271 - Best of 2025! How to Build and Sell Apps with No Code with Billy Howell

0:00I'm thinking about how to turn this into an app. If we took like that recording, 40 seconds of you describing the app, we could throw it into ChatGPT and say, turn this into a PRD, then throw that in Replet, and it would have the functioning app exactly as you described it in like five minutes. It's like ridiculously good to the point where like I've got a couple apps that I've been slowly chipping away at for a month or two in Replet. I almost want to just blow them up and start from scratch because this agent is just so good. I officially sold 10 of those 70, which isn't a bad hit rate. I think this might be the easiest way to do it.

0:58helping entrepreneurs either finish or even build their own custom apps using AI. Correct? Correct. Yeah. I probably officially sold maybe 10 of those 70, which is a bad hit rate, I guess. Yeah. That's an amazing hit rate, dude. So I reached out to you because I wanted you to come back on so you could teach me how to build a chat bot. It's something that I've been messing around with. There's a couple of use cases for it. The biggest use case that I was thinking of about for my podcast because I've had a bunch of interviews. It'd be nice to go query and say like, hey, who did I talk to about X, Y, and Z or which companies were the largest that I spoke to, whatever.

1:36But I think there's also a lot of small business use cases for chatbots like onboarding, orientation, HR questions, the sales process, just internal SOPs, et cetera. I'm excited to learn from you today on how to build chatbots. There's so much SaaS out there for getting custom chatbots and it can get expensive. I think you could pay 30, 50 bucks a month for one that slaps into your site. And it's really trivial to just kind of spin one up in Replit. All you need is a Replit account and an open AI key. So I think this is a good project because anyone can do it. And once you've done one, you could go show it to like a local business and say like, hey, can I build this for your site?

2:18And then charge them monthly to use it. Right. Or don't. or just charge them the one-time fee and get them set up because that could be part of your sale is like no monthly. I think the thing for me that's been the hardest thing to like try to figure out and we'll dive into this is the context window is limited and I have a ton of transcripts and documents. So I'm trying to figure out like how do I get all of the pertinent information that I want to within whatever the parameters that are necessary in order for the chat bot to be successful. Yeah, that's definitely a hard one. There's a couple of different ways.

2:50you can kind of expand that. Let me open like a screen to share. Because the software that you're talking about, like that there are already software tools, are those tools that kind of already have a plugin to like say Slack or your email? Or are those tools that I've got to go out, train it on, and then it becomes a chat bot? Yeah. It's probably both. This is chat based. This is one of those plug and play solutions I mentioned. Okay. So I haven't used this, but someone on my team has used it and kind of recommended it for a project. What's it called? Chatbase. Okay. And like I said, I think you can get started for zero bucks and then go up to 40 for more messages, more agents, more models.

3:36We have 10 links to train on. And I don't know the difference between... It looks like you're adding your sources. I'm just like scrolling down on the website. It looks like you have to go in and add the sources that you want. Yeah. and sync it with whatever which if you're a company that you're like oh our crm is hubspot cool maybe it goes and trains on hubspot and then you have a an agent that you can go and query for specific deals within hubspot as opposed to having to go to hubspot and pull stuff exactly let's see i'm not exactly sure on how we can do the context window thing my instinct is to use open AI assistants.

4:16So think of like, there's two different types of endpoints you can do on open AI. There's more, but like, there's the main one that's like chat, like you send it, and it just takes a prompt and it responds. The assistant is like, you get a unique ID, and it just creates a new assistant as if you'd open a chat GPT window, and it will remember context. And so I use that when I need context. But I don't know when that window evaporates and how big it can get and how we can kind of figure that out. That's a really interesting question because I know that like chat TPT upgraded their memory, so to speak.

4:52So it's like, oh, it remembers other chats, but it doesn't remember everything. And obviously there's a fundamental tension between speed and accuracy. Many times it optimizes for speed over accuracy. So it's not always pulling the correct information. What I've actually started doing with some of my chats is after a long chat in particular, I'll be like, please create me a context document that I can upload to a project that will allow me to pick up where I left off essentially is the prompt. So that's been good. But for these larger, like I have 200 transcripts, right? And each transcript is like 45 minutes.

5:27And even if I clean it to the place where, Hey, get rid of filler words, get rid of ums and ahs and likes, and it's still going to be far more than a million tokens. So I don't know if that's where then you get into a situation where it's, what's that called? It's like, not RSA, RAF. Oh, RAG. RAG, thank you, RAG. Yeah. What is that? Yeah, so I'm not sure if this is what it is, but like the solution that you're talking about, there's something called like embeddings in OpenAI. It'll take in like a bunch of like text and kind of condense it down, I think into like that AI language that they can parse quicker.

6:10Do you know what I mean? Where it's like the weird symbols. Totally. Yeah. Yeah. It can query that faster and use less memory. OpenAI's text embeddings measure the relatedness of text strings. Embeddings are commonly used for search, clustering, recommendations. So this is a possibility we could use embeddings. So that's like you got like your assistant that we talked about. And then it'll have kind of these embeddings where it stores kind of this data on OpenAI. and the first time I used this I used it for one where it was querying a bunch of SKUs from a CSV for e-com for store to write SEO articles about them and I ran it and it's kind of slow and I was like huh and there's like 60 ,000 SKUs and I went I looked at my usage and just like one run blasted like 10$20 worth of tokens because yeah because on accident there's just so many embedded skews so i think you just got to be careful about it i definitely wasn't doing it correctly okay so we don't let's not do anything super complicated but i think we're gonna ask chat gpt like oh okay i want to make because it'll know better than us i want to make let Let me dictate real quick.

7:21I want to make an app where I can upload transcripts from my podcasts and then have an AI chatbot on top that can answer questions about those chat scripts. I have about 200 transcripts and they're all pretty long. What's the most efficient way using OpenAI's documentation to give the chatbot context of all those documents? So we're going to ask about ChatGPT and hopefully it'll tell us. To give the chatbot context for all those documents. I love hearing how people do props. I feel like I learn something every time I hear. Because everybody does it different. And at least for me, it's not like you hear everybody else how they do their prompts.

7:56So it's just nice to hear. Anyways, it's cool. I've noticed lately, think about when ChatGBT first came out and how we would all start with, you're a marketing expert. You're a software developer. And now we all just kind of defer to it as the expert on everything. We don't say that anymore. Isn't that strange? Yeah, that is interesting. I do still use it if I want it to act a certain way, not necessarily have a certain knowledge base. If I'm like, hey, I want you to be a no-nonsense evaluator of business processes. Don't sugarcoat things, but don't be rude, right? It's like, because many times when I ask it a question, it'll be like, what a thoughtful question.

8:35That's a great idea. I'm like, I don't want you to be happy for me. I just want you to give me the freaking best answer. Okay. So it pretty much gave us like the answers we talked about. So it's talking about chunking. So just like pulling out important chunks of the transcript. I was actually thinking a shorthand way to do it would be have like OpenAI, like summarize the documents and just put the important information. So you could be like, hey, like it might have so with so-and-so and then it would know to go in there and then query that. Can I show you something that I actually did before?

9:08It's not up to date. I wish it was up to date. I probably did this a month ago, so it doesn't have my most recent podcasts. So what I did was I have all my podcasts in a folder, transcripts. And then I went to Google Apps Scripts. I ran a script, essentially, that went and populated this spreadsheet with OpenAI's summary of my podcast. And so I asked it to find guest name, business name, industry, business type, whether they started or bought it, location, revenue, employees, funding type. all these things. What was the key tactical advice? What was the best quote? What's a one-line summary? Just everything that I thought it would need in order to have the context necessary to answer good questions.

9:50I think this, it might be the easiest way to do it is to just create this podcast database spreadsheet and then a chat bot that goes and queries this. What do you think of that? Yeah. I mean, I think we are going to need something like this anyway, so that it could be like so if you ask like hey what do we talk about on sam thompson like you could do that quickly yeah first let's just try that yeah all right let's do that i just sent you the link to the spreadsheet so let's download the csv of this transcript and then i'm going to share my screen hey i don't know if you remember this but when we started this podcast we entered into a social contract i would spend time energy and money producing this podcast interviewing these individuals and giving you insights into how to build, buy, start, grow your business.

10:37And you would like subscribe and leave me five-star review. Now, out of that, we both get to talk to really cool people and hear really cool insights. We both get a ton of value, but I just want to help you keep your word. So would you do me a favor? Will you go leave a five-star review for me on Apple or Spotify? It would really help. And if you want, even share this with a friend. Okay. So I've already talked with chat CPT about how we're going to do this. I'm going to give it this download. Okay, here's what our database of podcast info and transcripts looks like. Can you walk me through a web app that has a chatbot look at the title and the date and the snippets and the summaries that I have in there?

11:31and then finds like the relevant episode and then looks either in those snippets and then into the transcript if it doesn't find what it needs in like the snippets or the episode information to answer my questions. Let's see. I see you use talk to text a lot, which is cool. Do you use the, I don't even know what to call it, where you're like talking to the AI? The chat GPT, yeah, has like just a straight to audio. I do that a lot. I do that when I'm like driving and I'm trying to like work through an idea. It's pretty cool. The only problem for me is the latency. Why? It's too fast. I'm serious because I'll be talking and then I'll pause like I just did there, right?

12:10And it'll jump in and kind of cut me off. So many times mid-idea. What I need to do is at the beginning - Can't you just hold your finger down or no? No, that's the talk to text, but you can actually be talking with it. Yeah, on the iPhone. Oh, maybe. I don't know. What I need to do is just say, hey, I'm going to talk to you. don't answer me until I say, okay, ChatGPT, it's your turn. You know what I mean? That's what I need to do. I just haven't done it. Must be user error. Mine works pretty well. I'm an idiot, Billy. I never said it wasn't user error. I'm just telling you why I don't like it.

12:41Okay. All right. So what did you just ask it? So I asked it, can you walk me through a plan for building this web app? Okay. And it kind of responded with the goal, the data fields that we have, it looked at that CSV and read it. And then it's giving a stack here of what we should use. oh cool and it's giving us like some example code i'm going to tell it okay now write me a prd product requirements document outlining this plan for our ai developer in replit don't be opinionated about what stack to use except open ai for the ai and don't put any code snippets just write it outline a plan on how to build the app okay i've never heard prd what is that a product i think it's product requirements document or project but it's basically what you like hand an engineer to have them build a feature or an app oh and so telling it telling chat gpt like i'm trying to build a prd it knows that there's certain things associated with a prd that it needs to include in there yes and i've kind of just i've just kind of stolen this term you know i've never worked like as an engineer i don't think i've ever seen one of these like in a real workplace so it's pretty much just like a planning document but you can see like it does user flow technical requirements success metrics so it's templated based on something okay so we're gonna take i'm always like that i'm just like there has to be a template for it i don't want to reinvent the wheel like what needs to be included in this so that's prd i love it and then i have this funny thing i've had it up for as long as i've had open ai we're in the prompt i say like ask me three questions at the end of every response about like something i might need to know more about and it's helpful sometimes i'd never thought of that until i started doing a lot of deep research and deep research is the only thing that actually asks you questions and i was like oh that's nice to have so i've started doing it i think that's freaking genius you've been doing it for way longer can you see replet i can all right i believe into the fun stuff now Now I've pasted this PRD in and I'm going to say, build a web app based on the attached PRD.

14:51We'll need a database. I don't even know. That might confuse it. We can add that in after we start. I'm going to let it do auto. It'll probably just choose modern web app, but just not going to touch it. And it's going to have this PRD in the file tree. It's going to be saved so that it can refer back to it. So if you're ever talking with the Repli agent or assistant, you can add the file that you pasted in and say, hey, go back and look at that plan that we made. Dude, can you do that in ChatGPT? No, I don't know. That would be sweet. Because I've uploaded so many documents. And then later in the thread, I'm like, can you actually look at this language of this contract?

15:30And it'll be like, I don't have access to that anymore. You would think because ChatGPT is local, they'd be able to make it work. I mean, it works for projects, right? Like I have a chat GPT project, like go refer to the project docs. But if it's just a chat, it doesn't. I love how that's freaking cool, man. Okay, we're looking at replets, kind of going through the planning process of the app we just told it to build. And this is kind of a newer feature in replet, but we've got kind of like this plan and approval screen where it's saying this is what the initial version is going to be. Here are all the features and we can get rid of ones that we don't want.

16:12So if we could go through them. Yeah. So instead of waiting for it to spit out the version and you're like, oh, I don't like this, this, and this, you're able to approve beforehand. Right. So you don't have to go back and remove bad code. It's going to go ahead and do a visual preview, but it's just the front end. Okay. So we want the first two. We want the third. This is good. This will save us tokens. attempts to answer using summaries I will see that works like ideally like future state it would be like oh yeah you talked about this on this episode here's a link to it and pull up like a nice card that shows all episode information you like click to expand and see the transcripts I really like this approval thing because so many times I've tried to play it I do lovable but replet and lovable obviously are similar it'll spit something out and then I'm like oh I don't like this change it and then it'll change it but it'll also change something else or now it's in the code and the fix is like a fix but it should have just you should have just gone back and recoded it the right way the first time you know what i mean so i think that's a really cool feature i haven't really touched these later ones i don't know how much it really matters i think it'll just give this as like hey do this after it's done all of these so but then the other thing i don't really know someone from replet tell me how would i even say I want to remove this and then add a new one.

17:38How would I do that? Because all I see is the only input is approve or delete. So it's a mystery. I have no idea. Come on, Replit. Approve. Yeah, come on, Replit. Get it together. Why isn't this perfect yet? Why isn't it perfect? All right. So now we're in Replit. And we've got our preview of the app here. This isn't like a real live preview. It's just like basically showing us like HTML, like the front end code. Like it's not attached to anything. It's not interactable yet. And then this is the agent. So it popped in that stuff that we saw earlier. And he's going to work right here and tell us what he's doing.

18:19And in the meantime, what else? I think I'm going to go ahead and just create a database. Look, it's kind of it's doing some CSS stuff. I know what CSS means. But like for people who are listening that don't, because I do. Right. what exactly is css it's cascading style sheets so first there is html which is what you're looking at here this is like what websites look like in the 90s and then css came along and it's like an extra file that makes rules for okay like make everything as a header blue like make every button styled like this so it's styling in like a machine and we kind of evolved past like most people use tailwind now which actually is just like they write those little rules like make this blue and make this 30 pixels they just write it like in the name of like the element i think it's easier for ai to do it that way but that's kind of how it goes now so it's essentially a style guide yeah it's a style guide to make everything cohesive that's the way my neandroid ball brain needs to understand what it is.

19:28And well, I saw this guy make like a inflammatory site that was like this, like it's like effing HTML.com and he dropped it. And it was like, this is like an HTML site. Like you don't need like JavaScript and all these complex stylings and animations for your simple informational landing page. Like this is a button. Look how good it looks. It's just HTML. and I kind of love that. So I want to try to build an app or a landing page that's just HTML. I think it fits really well with the ethos of my agency, which is stupid, simple apps. Oh, for sure. Yeah, that's beautiful. Because I don't care about perfect UI that much because most of the apps I build are internal tools for businesses and they just want it to work, right?

20:18Like this chatbot. Yeah. I don't care how cool it looks. it's not flappy bird or i can't think of an app with like amazing ux that was the first one that came to mind but you know it's on iphone app right table one that'll be my example my buddy's company table one has really good ui and it's a mobile okay it sounds sounds amazing back to the task at hand here i'm going to save us some time and i'm going to come in here and create a database because i know we're going to need to store the csv information i'm just going to click create manually because oftentimes the agent won't it won't create one for you unless you tell it to so i'm just gonna have it here and now it'll just see like oh an open port to a database and it should know that it has something that it can build wait do that one more time where did you add the database i went down here and i clicked on database and then there was a big create button there's a create manually which i clicked and there's an ask agent to set up for me and i just You said that the nicest way anyone has ever called me an idiot, and I appreciate it.

21:21You were like, oh, I just went here, obviously you weren't looking, and clicked this button that says database, and then the big button that says create new database. So that's how I did it. I mean, I'm just so used to this workflow now that I need to slow down sometimes. Because even if someone's really good at AI coding, I think they may be used to the lovable interface and everything slightly different. Well, because I would have, me using it, because I'm not as skilled as you, I would have been like, okay, well, now I got to do the API call to my Google Sheet, right? Right. But you can just add the database here, which is awesome.

21:57Yeah. And I just don't, whenever we do these quick builds, I want to stay away from like Google Cloud integrations because it takes like five, 10 minutes every time. Go create a new cloud project. Set up a redirect URL. It's very easy. It's very frustrating. And I know there's probably better ways to do that quicker. If you were building this long-term though, would you have done it the other way? Maybe, but... Like, is it better to have a local database like you're doing here, building within Replit, or have it go call out to wherever the data is? A real computer engineer would probably say a database because you have more control over the structure.

22:33Like, you can't go into a bunch of emojis like you could into Google Sheets and break the whole app, right? And to me, it's less complex, which I guess would be the word simple. It's simple to just input a CSV each time as you need it. There's like, I think there's less things that can break. But yeah, I guess long-term, like a consumer internal app, I would do it, I could do a Google Sheet. And it's still going. It takes a decent bit to get these initial apps built, but it does such a good job. And it's using Claude Sonnet 4 under the hood. Well, I don't know what for because the way Replit works is a cocktail of different AIs.

23:18So it may be like, okay, Claude's the best at databases. We'll have him do the database task. OpenAI is the best at generating interfaces. We're going to use that. Gemini is best for this, right? But I just know that they added Claude for whatever recently, which is like the hot new one. So when you created the database, did you like copy and paste or did you put the CSV in? I haven't done anything. It's literally an empty database right now. It's not an empty database. It's already migrated in. So it's matched at CSV. Did you have to tell it to match the CSV? No. And you know what? I should have, at the beginning in that prompt, uploaded the CSV so it had it to look at.

23:57But I believe the PRD probably did a good enough job describing it. And so this is nice. So I won't have to go expose an OpenAI key. but if you're doing this at home you would go open ai api dashboard and it's like so many clicks to get to i need to bookmark it so here platform.openai.com and then you'll go into api keys and create a new one and paste it in and then replets nice it'll ask you do you want to save this as a global variable so i've just got one one api key that's stored that i can use for demos like this got it but it's important like if you're building on a real app to get your own unique keys so that in open ai you can monitor like oh this app's using this many credits right so you can kind of tell so how much money do you spend a month right now on just personal apps not for your dev shop where you've created things for people but in either open ai or just token token spend probably like 50 60 bucks on replit just like kind of running the agent and then the apps themselves don't use a ton except i have one like internal app that my team uses setecom one that uses like a decent amount of replit database storage because it's i did it super inefficiently and there's so much data so okay so let's say 50 bucks on reddit replit tokens 20 bucks or whatever for the membership so it's 70 then i mean like the open ai you could see can i tell you a cool tool that i think you should build yeah for people who are vibe coding i think you should build a dashboard that tells them how much spend they're at either that day or per month or whatever because every single person i've spoken to who vibe codes has been like oh dude i lost 500 bucks that day yeah everybody has had like a mistake just because they didn't go and look and see what the usage was.

25:49So I think that would be a cool tool. That's a great idea. So I think the tally is 70 bucks on Replit. Looks like 10 bucks on OpenAI calls and then 15 bucks on Windsurf, which is what I use when I'm editing code locally. But that would be really great and probably easy to build to just make a usage dashboard for all of them. So it would be really great. I know there's like solutions like Rocket Money that will look at like your credit card statements and tell you like your SaaS expenses. I would love to pull in like all my business costs, like SaaS and then like API costs right into one dashboard.

26:27And even it would be cool to segment it by like client or project. That would be sweet. Well, and it would be nice because you can go into your credit card and that's a backward looking statement of what you've already paid for usage. It would just be nice. you could set it for like every three hours or at the end of every day where it just goes pulls the usage and then like populates it in your dashboard just so you know like oh today yeah wait why such a big spike today so you can cut it off early as opposed to wasting a couple hundred dollars yeah yeah i think most of them have this but like replit and open ai you could set a threshold to like send you an email and then like a threshold to like cut off so you're not going to accidentally run up a hundred thousand dollar bill yeah agreed okay it says built podcast q a web application created chat interface added csv upload functionality implemented smart search set up episode transcript viewing system okay let's give it a shot i think it goes to like 150 episodes let's see 152 yeah just ask like a random what zero episodes uploaded successfully 137 errors.

27:34So I'm pretty sure I know what it is though. Okay. The upload failed. Take a look at our CSV and make sure the app is expecting the correct header columns, update the database schema as needed to. So I just think it was guessing on what the columns would we call it it didn't know because we didn't feed it at the beginning did you add the transcript text in here i added a column and i went and manually and pasted in transcript text to like three of them because it just it gets sticky with okay i don't even know how i would get open ai to go open a google doc right like it's like ridiculously good like to the point where like i've got like a couple projects or like apps that i've been kind of slowly chipping away at for a month or two in replet and i almost want to just blow them up and start from scratch because this agent is just so good from the get-go it's improved a lot really yeah just in this just in the like how long have you been last month or two yeah i would say this agent's gotten a lot more powerful and i mean just as the you know open ai anthropic improve their models then it's going to you know raise the tide of all these ai builders it's incredible so we're pretty much like the start is like anytime like nvidia chips get more efficient or whatever in three months billy's gonna be able to make a repli tab better right oh i do have something cool i want to show you at some point go on something i'm working on this is so bizarre but i was thinking about this game that i used to play like with my family on like vacation a lot called ultimate werewolf or something and it's like cards and it's like that game where like there's like one or two werewolves everyone else is a villager and the werewolves kill people at night and then you argue and you vote someone out and you try to get rid of the werewolves another game is like mafia mafia was that game exactly it's like mafia and i was thinking like how how fun would it be to make like an online version of that with ai as players but you don't know who's ai and who's people and that way you can just keep having games going because ai will fill it in so i made it in repli and it's like i think i'm gonna launch it soon it's like it's very funny dude well i want to have did i talk to you about the game that my son and i play every sunday no it's a geography game it's so stupid it's like i shouldn't even call it a game but i've just got a prompt when chat and i'm like hey my son is a fifth grader i'm a 39 year old man create a quiz for us on geography that makes it equally as difficult for both of us keep score we're gonna do three rounds and it does the rounds and like, it's awesome.

30:24It's so fun because it, you know, asks him a question first and then me a question. And it's like 10 questions per round. The best part is at the end, I've got this prompt where it's like, all right, write a victory speech for me. And so it'll be like, in the victory speech, it'll call out all the stuff that he missed that should have been super easy. That's so good. Yeah. And then I'm like, and then I'm like, write a concession speech for the runner-up and so then it writes a speech for him it's like oh that's so well dad you've lived for almost 40 years i'd be surprised if you didn't know the answer to these questions it's just a roast it's so anyways it's it's fun and we play it every sunday night so i'm thinking about how to turn this into an app and it literally if we took like that recording 40 seconds of you describing the app we could throw it into chat gpt and say turn this into a prd then throw that interrupt it and it would have the functioning app exactly as you described it in really five Five minutes.

31:17Yeah. I need to do it. I need to do it. That's really cool. I like the speeches. We always call mom in and we're like, mom, we know who won tonight. And she can hear the roast at the end of it. And, you know, usually she just rolls her eyes because we're idiots. But that's a fun conversation. Okay. So I think we let our little engineer get too far away from his desk here. It's kind of spinning its wheels. And I think what happened was we didn't restart the app so that the changes it made weren't implemented. And so I think it's freaking out thinking it was breaking. That's always good. Which is super avoidable if I just flip.

31:53There's like a setting I turned off that's like, let the agent restart your app at the end. And I turned it off because it was annoying me while I was like playing with the app. So most people probably won't have this self-inflicted problem. Okay. Oh, perfect. It's working. Wait. How did it? It didn't really do that. no it's lying to me let's become self-aware and is now lying dude okay this is another weird thing that happened to me i was using grok i don't know do people use grok do you use grok i always hear that it's super useful so i've been trying to experiment with it more and i was having it do research for me on some like real estate data for an app idea this guy came to me with and i was like asking for the data and it's saying you can get the data here here and then like mid-sentence it stopped and it said an email has been sent to the xi developer team with the subject line request for real estate data and then it kept going like nothing happened and it was like dude like what was that and it said oh oh it's just it's an old feature no email was sent don't worry but i wonder if they're just like spying on us and like i don't know if they're mining us for ideas or it it wasn't like an illegal request it was the weirdest thing random dude I wouldn't be surprised.

33:13Oh, don't worry about that. Gosh. Is your name Sarah O 'Connor? No? Don't worry about that question. It was a dumb question. Okay. We have a preview of this app. Okay. Uploaded successfully. And I'm going to go cheat real quick and try to give it a layup. So I'm looking at your spreadsheet. let's say what did rafael say about small businesses make the layup don't be angel reese that's brutal i had to i mean watching her try to make a layup is brutal really i'm gonna be honest oh i'm going here and to wrap it real quick and checking in the console because you can see logs of what's going on okay we got an error so we're going to copy that in i got an error please thought it was just supposed to work.

34:05Please find the problem and come back to me when you're 95 % confident you have the solution. This was user error because I didn't give it the CSV to start. And then I did the thing where I didn't restart the app after it changed the code. So this is my fault. I wasn't going to blame you, but I knew it was your fault. I'm just kidding. So I have a really cool one. I don't know if I've mentioned this yet, but it connects to Gmail. I'm building this for a client and it pulls in your emails and your spreadsheet, not your spreadsheet, your calendar, and it'll pull all the contacts from there and then zip it off to like Hunter IO, enrich the data with like the business information.

34:51And then it'll tell you like, hey, you've emailed this person, you know, five times in the last month. And it'll kind of give you like a green if you're really warm with them or like a red. It's like you haven't been to an event with this guy in two months. So it's like a kind of a business relationship manager. Claude does. Well, Claude doesn't do that. But Claude's the only one that I know of that actually integrates with your email and your calendar. Oh, I had no idea. I literally, like two days ago, did it for the first time. I bet that's incredible. I wish the ChatGPT did it, though. That's the only thing.