna manage it is up to you, but particularly in the beginning, I recommend you use Claude to manage this. Now, all the prompts I have provided for download mention specifically that the person giving you this prompt is not technical. Please explain everything before you do it. That will help encourage Claude as you ask it to do things to be a little more circumspect about how it works on things. But with all that, we're going to talk about now, if we're using GitHub, we start to be able to work on Lovable and Claude both on the same product. But there's a question, right? I thought we were using Lovable. What the heck? Why do you keep talking about Lovable? Plus Claude. Now there are a few reasons why Lovable by itself isn't enough, although they keep adding new features to make it more powerful all the time. One of them is by default Lovable is using an older model that's not as powerful as some of the more current models. Because of that, sometimes you're going to find yourself where a really complex product it's too big for Lovable to really control. I will give an example of a product I built. We'll talk about more later. This is Insight AI, a case study we'll discuss in a little bit. It has 500,000 characters of code across, I believe, 250 files. And this started as a Lovable project, but as it grew and grew and grew and grew, it gets to where Lovable really struggles to add new features, understand what's going on, and the, the backend has reached a level of complexity where it's just not easy for AI to maintain. There's ways you can manage this with a more controllable tool. Specifically, uh, large products also, just even if it's not very complex, but if you have a very large file that uses up all the context window for your AI, and Claude is very efficient at using its context window, unlike Lovable. And you'll reach points where you get into kind of a development bug hell where you will say, fix this feature, and it will say, I fixed the feature. And then you say, great, but when I tested the feature, it's still broken. And it will go, oh, but I fixed it. And you go around and around. And when you find yourself in that loop is when it is really valuable to try a different AI model because it'll have a different approach. Just like if you're writing a PRD and you drafted one in Claude Desktop, then you go to Gemini and ask it for a second opinion. If you've got a code base that is not working for Lovable, you can go over to Claude and ask Claude to help. But also, if Claude has reached a point where it is spinning out, you can go back to Lovable and say, hey, these specific files are not working, can you think of a way to fix it? And sometimes Lovable will block things for you. Now, Claude by itself is a really powerful tool. But I don't find it does front ends particularly well. Like something like this little front end here, or where's the other one I have running? This one here. Let me pop this open in the sidebar. Right, I rebuilt this app twice because I wanted to do a trial run before this. Basically the same functionality but a little further on where now I need to give it my API key. If you will excuse me, I will do that in a different browser window because I don't want you all to see my API key. The need— there we go— is I have it Save it here. Okay, and paste it in. So, whoops, uh, this, and then it, uh, fortunately anonymizes the key for me, right? So I've got this connected, and now I can send queries in and go in. Uh, this is developed, let's say, a little further on than this mockup is, but getting a mockup designed this way from Claude directly, I find the UIs look a lot more elementary, and as a result You don't get the same experience for your user. And Lovable here is going to take care of a couple of problems that are more challenging to deal with for Claude. One of them is Clovable— Clovable— Lovable has an easy integration to Supabase to store data. Lovable also has an easy integration to GitHub. To synchronize this data. So we're going to connect this to GitHub right now so that I can then take this code and begin to look at it locally. We're going to connect the project to— I'm in the wrong browser window. Sorry, this is the wrong one. I want to go back to this one, connect this one to my GitHub, and we're going to make a new repo for it. To me. I can now talk to Claude and I can say, hey, I'm going to make a new chat real quick. Hey, I want to pull a repo from the GitHub repo. From the cloud to, to my /dev folder to begin local development of it. It is there. Now, this is going to take the code that now sits in GitHub, and every time I make a change, So I'm going to say, hey, make it so I can connect my OpenAI, spell this right, my API key, and use it to analyze, you know, wisely the prompts a user provides. There we go. Now, as this makes changes to the code base, it's going to automatically push them to GitHub, which means this product that I'm cloning right now, unless it takes a few seconds to clone, is going to end up being out of date from what's in the cloud. Because this is now writing software and pushing it. So I would note all of these things that are happening right here is the kind of thing that if you are new at development, this is kind of intro to, uh, version control, but you don't have to understand how it works. All I've done is use the terminology we just discussed, right? Pull this down from the cloud, save it in the dev folder, and you're good to go. And then this over here is now writing and adding new functionality in the background because it's automatically pushing it to GitHub. It's keeping things up to date for me nicely. And if I were storing data in Supabase, this would make it very easy for me to build my entire cloud architecture without having to worry about how do I spin up a Google Cloud server or Amazon web server or Azure server or whatever, whatever, whatever. For like $20 a month to Supabase, and I think $20 a month to Lovable, you can have everything you need to build out a web app that will work. And then pay $20 a month to Claude for Claude Pro, and you've got for $60 a month something that looks a whole lot like a junior developer who does whatever you tell them to. Now on the downside, He does whatever you tell him to, and if you don't tell him the right things, you're going to end up being frustrated. So I've been talking a lot about using Claude Desktop to write this software and to write your code. This is not the tool that most people who are in the development space talk about using. They talk about using a tool called Claude Code, and I'm not going to go into the differences between the two in great depth because I think for most new founders, Claude Desktop is a superior way to approach the problem. There are two reasons why you might want to use Claude Code instead of Claude Desktop if you are even a fairly early stage non-tech founder. One is, as you've already noticed, I have— when I run something here in Claude Desktop, I've only got the one window. It's going to run and do its stuff all on its own. I have no control over it. With Claude Code, I can open 2 or 6 or 10 individual windows with Claude Code and have them all be running at the same time. Also, Claude Code, unlike Claude Desktop, has something called subagents or agents where I can say to Claude Code, hey, I want you to do these 3 things at the same time. So create 3 agents to do it all at once. And that will go much faster. I would note the AI credit consumption is basically the same. So if you're using Claude Code and you're on the Claude Pro $20 a month plan and you try and do 6 windows each, which has 2 agents, like you will use up all your credits in 10 minutes. With Claude Desktop, you have to think linearly instead of kind of in immediate branching logic. But it is pretty hard to use up all of your credits going through this one step at a time. So the repository has been pulled down. It is now stored locally in— where is this? C:/dev. So if I go to my hard drive, there should be dev and This will have the file structure that we just got from Lovable stored locally. There we go. Monkey's Paw Oracle. Oh, it hasn't pulled down yet. No, I guess that means it hadn't copied over yet. Let me check in GitHub. My repositories should have new, this one update. All right, we're just going to grab this. No, please pull the files from GitHub into the local repo. So this will handle— there we go. Probably when it checked initially, it was an empty repo. Now it's set up properly. Okay, so from here, what we've now laid out is the means through which I can iterate back and forth with AI. Either in Lovable to work on my front end and to add new features, or in Claude when Lovable gets stuck, or when I want to add complex new features, or even just I go, hey, um, the prompt we are using, let me find it, uh, this would be the prompt analyzer, and in it, it has basically what's going to get passed off to OpenAI. Which, if I'm trying to make a prompt that will catch every possible weird use case where someone has written a bad prompt for their AI and they aren't going to get a result they want, I will probably need a better prompt than this to guide the results. Now I can edit here, but this is not the most user-friendly place to be working in. I can also edit here in VS Code, or I mean, you could open this thing up in Notepad, like anything you want to take the local file that you have that you wish to analyze. Let me go back to here. This should be updating. Not yet. Still working with Cloud Desktop. You will want to, any piece of software you want, you'll be able to use to add the text here. I would note now it's saying, hey, there are no files in the repo. Now there's a couple of ways I can be like, what's going on? One is just to say like, hey, look at the repo. At the repo, I can see files there. This repo is public, I'm pretty sure. Let me check and make sure. I believe it is. Yes, it's not set to private. It's going to go look at where I've directed it to. And ah, you're right. There's information here. Now, I would note the AI lied to me. I didn't set this up that way, but it's convenient. That's my next slide. I would also note, again, if you've missed the chance, right? Free download all the prompts I'm using. Except for the ones I've just typed live right now. All the pre-planned prompts I'm using are at crowdtamers.com/aiforfounders, or scan this QR code and you'll be able to grab it. And if neither of those work, go to my homepage, crowdtamers.com. There's a big banner on the bottom of the website right now. Click on it. It'll take you to the right page. Okay. So AI lies to you. AI is nice. But AI is not kind. And but what I mean is like, When I, when it was lying to me, it's like, oh, you're right. It'll say, oh, I'm sorry. It will lie to you totally barefaced because it, it's a language model. It doesn't know how to be kind. It just knows how to show the, the, uh, the forms of politeness. So it'll say stuff like, there's no files in this repo. Or it'll say stuff like, if I go all the way back to the first answer here, this is a fascinating concept, right? The monkey's paw. Is this really a fascinating concept? Like, I don't think this is a business deal. I think this is maybe a tool that's fun to work on, but if you read this and you go, oh wow. The AI thinks this is smart, so I must be smart. It does not, for two reasons. One is it doesn't know anything. It's not really intelligent. But the other thing is it is nice, but not— it is kind, but nice, but not kind. But it'll, it'll say nice things to you. As a reminder, here in the very first step, I did everything wrong and it didn't go, hey, I can't build this yet. Instead, it gave its best attempt at building something that is impossible to make because core pieces of the infrastructure have to be set up. So as your AI is working for you, it is important to note, right, whatever it says to you, don't assign an emotional weight to it. Don't assign a belief. That if it says this is a good idea, it is actually a good idea. And also frequently it will take you the— down weird, strange side corner paths to results that are impossible to achieve simply because it also lacks an understanding of your product. And it may try to build architecture that is impossible to have work. And using that fundamental memory of, in your head, that the AI will do its understanding of what you tell it to. It doesn't really have an opinion, and it will tell you stuff that's not true until you correct it. That's stuff that's obviously not true. And now it's like trying to, uh, execute on a task I gave it that in the end is impossible because I haven't set up the tooling that it needs to do it. So previously it was saying, sure, I can do this, sure, this will all work for you. And then just like here when it says I'll build this for you and it can't, here it tries to build it and it can't. Now It tells you how to fix it once you poke it enough to make it see the truth. And that's the last part of being a good non-tech founder using AI as your technical founder is you have to go at this with a very critical mindset because what AI is going to tell you as you develop your products is going to be true sometimes and somewhat true most of the time and occasionally completely wrong. If you interrogate it, that is when you will get it to say things like, oh, you're right, I had that completely wrong, or oh, that's impossible, of course I can't do that. I've had it tell me it can perform features in Claude Desktop that no AI can do, just because it suddenly assumed it had the capacity, right? So getting to the point where you know to trust the AI a little bit, but validate its outputs before you actually execute it is really important. When you are working in product development with AI, you've got this, this tension, right? The AI can go just preposterously, ludicrously fast. It can, uh, and because it can go so fast, you can make the assumption that anything is possible. But there is a necessity to be very specific about what you give it. To get the best outcome. But if you're too specific, then you don't have room for the AI to actually develop some creativity. Now, as a non-tech founder, you are not going to tend to be the person who's going to say, hey, you have to write this in React, you have to use Tailwind, you have to use TypeScript. You don't tend to be forcing these technologies onto the AI. So some of this creativity paradox isn't going to be a problem for you. But if you are technical enough to know these things, and you're using AI as the way to ship faster and better, it is very easy to be mistaken by telling it do this and not the other thing. And then you will force it into an unsuccessful outcome because you put constraints on that the AI can't handle properly. The best way to do this is to work in small pieces, even in a big project, right? Even in something like Insight AI, 500,000 characters across 240 files. In Lovable, let me actually go to the project file real quick. In Lovable, where I've got this project, yes, thank you, go away prompts. I can specify to this, hey, I want you to, uh, focus on making changes to just this file, this file, and this file, and change something about the way they work. This requires that you understand the architecture enough to say, oh, those are the files that this feature sits in. But if you force it into small testable chunks where you don't say something like, you know, hey, for, for this product here, don't say I want to switch out all these cards for, I don't know, a, um, a stacked card methodology where I can automatically scroll through them with my mouse. That would be a really big UI change. And to try and one-shot this would would probably work poorly for you. The first thing you would obviously do is make a branch so that whatever changes you make don't affect the production code, right? Do that in GitHub and tell the AI to do it. But the second thing you'd be doing is you need to say, okay, well, uh, what's the smallest piece of that big thing that I want to ship? How would I make that smallest piece happen? What are the fewest files I can have the AI touch? There's a simple framework to make sure that you are bringing clarity in your, your prompting, to make sure that you are giving something that is concise enough that the AI can manage without being so, so concise that it is unusually small. What you need to do is if you go back to think of your PRD, and in your PRD, if I go open that up again, hang on. In your PRD, this window, let me go back to the right project. Or actually, I've got one here, I can show you this. If I go, in this case I called it CloudMD, which is not the current best practice for naming, but that's what I called it back at the time. There you go, Project Manager, beg your pardon. This has like how this is laid out. If I'm going to ask the AI to do most anything. Once I know my design doc is built out, I'm going to ask it to touch one or two pieces at a time and no more, because that way I'm limiting what it's doing and I'm telling it specifically I want to work on this feature, so do this and this and this to it. You don't have to specify what file it is. If you know what the feature is called in the PRD, you'll have to specify language or variable choices or any of this, but just say we're working on just this feature, let's go, right? Making it that way is how you're able to build out a scalable, workable model to iterate slowly through your systems here. Okay, let me talk about 2 case studies real quick, and then we're going to go on to questions, right? The first one is Insight AI. It is this product. And if I scroll all the way back with Lovable to the very first prompt, if I can get all the way back there, I've had a lot of conversations back and forth with Lovable. Give it a moment to load. This did not start the way you are supposed to do it. I bring this up because this Project has become hard to manage, in part because I started this the wrong way. I've been developing on this for maybe 3 months off and on, and, um, what this began with was a classic vibe-coded kind of product where I just said, hey, I've got a cool idea, let's, let's build something, and From here, and you can see any of the orange boxes you see that I'm scrolling back up here, these are when something went wrong. The product wasn't able to build. Another reason why you use Lovable is if you're not technical, it will instantly catch when there's something wrong with the product it has built, and you are then able to, uh, go catch that problem before you spent 4 days building on something that actually can't assemble the code. So, I mean, I may not be able to scroll back far enough at this rate. This— I go back far enough here, what I'm doing is there's no design doc or project manifest to guide things. I am asking it, Hey, do this. Hey, do that. Maybe we can develop this. Maybe we can add this to our Supabase tables. Like, it's very kind of interrogatory, right? I don't have a plan. I'm asking the AI to do things. If I were to scrap this thing and build it again from the start, it wouldn't be a 500,000 line, or sorry, 500,000 character piece of code. It would probably be more like 40,000 or 50,000 characters. If I began with, oh, okay, I know what the end goal is in mind now, I can make that happen. But I started with, hey, I've got some cool ideas, like what can we add to make this a real product? And kind of navigated my way through to something that often works, sometimes doesn't. I would sometimes spend a whole day trying to solve one bug because I was working only in Lovable. That's why I started to bounce back and forth between Lovable and Claude, that I found out, oh, there's— when you get in this loop, you can get back out of it. That's a, uh, I'd say a slow-growing, relatively, uh, fragile product. You can go check it out yourself if you want to. I believe it's publicly available. I haven't actually tested it. For a while to see if it's working, but insigth@insight.crowdtamers.com, and it'll ask you to log in for free. And I think I'm using Gemini AI credits, like, of my own, so don't send like 10,000 queries in, please. But, uh, Gemini is pretty cheap to run, so it won't bankrupt me. Uh, and like, the whole point of this product was just, hey, I want to look at someone's website And I want to identify opportunities in messaging and then suggest ways to fix it. And then like I have some other screens on who your persona is and like what research we've done. But as I mentioned, it's kind of fragile right now. It doesn't always pull all of the data. So with the rewrite, this would be more dependable. Not great. Uh, when you do live coding like that, you will end up regretting it. A product that's working pretty well and in production right now is this one, Light.ai. This is an AI product that is also a service business on top. You can punch in your phone number, get a phone call from it right now. And the idea is that it is opted-in phone calls plus lead generation plus a bunch of other things. And this was built over the course of a couple of months from start to finish, but is in production for a couple different companies and, you know, making money for the business right now. But in part because there's almost no backend to it. The backend is all built-in tools like n8n and Railway, and it is not something that's client-facing. That is, if you want to build something quickly and you're not using— as I didn't, uh, this website here is built with, uh, with Lovable, but the backend is not. If you want to build something quickly that is, uh, some way to use AI right now, I'm finding it's way more effective to build something where you do 5 or 10% of the work and AI does the rest, and then you, you bill people for the human labor in there and it becomes something pretty scalable. Uh, the one last one, just because it's interesting project is CLAR. There we go. Clarity AI is an attempt at a project. If you go do GAUNTLET, GAUNTLET AI— I didn't spell it right. This is an AI, like a crazy AI training program that takes, I think they promised, like 120 hours a week is how much work you're putting in for, I think, 12 weeks, and then you are like, you are hireable. And the very first project that they do is, can you rebuild, uh, Hemingway— Hemingway— nope, uh, app. Can you rebuild Hemingway— I've got it wrong— Hemingway app. There we go. Like editor. There we go. Uh, can you rebuild this And that's the first project you tackle in there. And after I heard that's the first project you can tackle, I went, well, can I do it? And the answer is yeah. It's not like, it's not quite one prompt. I think it took me maybe 2 or 3 hours to build this. And Honeyway Editor is like, they make money. They're a full app, right? Like you can pay them money and they will deliver services for you. Uh, or you can write this yourself in a couple hours and have the core functionality that you want duplicated. So this came with a— it was easy to build a PRD because I went, what is— what's Hemingway App do? Let me think about that app, its features, and then let me write a project requirements doc from there. Very easy to build when you have something that's complete to copy from. It's okay. Those are like ways to build and what you can build. Um, this app that we're talking about right now, right, the monkey's paw. If I connected my API key to this as well, which I'm not going to do because I don't want to spend too much money with you guys hitting up my APIs, I could put this live right now, right? The results would be meh. It wouldn't be great because I haven't tweaked the prompt. It's not a fully featured app. But I could, in a matter of a few moments, take— well, let me, I guess, let me grab this real quick and, uh, go back to dashboard. We'll go to the Monkey Spa, uh, and I'll put in over here my API key real quick. Oh wait, no, you put in your own API key. I can publish this right now, right? This will take your AI. Hmm, interesting. I've never seen that particular problem happen before. Let me try the easy fix to reload the tab and then try to publish. But from start to finish, and with me only really working on the app for about 10 minutes of— well, that's fun. I guess I can't make it live. I could do— hang on. Oh yeah, this is having difficulties. M-O-N-P-E-Y. That's your subdomain. We'll do monkeypaw.crowdtenors.com. You can obviously put any custom URL you'd want there, but this is going to ask me to authorize. I'll just go over to a different monitor so you don't see my password here. This is— if Laravel will publish it, we'll see. Yeah, so that means this is now live, and theoretically you could go right now. I'm unsure if it's actually going to process that fast. Yeah, it'll take a few minutes before it's live, but I went from Beginning to end, right? 60 minutes to have a very minimum viable product working. Next week, we'll talk about AI productivity. We're going to talk about how once you take Claude code, or Claude Desktop, I beg your pardon, and once you install a couple of MCP servers, those are Model Context Protocol servers, you get the capacity to, like I was doing, Add-in, uh, we have Desktop Commander. This is the one that lets you read and write files to your local hard drive. Among the prompts that I have included in the prompt guides for this session, because if you want to start using Claude to develop locally, you will need this, is how to have Claude set up Desktop Commander for you. And then if you want to try Claude code, it'll also— a different file will walk you through how to set up Claude code as well. Those MCP servers, again, run at your own risk, because they are reading right into your hard drive. But they give you the capacity to have Claude work right in a file on your hard drive and make changes for you. You take that and you level up a bit with things like PowerPoint, or Google Search Console, or Google Calendar, or Google Drive. And we begin to have something that can actually control my day-to-day life, make things more efficient, and do things like, say, uh, there we go. I'm not going to open this up, but I had Claude AI go through my email and find, uh, all of the emails that were sales emails that I had a proposal out to that I have not replied to yet. And I, you know, it's easy as a founder, you're doing a bunch of things, to miss out on, uh, what it is you want. But I said, hey, look at my sent folder, figure out what I need to reply to. And then it went, cool, I will do that for you. And it found like $20,000 in proposals that I need to send replies out to. You can get the same functionality and a lot more. Go register here. Thank you for your time, everybody. Hope you found this useful. I know there's Q&A, there's a bunch of questions that have popped up. Uh, but that's all I have for today.

Awesome, awesome. I hope you can see all the emojis, Trevor. I think the folks loved it. Uh, pretty clear to see. So before we really get into the Q&A, honestly, how frustrating is it when your prompts don't work compared to, you know, when you've been working with a human being? Like, what's the difference there?

I'm sorry, how frustrating is what?

How frustrating, frustrating is it for you compared to like when you're actually working with a human versus working with AI?

So I personally, and maybe I'm weird in the head, I find AI is in some ways refreshing because it always does what you tell it. The problem is if you're bad at telling it, then you will not be happy with the results. And I personally, I mean, both my parents were English professors. Uh, well, I guess my mother, my stepfather, my father's a lawyer. Like, every way I can, both genetically and like growing up nurture-wise, was all about precision and language. So I have an uncommon superpower with being able to say precisely what I want that a lot of people I don't. So I find it is like AI won't outright lie to me with saying like, yes, I will do that, and then instead go out and go drinking. So at least it doesn't do that to me.

Honestly, I think you're probably built to be a prompt engineer. Like, you have everything there is, you know, for it. Uh, yeah, cool. So before we really get into the questions, uh, a quick thing here, Trevor. We had a couple of folks, uh, letting us know that they had issues with the CrowdTamer's link. Uh, some people, they weren't able to get the assets. Uh, something about not being able to enter a non-US number as well. Just wanted to flag this with you.

What will go for, for the, uh, the HeyVail website, the phone number? Yeah, links. Oh, you can't do a non-US? Really? Um, okay, well, I will check that out real quick. Um, I don't have it set to only accept US numbers by default because I have tons of clients in Europe and in Asia. So, um, I can— I can— the easy thing I can do real quick, let me just do this while we're talking, is I will go remove the, um, the mandatory for the phone number, uh, and then that way if it doesn't accept your number, it won't matter because you don't have to put your number in. So give me one second to fix that. That may be the problem folks had last time, but I didn't have specific enough info on it. And my apology to folks from anywhere but the US that you're having a problem putting your phone number in. Uh, wasn't my— wasn't my intent.

Okay, cool. So folks, generally what we do is we do send out all the resources that we're talking about, which includes, uh, the CrowdTamer's link as well as the replay here. So you'll get it maximum by the end of the week.

All right, cool. Uh, I definitely do like recommend to you guys Once you have access to these, right, if you, if you are in FI and you're trying to build stuff, or if you're trying to think of what to build before you join FI, seriously, for $60 or $80, have all the tooling you need, and you can almost instantly begin, uh, making progress on, hey, that thing you've always dreamed of doing is now a reality. Awesome.

Exactly. Okay, so Trevor, are we ready for a couple of questions?

Yes, sir.

Okay, so, and we got different variations of this question here, but I wanted to kind of put it into my own words. How do you pick your first AI tool? Or better yet, like, how do you like select your kind of, you know, Avengers of AI tools where you kind of start off?

You know, what's— do you pick Replit?

Do you pick Lovable? Do you pick Bolt? And then how do you decide between maybe ChatGPT and Claude and so on?

Sure. So I have a current and active subscription for at least the pro tier of every tool, right? 'Cause I'm testing them all the darn time. I even pay for Groq, heaven help me. And the tools I'm recommending here, and specifically these tools, are because they have the combination of user-friendly and powerful. Claude Opus 4 is still by most standards the best programming model out there. Google says that 2.5 Pro is better. In my experience, no, but maybe you'll have better luck with it. I don't find that OpenAI's OpenAI models are any better as a technical developer. And the downside to OpenAI models, if you want to use them to read and write locally on your hard drive, you have to install a tool like Cursor or like VS Code, which are not as easy or user-friendly for most non-tech founders to set up. And then you have to like, it's it is more challenging to get the local experience to do everything you want. With Claude Desktop, you download it, you install it, you run the prompt that I've given you to have it install Desktop Commander for you so you can write to your local files. And then from there, basically everything else you want to do, you can instruct Claude Desktop to do for you, and it will help you with setting up GitHub. It'll help you with setting up, you know, whatever, npm and Node and all the other tooling you will need to not just build locally, but also to deploy locally if you really want to do that. And none of the other tools are that easy. Um, Lovable, I just find is the quality of the output is better for, um, for shipping front ends. And so for like everything we're doing in my agency is now we're using either Cloud Sonnet or Cloud Opus for writing or for code, and then we use Lovable for front ends. I still have, like, CrowdTamer's website is a WordPress website, so Lovable actually doesn't work for WordPress websites, but I've had WordPress for I think 11 years on my domain. I'm using Claude just natively to make edits to that simply because of, like, I'm on this old piece of legacy tech over there. But if you're building something new, I would start with Lovable. I would start with Claude and Claude Desktop. And then only when you're comfortable with those, I would start to look at maybe I want to try some other tool.

And on Claude, we had somebody ask why even use Lovable if you could technically do everything on Claude, right? It's obviously a question of how comfortable you are.

Yeah, well, a bit, but also like with Lovable, I get instant access to Supabase with basically one click. And Supabase is both my database, it's my hosting, it's edge functions, so I can have a server do a thing to another server. Like, all of that comes free, and that's just one click. You have to pay Supabase, obviously, but the setup is idiot-proof. And if I'm only running Claude locally, I need to be comfortable setting up GCP or AWS or Azure Web, like any of these services, to then do all the servery bits that I want to do. And this is MVPs for non-technical founders. I assume that you don't want to be doing that.

So we have a ton of interest in, uh, cloud, to be honest. And I think this is a problem like everybody faces, and I've had this happen so many times, but how do you circumvent like the limit rate issue with Claude? Like, what are some tips and tricks there?

So Claude is the only AI service I pay for the max plan. I'm paying them, I think, $200 a month for the— for everything they can give me. I would note that Anthropic is losing money on that deal. I know I'm losing money on that deal because when I look at— I used to spend more than $1,000 a month on API credits, and I was always mad because I cost a lot of money. And then I switch to Claude Code Max or Claude Max Pro, whatever the name of the thing is. And I've increased my usage because now I'm not paying per API credit and I'm spending a fifth of what I did, right? So this is a leader, a loss leader for them right now. But that is, you have to work really hard to use up all of your token limits when you're using the Claude Code Max plan for $200 a month. You can, uh, I don't think you could do it if you were running only on Claude Desktop, unless you just sat down and used Opus Pro for like 8 hours a day. The other thing, and I've developed this, um, with Claude Desktop, because Claude Desktop by default is not super efficient at memory allocation. Unlike Claude Code. So I have told Claude Desktop to, whenever it's making a code sample or whenever it's making a piece of copy, don't put it into the desktop app, save it to a file on my hard drive locally, because that way it's not taking up context window. So like, you can do tricks like this to reduce how fast you consume your context window, but also you just, you're going to have to every you know, 150 messages back and forth, stop, summarize the conversation, start a new one. Like, otherwise you'll eventually find you've just gotten stuck.

And also, like, a slightly technical question: can you actually work on a pre-existing code base with any of these tools? Let's say Claude.

Oh yeah.

Does it have to be, like, you know, built-in Claude?

Well, no, Claude is excellent at reading a code base and giving everything from just suggestions to doing actual refactoring. I mentioned how my Insight AI tool started as pure vibe-coded all in Lovable, and I kept reaching stuff parts where I was stuck. I couldn't go any further. And so what I did then was I got Claude Desktop. I installed Desktop Commandments so I can read local files. I downloaded the repo on my hard drive, and then I would tell Claude, this is the bug I'm seeing. Go look in the files, tell me what's wrong and fix it. And so it was pre-written code. Lovable wrote the code and then Claude came in and worked on it further. You can even have really big code bases, but if that's the point, very much don't use Claude Desktop, install Claude Code because it is way more efficient at reading files and using the context intelligently.

So let's say somebody, you know, they want to get way better. Creating apps using AI, they've seen your workshop, they have all your resources, like what's the next step? How do they work on the skill?

Only way to get better at something is by doing it. And when you start at it, you will suck. That's a guarantee. I did not the first time I sat down to write, I don't know, a story when I was in, I think it was second grade. Was I good at it? Of course course not. I was in second grade. The first time I tried to write product with AI, was I good at it? Of course not. I've been doing this for 2 years, grinding away and smashing my face into this problem over and over again. It's easier now than it used to be because AI is smarter than it used to be. But just accept the vision you have the first time you do something. Odds are very high you're going to work on it for a day or two, and already you will have learned enough to go I'm going to scrap this entire project and start over again because I've now learned a bunch of things I should have done differently the first time.

Okay, so what I want to do is go through, uh, 2 or 3 more questions super quickly because I realize we are like way over time right now. Uh, so these are a little more generic, not necessarily related to Claude or Lovable, but question 1: What exactly can you and Can you not build with AI? Because somebody was asking about whether or not they could create APIs. So what's your take on that?

You can create an API. You can create crazy stuff with AI. I have seen somebody share that they had written in assembly instructions to run on a silicon chip, like natively on silicon at the bare metal level. And Claude code was what was generating this for them. Now, they were intelligent enough to know the output quality and to give it some steerage, but they mentioned they were by themselves shipping as fast as an 8-person team because of what they were able to do at like— that's extremely niche code, extremely niche use case. This person was basically writing software to control a Bluetooth chip on an actual breadboard. And to be able to make something that hard with AI indicates you can build most anything. Your real limiting factor is AI's context window isn't that big. Gemini has the biggest context window. It's why it's my fallback. Is Gemini will— Claude will hold 200,000 tokens, which is about 100,000 words. And Gemini will hold a million tokens. Right? So it comes into about half a million words worth of context. With that limited context, it's easy to consume it all just reading the code and not being able to do output. So once you start to ship bigger code, you have to get very good at being very specific about, I need to touch this feature on these files, only look at these bits. If you can do that, I mean, I'm sure there are limitations to what you can do with code and AI, but they are pretty few.

Okay, so let's say we have an application, we built something. When should we consider going out into the world and, you know, getting customers? Like, when, when should you think, hey, yeah, I think my MVP is ready?

Well, Honestly, as soon as the landing page is done, you should start. And since we just saw you can build a landing page in 10 minutes, any of you who's listening here, what is today, Wednesday? If you have an idea that you want to make happen, um, build it and go to market with it by Friday. Because as soon as you start getting people to see what you've built, they're going to say, oh, does it do this? Oh, this is where I thought it should do this, or Hey, your idea stinks. I don't like you at all. Or your landing page sucks or whatever, right? And that feedback in is invaluable to building the product. Until you start to see what your user actually wants, you can spend months or a year making the perfect thing that no one wants. One of the key things about what AI enables is you can ship the smallest thing fastest. And by being able to ship something tiny and quickly, you're able to see if people like it. And if they don't, do a different thing. You didn't have to spend months making the first version work. You spent a day or two. And if everyone goes, oh man, that sucks, message received. Let's do the next thing.

And kind of on the other end of the spectrum, for my final question, when is the right time to go out and maybe get a technical co-founder or find a team to help you build on what you have with AI?

Sure. So fundamentally, a lot of the principles around good application security don't make sense to somebody who is not trained in it. Once your application starts to attract users and start to make actual money, as hard as you might try, to implement good security practices, odds are pretty good you won't. And having a technical person come in and evaluate where you are and where you want to be as you start to see real users is a good time to do it. Uh, there are some industries where you just, like, you can't launch in with a purely MVP product— health tech, legal tech, Finance, like any of these, the security standards are so high, your product can't actually onboard users yet, but you can do sales demos. You can do investor pitches. You can show the vision here and being able to show the actual workflow you want to a real developer makes it very hard for them to get it wrong. Even if the backend has to be changed, if the frontend works correctly, you're golden. So I'd say where you have to bring on a tech co-founder or a technical partner is going to vary based off of the industry that you are in. I would normally look to see, once you start to have a handful of active customers who are paying you every month, is when I would start to say, okay, at least bring in a contractor. To evaluate the code quality and say, hey, these are the good parts, these are the bad parts. So you can go back and say to the AI, these are the bad parts, how do we fix it, right? And it may be you discover you want to bring in someone full-time, a co-founder, like anything from there. Once this is earning you money, it's time to start to get some more eyes on it.

Awesome. And in 90 minutes we built an MVP, we answered a bunch of questions. Trevor, thank you so much for doing this. As always, it's really great having you here. And I can tell you by the reaction that we had and the number of signups, by the way, folks, we had almost 1,000 people signed up for this. It was pretty great. And we have another one coming up. So we'll be focusing on kind of supermaxing your productivity with AI next week. Same time right here on Airmeet. We have our producer Sana sharing a link in the chat chat. So do check it out to sign up, folks. Uh, Trevor, before we end things here, uh, any words for our audience?

Two last things. One of them is I have edited the form on, uh, crowdtamer.com/ai-for-founders. The phone number is now optional. If you had any problems, go back and try again. You should not need to enter your phone number, so that will get around that bug. Uh, thank you for people who were specific about it, so I do want to fix this time. The other thing is I encourage all of you, stop dreaming, start building. This is the easiest it has ever been to go build things, to go make a difference in the world. Every founder is a little bit crazy because they— there's no reason for a business to exist except that you've manifested it there. So go out there, go manifest it, go make it happen. I want to see what you build. Thanks, everybody.

Awesome. There you have it, folks. Uh, go out there and build stuff. We'll see you in the next one. Uh, do remember to sign up. We're also doing a bit of networking, so if you want to stick around, maybe connect with founders from across the world, uh, please feel free to do that. We'll see you in the next one, guys.

Take care, everybody.