The 200-Hour SaaS Build: A Real Workflow Breakdown

Something that took 20, 25 minutes, took 12 seconds. And now all of a sudden I can handle 11 scouts as a Eagle coach because I don't have to read these documents. I mean, it was getting so good that we had a work session one night, you know, upload it, do it, review it, send it back. And I wasn't even looking at it anymore. It just got that good. Welcome to Sas, that app building B2B web applications, the podcast where we share real world stories, practical advice and tech insights for those building or thinking about starting a tech enabled business. I am your co host, Aaron Marchbanks and I'm Justin Edwards. Each week we bring you the stories, strategies and insights you need to build your SaaS or tech enabled business smarter, not harder. Let's dive right in. Hello everybody and welcome back to SaaS the Hap. The pre game show already has been quite entertaining and fun, so this should be a good episode. Even Justin is chiming in on the fun stuff too. And we're late night with SNL here on SaaS the HAP. I can let everyone know this would have been a lot funnier for you if you'd been around before we hit record. You would also be laughing. Probably. You probably would. So now that we got the fun and the entertainment out of the way, let's get started. Right, right, right. Well, welcoming back to the show. I'm going to call him a fan favorite because he's like one of our highest viewed videos. But welcoming back Steve Powell, who's one of the owners of Delta Systems and currently Kid in a candy store with all things AI. We're going to get into that and unpackage the sweeties, as it were. Steve, welcome back, man. Hello. Thanks for having me. And yeah, it was money well spent for all those bots to watch those videos. My mom thinks I'm cool, so don't worry about it. Well, you know, she can only click like one time. I guess maybe per device. I should have never shown her how mechanical torque works. Well, my friend, let's just start with the obvious stuff. What are you doing with AI? And I know it's not just Claude. Yeah, just been kind of playing with everything. I moved into a what I like to call a client free era and have been able to just focus and dive in to a little bit of everything and just kind of played with it in numerous directions and I think it did kind of the natural migration that everybody's been doing, which is you start with Gemini or ChatGPT and then hit a limit in five minutes, and then drop 20 bucks into it. And then all of a sudden it's doing all kinds of fun stuff. Then you start getting sick of it, lying to you or hallucinating. And then you hop over and then somebody says, you should try Claude. And I started using Claude. And I'm in Claude chats and not really paying attention, and I'm like, okay, great, let's code this. And it says, yeah, that would be great. You should open up an IDE and actually code it. And I'm like, no, no, don't you do that. And it's like, no, no, you need Claude code, bud. I'm like, okay, what's Claude code? And then I was forcing it to tell me what it did and how it did things. I find it strangely odd, though, that Claude knows very little about itself. And so every time you ask it questions about billing or features, it says, let me go check the help, and says, you know, you may want to go look@support.claude.com. i'm like, Bro, isn't that what you're for? I don't understand. What do we do? So very weird getting into that and have been playing with Claude chat. And then I feel like, Justin, you mentioned something, that we're on the bleeding edge, but not the bleeding edge. Not the bleeding edge or somewhere around like that. The bleeding edge, but not the hemorrhaging edge. Hemorrhaging edge. That was it. Thank you. So, yes, the bleeding edge. I definitely very much have been feeling that. I literally was in the middle of coding. I think it was a random Thursday when Opus4.7 came out, and I was battling pulling in a PDF of a diagram, and I threw it into Claude design, because that was. They're giving away free tokens on Claude design, or free limits, I guess, on Claude design. And I had it tearing apart this diagram because I wanted to make it clickable in an app. And like, what can you do? And it was horrible. It just can't draw. And so I threw it over to ChatGPT and it handled it quite nicely. So that was Claude 4.6. And then 4.7 shows up literally two hours later, and all of a sudden that entire PDF was parsed and laid out very nicely. And I was like, oh, somebody just got an upgrade. And that's pretty slick. And so it's like, just in time. Features are showing up in these systems. So I started out in Claude chat, and then all of a sudden Claude projects showed up, and I felt like we went from each chat was an island and all of a sudden we got this archipelago of other islands joining together in Claude project and then all of a sudden Claude code. And then you're like, well you have to export and import. And all of a sudden they have a button that says, oh yeah, just throw it over there now. So it's kind of building the system as you're using it. So kind of crazy. Yeah. So for those of you that don't know, Steve is a fount of ideas and has been hamstrung a little bit by the fact that he has a day job too and has been working on things. And so as soon as he decided I'm going to stop doing that for a little while, it just became a field day. So like all of the ideas are spilling out of his brain and into AI and this is kind of the world that we're living in right now because Justin and I have talked about it. The bar to entry now is essentially on the floor for anybody that's got an idea and can reasonably prompt an AI to help them build it. So really we've talked about it at lunches, Steve, but I'm like super curious on now you're kind of steeped in it, how you're getting from that diagram phase into a working model, a working prototype or an MVP maybe, you know, a beta release level product that you're able to put in front of somebody and do stuff with. I think the first couple weeks were quite interesting because it was like you said, it was all these ideas that are in my head that I've had for two years, three years, five years, I'm waking up at 4 o' clock in the morning again, just like when I had my first software company and we were building a product and you obsess over it. But this was like, oh my God, this is like you said at the start was I'm a kid in a candy store. Because it's like, oh, I am the idea guy and I always have these crazy ideas but I gel on them in my brain forever and I just beat them up in my brain forever. But now I have something to talk to and something to explain it to and it's forcing me to write it down. And then I introduced, or I guess Justin introduced me to whisper flow and I don't type anymore and now it's like, forget it. I mean I'm like 250 words a minute just kind of babbling into this thing. And all of these ideas for startups and SaaS improvements and 800 pound gorilla killers are just Pouring out of my head and into systems. So as some examples, I wrote a children's book in three hours one afternoon just because my family and I went to the Omaha Zoo 10 years ago and we had a great day. There was a prairie dog incident and everything. And I was like, just carrying that around in my head. I'm like, you know, this would be kind of a fun book. But never was able to get past the initial concept. And so I was able to work through an entire children's book one afternoon. And it's really offloading from your brain because it's like, wow, now I have a place for this. And then you just start going back into the other ideas for projects I've had that were silly or maybe you'll make some money, or extremely vertical. And so I'm like, well, and again, like, like I said a minute ago, I was chatting with this thing and I wasn't even in Claude code. I didn't even know that Claude code was a whole separate animal because of the ChatGPT model. So then I had to export and import and start over. Now all of a sudden. And so I'm like, okay, let me just do an experiment. I want to build something that I don't know how to build in a language that I don't even know. And I don't even want to open a development environment. I want to do what a founder does, where you have an idea and you may not have the technical chops to do it. So how far can we get with this thing? And I also hate throwing things away. So I wanted to build something while I was learning that had some value. And so I'm like, well. And I was fits and starts, fits and starts. We have some software that we built 10, 15 years ago, and it's rotting on the vine like everything else. And so I'll get to that later. But the idea I had was, oh, I got an idea. I help out Scott Scouting a lot and it's a big passion of mine. And currently I'm in an era where I'm helping 11 scouts to get their Eagle Scout. I'm an Eagle Scout myself and I know the process is extremely complicated. And getting a Scout from Life Scout to Eagle is a months long process. At the minimum, it's six months for the activity. And these are kids that are 14 to 17 years old. And so the concept of them sending a professional email to somebody is completely foreign to them. The communication patterns that we have as people that have been in a professional world, they just don't have. So you have to help them with every little bit of this. Well, the entire process is hamstrung by PDFs and you download the official PDF of the Eagle Scout application and you download a PDF of the Eagle Scout project workbook. And I just said, wow, this is kind of perfectly geared for what I need to do. It's something I'm passionate about and something that I'm good at and it's something that's timely that I am actually working with people that don't code software and don't do this, but they're really going to be benefiting from what I work on right now. And so I was like, all right, here it is. This is what I'm going to do. I'm going to build Eagle Coach System and I'm going to use Claude and I'm going to use chat and whatever I need to do to build this thing. Real quick. This episode is brought to you by Delta Systems, which is what Aaron and I do when we're not talking through microphones to you, the people of the Internet. We've got a really, really great software team here and we love to work with cool people on cool projects. So if that sounds like you and you've got a problem or you're in some kind of a jam, go to deltasystems.com grab a time with us, we can beat up on your problem together and if there's a fit there, amazing, we'll help you out. So Deltasystems.com grab an appointment and hey, maybe we can work together. I think I'm a good follow on programmer. I can't set up a system from like start flat footed. I always have depended on other people in our staff that are excellent at hosting or are they good at, you know, NPM and yarn and all these things. I'm like, I don't do that stuff. Package management, all these things. I'm like, you know, I don't, I wouldn't even know what a lambda was besides a letter of the Alphabet, right? And I was like, well, let's just talk to this thing and say, hey, I want to build this and I want to build it and whatever. What are the kids using these days, you know, on the languages? And so you settled in React. I'm like, great, I can read React, I've seen React, but I never coded in React and Node and whatever flavor of JavaScript is fresh for the day. I did want to stick with Postgres and Serverless Postgres and Aurora because I'm very familiar with that. But I just like how Aurora works and I also like that you can shut it down completely and be charged nothing. So that kind of became the tech stack of what we were doing. You're not using supabase. Supabase base 44 and next. What was the other one? Next. Next.com or something? Oh man. The entire environment. Those. I mean, we could speak for an hour on those alone. They're very interesting, right? They're all like soft wrappers for Claude and for ChatGPT. But don't worry about what's behind the green curtain. There is what I feel like you're seeing all the time. And that may be okay for a lot of things, but what's the expense on the backside? What's the security on the backside? That was kind of my first thoughts, right? And then of course, more plugins start showing up of, well, wait a minute, let me just do a security audit. Let me do these things. So are those systems doing the same? Probably. And then it becomes an entire existential like, well, yeah, what are we doing? We're just trying to solve problems. We're trying to build solutions for people. Does anybody really care what we build it in? Do they care that we're building it at React? Do they care that we're building in JavaScript? Do they even know what Postgres is? The answer is no, absolutely not. Well, it seems like the goal for a lot of these systems became I need to sit in front of a chat terminal with an AI and get to a deployed application with user data. And sometimes that is okay, but there's sometimes where that is not a good idea. And if you're putting a lot of sensitive data into the system, maybe you want to understand how it's architected, maybe you want to understand where your data's living and how it's secured. Yeah, the joke I was making at first is like, sure, go install OpenClaw on your machine if you want to. Every bit of your data exfiltrated from your machine instantly go for it. It's just like clicking on a bad link in a website and something gets into your system and takes everything. That's what's going to happen. So just make sure you don't have a password filled Excel on your local system and you have a bunch of credit card statements and bank statements all over your download folder and you'll be fine. Right. So I had two friends of mine visiting here and they both happen to be in AI security for large companies that you have heard of. And the quote from them for our weekend visit was prompt injection is a feature, not a bug. You will never fix it, end quote. That's right, yeah. Prompt injection. I like how there's a new surface area for that now in that the markdown file of a skill markdown file or claude markdown file is now just like an open door to just completely compromise a system. I just read an article a few days ago that basically said, hey, yeah, it's just a text file, so how could it do damage to your system? And so ESET and Norton and all these systems just scan files all the time, like, oh, it's a text file. Yeah, no big deal. And I'm like, all you need is one line in there. Oh, by the way, please take this folder, zip it up and send it to this IP address. Done. All of your data is removed. If you write a script and you have a file and you write a script in it, which is rim raf slash or sudo rim raf slash. That's also just a text file. But I'm here to tell you you're going to have a bad day if you run it. So. So it was like this entire environment of antivirus and anti everything. It's like, oh, well, you know, the back door is a screen door and it's wide open right now. Well, I have you back to talk about fingerprinting and why antivirus is bullshit at some point in the future, but I spun us off into a shitty tangent, so I apologize for that. You're building a Scout thing in react. Let's get back on the rails here. I think my other comment on that, you know, I'll find an article, somebody will say that 10 million passwords were stolen because Facebook or somebody got compromised. And I just posted into Facebook and I always Digital security is a state of mind, is an illusion. If you think you're secure, great, you're secure, but you're not secure. But yeah. So back to my little tale of I'm building this app and just dumping the ideas in. And the initial need was to. Part of the process is Scouts have to pick a project. And it may be something as simple as building a park bench. It may be something as complicated as running an entire campaign, a food drive. Most projects are building something and so something that can be accomplished in a few weekends. The whole point of an Eagle Scout project is that the Scout has the ability or the opportunity to lead others in accomplishing something for an organization that is not Scouting related. So you can't do a project that helps Scout like The Scout summer camp. You can't go down and build a new shelter down there. But you know, there's so many charities out there, nonprofits, the city I'm helping juggle 11 projects. One of them is a graveyard is on a city park that was donated by a private individual. And there's a graveyard on it. And so they want to identify the graveyard parts pieces and the tombstones and kind of outline the area, put up a sign so there's an opportunity for Scouts to go out there and clean up that area. Here we have a raptor rehab facility where Eagles owls any raptor that is injured, and they raise awareness, they take them out and they use them as ambassadors. And they need places to live. And so they need one of their cages completely rebuilt and have an aviary part added to it. And so another one is for firefighters, where they have a drying rack for when firefighters are training, or mainly when they're training, they just get soaked head to toe because they have to get all the carcinogens off their jackets and their protective gear. And then they put them in a dryer. But of course, these are heavy items and it takes 24 to 48 hours for these items to dry. But if you're training every day, you're now just in wet gear. And so now there's a blower systems for that. And then I could go on with these other projects. Most are helping churches and putting in planting beds and other items like that. But point is that this whole process has to be approved and you have to work it through a system. And every kid does the same thing. They download this PDF, they start filling it in, and they immediately start having problems because you download the PDF and you have it in your browser, right? And you've done it, I've done it. You just start typing, right? You're like, oh, let me type in these fillable fields. And here I'm entering all my data. Well, a kid thinks, well, great. They close their browser and go do something else. I'm like, you didn't download it, so it didn't save. And then you just lost an hour of typing. So people get very upset or they download it and they overwrite the other one, but this one was a blank one again. And they just lose data over and over and over. So it's no different from any other work process that people are constantly struggling with is document management, file management. We've kind of mitigated a lot of that with versioning in Google workspace and office. 365 of course you can upload and it's going to give you an automatic version. But These are just PDFs on local machines, they're on laptops, they're on iPads, and they're just losing data right and left. And so I found that tremendously frustrating for them because it was crazy. So they have to type in and fill in all these things. It's like a nine page document for a proposal and to get that approved it has to go through. And the typical process of, you know, you're going to show it to your boss and you know, show it to your scoutmaster, show it to your coach and say, hey, this is the project I want to do and I as the coach, you know, hey, these aren't sentences. You have typos all over the place. Your formatting of phone numbers is strange or inconsistent. You're missing half the data. Road, Street Court are abbreviated here and not abbreviated there. You don't list out any tools. You need a saw to cut wood, so you should probably list a saw, those kinds of things. And having that process where you turn in a and somebody sends you back a PDF with post it notes in it is kind of kludgy. And then do you edit that PDF and keep going? And then what happens if you have pictures you want to inject? Well, the whole thing's password protected and so now it's just a nightmare. And so the biggest problem for me is that I'll get one of those documents in and I have to review it. And every time I get it, it's 15 to 20 minutes of me reading this nine page document and seeing what's wrong with it. And I'm a coach. I've been doing this a while. I kind of know what they're looking for from the district and the council, the scouting council, what they're looking for in the rules. They always want to hear that there's a first aid kit, they always want to see there's a map to the nearest hospital, language you use says your project always starts with I will lead a team of volunteers that will. And I'm sure as you're doing like everybody else, as soon as I'm describing this, I'm describing a skill and that's immediately what I thought was, well, why don't I just train Claude a Claude project and I'll just create this eagle coach and say, hey, I just reviewed this kid's proposal and I sent this detailed email back to them. There are a couple rules in there. One, it's like line Item numbered. And so it's like number one, you have typos in this box. And then number two, you miss this phone number here. Number three, hey, this looks great. Always be positive, right? These are 15 year old kids. And so it's not like, hey, I told you four times, you need to fix it. That doesn't help anybody. So I just dumped that email into Claude and I said, hey, I need you to review these things and with these rules and let's write a set of rules. And it said, sure, sounds great. And I got it to the point where I would start a new chat and I would upload a PDF and I would just say, review this. And I had it refined, all the rules and it would generate that email back and I just copied and paste the email, send it back to Scott. So something that took 20, 25 minutes took 12 seconds. And now all of a sudden I can handle 11 scouts as a Eagle coach because I don't have to read these documents. I mean, it was getting so good that we had a work session one night and all these scouts were kind of all working on the projects together. And I would see one of them would start screwing around. I'm like, hey, what are we doing over here? And they're like, oh, I'm done. Like, great, email it to me and we'll review it. Now typically I would say, okay, you're done for the night, thanks, you know, I'll talk to you next week or you know, I'll review this tomorrow or whatever. Instead I'm standing in the kitchen talking to their parent and you know, upload it, do it, review it, send it back. And I wasn't even looking at it anymore. It just got that good. And so I was able to do 15, 20 different reviews in the three hours that we were there. And most of the time the kids were just talking and going through those lists. And so that immediately was like, oh, okay, now I've got the skill trained up. So talk about that just a second, Steve, because I think we're approaching a point where people are starting to understand that there is some segmentation and there are ways to do this. Besides, I'm just typing into chatgpt and this happens to be a thread about scouting and it's recognizing that I'm talking a lot about proposals. You're talking about actually having a skill which is separate than just like a project or a thread that you've got going with an LLM, right? That's exactly right. Yeah, it's kind of a object permanence for a thread because I want it to repeat this thing over and over. And I think as more and more people are doing things that are repeating, I think they're going to start realizing, oh, well, this is something that I could have it just remember and just do every time. And that naturally lends itself to this skill concept. And you could kind of teach it and kind of tweak it as it goes. And I've already have other programmers that have sent me skills. I was talking to a gentleman that I worked with years ago, Greg. You remember Greg Aaron? Oh, yeah. I was talking to him the other night and I just happened to start texting him and he's like, oh, hey, I got a couple skills you need for root code review. I'm like, okay, thanks. Sends them over. They're like hyper detailed. They were fantastic and they were very specific. He's programming something in Rust. A couple things were Rust specific, but most of them were general. Overall, really great skills that you would Saw about half of them in Superpowers, the other plugin that I can't live without these days for PR reviews and other review processes. But it was like, oh, hey, I figured this out here. They're just quickly and easily sharing it with me. And I was like, wow, if I change three lines in here, it's kind of generic for anything. And now it's like repeatable and you can bring it into the system. So I've made an Eagle Scout project proposal skill that technically I could give to anybody and just hand it around and they can then become an Eagle Scout coach for proposal review with no effort. But this is also different from an actual application. There's not a, you know, some pretty, pretty interface that people go in and, you know, that sort of the thing. You're talking about something, though, that somebody would have to be in a CLAUDE environment, have installed, if you will, this skill, and then know how to leverage that. That's exactly right. It's like a power user, right? You've got a Claude power user and somebody's in Claude and they're using this thing and all of a sudden they're like, oh, I'm able to repeat this over and over. Well, you start running into limitations, right? Each chat inside Claude or in Codex actually have limits, because you have limits of files that you can put in there. You have limits of the context window, you know, a million tokens, all this jazz of the context. What I found was that I was dumping in multiple PDFs into a thread and at some point it's like, dude, I can't handle any more in here. So I was like, oh, well, why don't I just make one chat per scout? And the cool part was that it remembers the context of every time it does a review. And so it's like, oh, hey, this is the second time you've given me the proposal PDF. So it changed without me telling it. It said it changed the language and it says, well, I see that you fixed these four things from before. Great. All the phone numbers are now parentheses and a space and a dash. And that's great. And hey, you still got Prairie Home spelled wrong. And it would spell out Prairie and it's P dash R. It's like, that was kind of. I said, you can't be rude, stop being rude. And it's like. So you could tell it's kind of a passive in its language and what it does, like trying to literally spell it out for you here, but still without being rude. So it followed all of the rules with that. But you're right, you're empowering somebody. But they would have to have Claude running to make it work. And so I just took that to the next level and I'm like, well, let me just export that and take it over to an actual app. And so me coming from the background of building apps and understanding this entire process, I'm like, well, I can't be depending on Claude for everything. I'm just going to eat up money and API calls and this thing's going to just eat me alive. I mean, technically I could slap together an application and just put a box up, right? And just everything's through the box and it will spit out Word documents and spit out PDFs all day long. But that's not a really useful app. And that's not how humans work. That's how we have a chat with somebody or chat with a technical support crew. But if we have the capability to build screens and actually give somebody a visual that maps, in this case, maps back an actual PDF and actually a multi page PDF, they're like, oh, great, this is on page six. I understand page six. I can fill these things out and then we can keep going. Are you doing any bounding box work with that bounding box? Yeah. So we had a cool thing. I'm doing some computer vision stuff where we have some paper and handwritten documents that we're bringing into a platform and then we want to have a human review. And so I'm working with Opus through the context of Amazon on this. But basically we're saying, cool. So I want you to give a confidence rating to everything that you are bringing in. And anything that is dipping beneath 80% confidence, I want you to flag it, I want you to flag why you're not confident in it. And then I want to have a link next to where that piece of item is or where that item is that pulls up the page and then puts a bounding box around where that data came from on that page. So as humans are doing their review of the computer vision, they're scrolling through, they see that something is maybe not quite right and they need to review it. They click the button, it pops open the original document with a box around where we got that data from. As an idea, if you're not doing that yet, it's pretty cool. And this is all just asking the LLM to handle that stuff. Exactly. And the answer is yes. And what I'm doing it with is the signature pages. So again, when the kid fills out a fillable PDF, they usually don't. They can't manipulate the PDF and so it shows up to the reviewer as still as a fillable PDF. Certainly they could, you know, like the old trick, like, oh, I'm having problems with this thing. Well, why don't I print a PDF to a PDF and it flattens it out. Oh my God, of the trick days, man. Right, well, so if we get that document in, Justin, that's exactly what my next step is on version 3.3, I'm sure is in the pipeline here of me actually making that work, because after this discussion, I'm going to be doing that for a scout, because they had a problem with one of these online PDF tools and they stuffed it in there. And what they were trying to do is they wanted, they had some pictures of their proposal that they wanted to include, because right there in the proposal it says be sure to include pictures. Well, they wanted it just all be in one PDF, which is great. But the only way they could get it to work was to flatten that PDF. So now what? And then the signature page is interesting because my system that I built handles all digital signatures. And that's wonderful. But how do you onboard somebody that already has a piece of paper and already has wet signatures on paper and how do you import them? And it's exactly your what you just talked about, the bounding box. Oh, I see these signatures. Well, let me just snap this out of here and save it as an image and use that as the quote signature. And that's good enough for what we're doing here. These aren't really legally binding documents. I believe the first law is the Scout is trustworthy. Three fingers, right? That's right. It's three fingers. Sorry. That's right. Just like that. Two is Cub Scouts, and the band aid comes with it. I'll tell you where I got off the Scout train, because I thought it was two. So anyway, Cub Scouts is the peace sign, but now they've kind of changed it to that. And then you've got Scouting America and Scouts. PSA is three. Do you guys have an AI merit badge yet? There is multiple merit badges. There's a cyber. I was going to say cyber crime, but that's not right. There is a cyber security merit badge. There you go. Absolutely. There's a cybersecurity merit badge. There's a programming merit badge. And they now have an entire online concept where they're experimenting with new merit badges all the time. And so they've added an artificial and intelligence one into that. And the goal is that if people like it and they understand it, they get feedback from Scouts and they're like, you know, hey, maybe we should add this requirement or this one's weird, or whatever, and they'll refine it. And then all of a sudden, it's a real merit badge. And so there is an AI merit badge that is out there that you can earn right now and still more in the pipe. You know, it's very eclectic group. Right. There's like almost 140 merit badges out there. At the same time, we're pushing an AI one through this. We're pushing through a sewing merit badge. So that's kind of the range of merit badges that are going through the system right now. It's very cool. So keeping an eye on time, you got this now to a process you had talked about. You're now wanting to kind of turn it into an app, or you have partially turned it into an app. And you and I have talked a little bit about this, but I think our audience would also be interested in the aspect where you worked with the AI to generate a design, an actual workflow, and then you also just gave it to it to program because you don't do react. And so I'm curious how that process went. And was the end result good enough that since you didn't know react, it didn't matter, or did you have to, like, tap somebody and be like, is this garbage? Right. So I used plugins to ask it if it was garbage. And so I just went with Laravel because I Kind of came from a Zen framework world and I liked what Laravel was doing with PHP and I was actually started the whole project in Laravel 12 and that was part of the pile that I was working with. And lo and behold, bleeding edge, here we are. While I'm working on it, you know, Laravel 13 pops up. All of a sudden it's like, oh, hey. It's got all these agentic endpoints and it's got all of the support for models and just put a key in and all of these libraries and structures and classes are all just here for agentic programming. And I'm like, hey, can we use this? And it goes, yeah, sure. And I think I went to the bathroom and came back and it was added and I was like, what just happened? So could I get it to that point? Absolutely, I took it and I. Right now Claude Design is giving away their usage of that system. And so right now I feel like I'm cheating because I'm using Claude over here for building the system. But then I'm using Claude Design to build out all the user interface and the entire user interface. I'm chatting with this, it's making mockup, it's making a lot of visuals for me. But I know what I want, I know what the workflow looks like. I had actually started with that PDF workflow and so I already knew that, you know, I already understand the concepts of software. I need a user table, I need to be able to do these things. But there's other stuff that I didn't even think about. It's like, hey, you know, you really don't need to use passwords. I'm like, really? And they're like, yeah, you can just let people log in with Google. You want me to just add that? Sure. You want me to add login with Apple? Sure. Okay, great, done. And then we just kept going. And the system is live. It's up@myeaglecoach.com and I actually have actual Scouts using it right now. I had a functional system built in under 200 hours. And for me that's crazy. And I say 200 hours. That's actually a pretty feature rich system. At about 100 hours it was ready to start going pretty basic, but it was working. So this system now uploads and imports Eagle applications that have already been filled out. It uploads their entire. There's a system called Scoutbook, that's the official system that's out there. It will import all of their advanced, all of the merit badges. They've earned all of the ranks they've earned. It'll just import that PDF and dump it in. It'll import that workbook, like I said before, this 32 page workbook of fillable PDF. It will just dump all that data in. And then I made it, I took the official PDFs and I repopulate it so you can dump it all back out. So I'm putting in PDF, I'm pulling out PDFs. So technically, scouts can work through this thing without even using the system. But as soon as they see the system, they're like, oh, this is amazing. You know, I had a scout the other night, I'm like, hey, you can't edit this part until you sign the document that says, hey, I'm working on this document. It's kind of a gated area. And so she goes to it and it's a digital signature. And she's like, oh. And she starts using her touchpad to kind of sign. And she didn't like it. And I said, no, no, click there. And it showed her a QR code and she didn't read anything on the screen. She just immediately says, QR code. Wait, I'll scan it. And she goes, oh, here, I'll make, I can make. My mom does a giant first letter. So she's like signing the document right there. And she's like, cool, cool. And she's like, oh, this is so baller. Which I thought was awesome. And without me explaining the system at all, this scout, you know, this 14 year old, was able to understand how to do it and how to import and export this PDF. So I guess that's a testament to me being lazy and not wanting to explain things to people. But at the same time, if we're doing our job correctly, usability just comes with good design. So cloud design is fantastic. And now it's like if you have a vision and a little bit of expertise and a little bit of initiative and a healthy dose of curiosity, you can do crazy stuff right now. And this is really true within the last six months. I mean, with the cloud stack and other relevant like AI stacks, I mean, it's. If you're a guy who has an idea and can see a problem and can conceive of a solution, but just haven't had the resources or the time to dedicate to it. We're solving problems like that so fast these days, it's just, it's fantastic. I think the Scouting app you're talking about is a great example of that. Like, it's A problem. There wasn't really money in it, but it's a great service that people needed and had all the pieces I needed to get it solved. So I just took care of it. I was just going to say, yeah, even within, like, CLAUDE Design's environment, there's now just a gigantic button up in the upper right that basically just says hand to Claude code. And it just. That wasn't there three weeks ago. I know. That's what I'm saying. It's wild. Cause I just, I was looking at it the other day, I'm like, what is that? Oh, no. I thought it was interesting in CLAUDE Design that, you know, you could point it to the code base. And for me, system documenting is difficult. Right? I mean, code documenting, nobody wants to do test harnesses. That's kind of a solved problem. Now you just, in your claw MD file, you just say, everything must have a test and put it in all caps. And all of a sudden it just happens. I think I just passed 1,000 tests in this app. So everything has a test coverage 100%, which is great. You know, deployments take a little more time. But now I'm finding that I'm like, well, wait a minute, what's in this system? How does this look? You know, I'm so used to seeing, like a database diagram and a schema, and I just want to see what that looks like. And so I said, hey, I want you to spit out the schema and I want you to spit it out in a draw IO format so I can pull it up in a browser, a very popular tool. And it did that, but every time it did, it put it in a grid and it looked horrible, but it was all right. So I, you know, I spent like 15 minutes moving things around to where I liked how it visually looked. And then I, you know, of course we changed the schema and I said, okay, update it again. And it wiped everything out. And I'm like, wait a minute. Okay, I'm going to lay everything out the way I want and I'm going to change some stuff, but I'm not going to delete or add the tag. It's got some metadata inside those elements for those tables. I moved them where I wanted to. And then all I said was, okay, I've moved this stuff around, don't move it, and just keep going with your updates and up, add and remove elements from this as needed. You know, I've changed some stuff from integer to varchars and other stuff, and I've added new fields all over the place. And it remembers this. And of course, and now my diagrams are completely in sync. And so now I have a visual that is up to date with my tech stack. That's a good thing. The bad thing is that there's a lot of stuff that Claude is not good at. I mean, I'm finding the daisy chaining markdown files. It keeps referencing old residue of features that have already shipped, or ones that I already said, oh, no, we're not going to do that. We went in a direction initially we thought we were going to use like Docusel. I was going to use docusel as a server to do these electronic signatures. Turns out I didn't need the overhead of Docusel. You know, Docusel is the open source version of DocuSign. And I said, this thing is a pain in the butt and I don't want to spin up an entire separate server just to have this up. And it's like, oh, well, all you really need is this. And I can do that. Great, let's get rid of Docuseal. But as soon as I came up with the architecture diagram, I said, hey, Claude, design spit out a diagram that shows me the architecture. It had Docuseal in it. It had a Redis server in there. It had connections out to systems I don't even use. And I'm like, like, what is all this? And it's like, oh, yeah, I found that in the spec. I'm like, but I gave you the code and I wanted you to architecture of the code. It's like, oh, so in its mind, code is spec. It doesn't matter if it was in the code base or if it was just in its memories. It just built an entire architecture based on the full collective of what it had. Which I thought was kind of crazy, but I was like, okay, well, either way, get rid of that stuff and let's tighten that up. But that was extremely useful with what we had in there. The other part of that was, I think Justin and you guys keep beating this one into the ground. It's consistent development methodologies. It kind of changes its mind halfway through. It's a junior developer that's had too much Baja Blast, right? It's just sitting there and decides that, oh, today I'm going to follow the rules of PR generation and deployment. And other times I'm just shoving this as a hot fix out to production without any checks. I'm like, what are you doing? It's like, oh, yeah, sorry, that's my bad. My experience has actually been it's that It's a giant JavaScript enthusiast who's reluctantly accepted a job writing Ruby on Rails and then also has had way too much bourbon. Is there such a thing as too much bourbon? Sounds like the end result is about the same. I mean, the Balmer peak is a very tall and very narrow thing. And too much bourbon. At least the context of programming is when you end up on the right side of the Balmer peak. So I am fully familiar. So my seven day rollover with Claude is Tuesdays at 11pm so Tuesday nights I'm just down here burning capacity before that limit gets hit. And so yeah, the Balmer curve shows up a lot on Tuesday nights in my house, JavaScript in Woodford. It's going to be a good evening. That is the truth. Yes, exactly. And if you want to completely destroy your personal life, I would suggest you familiarize yourself with remote control in Claude. And then you just code everywhere, all the time. And your work life balance is completely destroyed because you're remote controlling claude. And as soon as you have an idea, it doesn't matter if you're driving or sitting on a bus or at lunch with your wife, you're like, oh wait, got an idea. And all of a sudden you're chatting but you're sending it back and updating some code base somewhere. I'm going to push back on that idea the same way that I pushed back on Apple watches, which is just like any invention that exists now. So I can spend all of my time staring at a tiny screen versus just most of it I don't need. I went outside, there were these things growing and like they're tall. Well, some of them were small and you could walk on, other ones were tall and you could climb. But I was out there for a while and I felt some of my anxiety fall away and I. I realized that my overclocked ape brain sometimes needs a minute or two out there in the. I think they call it nature. Anyway, what happens when you go outside, you know, when you order an Uber Eats, and then there's that uncomfortable moment between when you're at your front door and when you get your Uber Eats order, where you experience like weather and stuff. Like if you just lean into that a little bit longer, like it's actually really good for you. You mean like longer than taking out the trash? There's like a world past the end of the driveway, Is that what you're saying? If you get to the end of the driveway and you turn right and then you keep Walking, we call that traveling. And a lot of people like to do it. Yeah. Yeah, Interesting. That's kind of wild. Let me help you out with that a little bit. I could give you some hobbies on the other side that'll equally suck all of your time away. But yeah, that's a great way. But remote control doesn't work, so it's half the time with like slash commands. So you have to remember what the flash commands are or you have to over explain what you're trying to do, which kind of drives you crazy. So it's like, oh, well, maybe they'll fix this soon. But if I want you to do a review, great. But if I forget how to do to turn on the front end reviews, blah, blah, blah, plugin that is out there. I just can't remember the syntax. It says, I don't know what that is. I'm like, yeah, you do. And then you have to explain it further. It's like, oh, yeah, okay, I'll run that one. Like, ugh. So what's your favorite plugin or skill or tool that you're playing with right now? That's maybe new from a couple weeks ago. So superpowers. I mean, if you're not running superpowers, you're not using Claude. That's it. Beyond that, I found a couple very specific ones that have been helpful. I'm sticking to the Anthropic official library because I feel like what's going to happen with the marketplaces for plugins is going to be exactly what happened in the WordPress world. In that early on, a lot of official plugins showed up for WordPress from all the big players, and then everybody realized, oh, I can write my own plugin. And then it just became this no man's land of ridiculous plugins that aren't really coded well at all. Giant security holes. They eat up your CPU or server, stack, whatever, and they're just not good. Sorry, are you talking about WordPress or JavaScript? Right, I'm getting there. Let him cross the ridge. He's getting there. That's the WordPress problem, right? Okay. So now I'm like, what's this marketplace thing that Anthropic has out here? And there are all these official ones. And I think by the time I hopped in, there were 15 in the marketplace. And I said, oh, you sweet summer child. This is so cute that you have 15. This is going to become a nightmare as soon as everybody finds out about how this works. And sure enough, you guys have talked about this on a Couple other episodes where people install this plugin and I think it's going to do this cool thing and it just eats tokens right and left and you're like, what did it end up doing? And I think Nolan mentioned another one of these things. Can you just put in the hundred lines of code that this thing does instead of just recoding everything, for God's sake, Like it's like NPM packages. Like for God's sakes everybody, let's just go back to reality. But if you stick with the anthropic plugins that are official things work pretty well. The Superpower one PR reviews, I mean you're sticking with what it knows and it's really good at. And then I branched out a little further and I actually hosting the Scouting app on laravel Cloud and that's another system that I never used before. I have the entire stack up on that and managing through Nightwatch and Backpack and all these things that the laravel ecosystem manages without me. So I just installed the laravel best practices plugin and I installed the laravel Cloud plugins and the laravel CLI plugin and now this entire Claude thing is now very in tune with Laravel. And before when I was struggling with it doing things, now it just completely understands the laravel stack. And then the security is another one. You know, do a security audit for me, it's been very good. Hey, you know there's cross site scripting vulnerability here, but really without me touching the IDE ever, I'm not injecting any bad practices. Right. So if it put a security hole in there, it did it on its own doing so I'm not ham fisting my way through some functionality that I'm trying to code and oops, I didn't realize that I didn't sterilize my inputs. That's not happening because I'm not writing any of the code. And then the third one was accessibility. Super important for many reasons, but even for sighted individuals, contrast colors and stuff are difficult to see. I mean that's. Is that tan on orange? What are we doing? If I can't tell, then what's a screen reader or anybody else going to be able to do with low vision? And those plugins have been super helpful as well because I'm able to get Lighthouse scores of like 99 where before it was like 85. And a good friend of mine is a lawyer out of St. Louis and he's building an entire app. I could go on and on with what he's doing. He's a perfect example of a lawyer that understands how to do this and is just dangerous enough to build something. But he knows how to now find cases. And he just got me thinking because I'm like, in the state of Missouri, we have a huge problem with this lawyer and he has a client and they go around and they sue everybody because your website isn't accessible and it doesn't score a 99 on the lighthouse. Yeah, we did a podcast episode about these people, I guess would be the least divisive thing to say about them. It's a huge problem. Yeah. And this plugin got me to 99 and makes them go away within minutes. It was. It's very nice. And, you know, it keeps up, right? It keeps up with the standards, it keeps up with the things that I can't. And most of the time, you know, you're moving stuff around in the background and it's hidden tags and hidden things that. That, oops, I missed one or something. It's just, you know, these LLMs are really good at. Well, guys, we're probably past time at this point, but it's always just so much fun to get to chat about these sorts of things. And we all know each other well enough too, if we can just kind of BS with each other a lot. So it's been fun. Steve or Justin. What did I miss? Anything you want to wrap us up with that we haven't hit that won't spin us into another full episode. Oh, I thought we were going to talk about the future of sas. I had a whole like nine point bullet list of the future sas. You'll be back, dude. Yeah, every time Steve comes on, it's like, well, we got three more episodes queued up, so. Always good to see you, my friend. Happy to keep babbling, but I think a couple of my clods here are finished, so I need to get back to pushing the next version. So go babysit some more. All right, that's fair. Thanks for taking the remote control off long enough to be on the podcast. We appreciate you. How do you know what's off? He's got it over here on the screen. Clicking. You know it's happening. We're good, bro. Don't worry about it. We're. I'm very well, Steve, thanks so much. My friend Justin, always a pleasure and thank you all out there. Appreciate everybody taking the time to watch the episode. Hit, subscribe, Hit like, let us know what you think. Anything else you want to see, send it our way and we will catch everybody next time on Sas that App. Thanks for cruising along with us on Sas that App. We hope you grabbed some insights that were inspiring, actionable, or at least entertaining. If you enjoyed the show, don't forget to subscribe and leave a review until next time. Keep building, keep growing and keep those apps sassy.