AI Meet Up — November 4, 2026 — Provo, Utah Full transcript (machine transcribed from the recording, lightly corrected; small errors remain) === Aaron Earnest — Agenda Insider (online) === [0:00:00] that we'll get started. [0:00:02] Aaron Earnest is online. [0:00:04] Aaron, we're going to turn the time over to you. [0:00:06] You're going to be our first presenter. [0:00:08] And then up on deck, Perry. [0:00:12] You're on deck. [0:00:13] And did you already go over how we're going to... [0:00:17] Do any of the presenters have a question on how to present, [0:00:20] how you're going to do that? [0:00:22] Yeah, I'm wondering if you see my... [0:00:24] Can you see my screen and can you hear me? [0:00:30] Can you see my screen and can you hear me? [0:00:35] I can see and hear you. [0:00:40] Okay. [0:00:42] In the room, should I start? [0:00:45] Yep, you're good to start. [0:00:49] Okay, thank you. [0:00:51] Okay. [0:00:52] Per the instructions, 60 to 90 seconds of intro. [0:00:55] I'm timing myself. [0:00:56] I'll be as succinct as I can. [0:00:59] About a year ago, my friend and I were talking about permits. [0:01:03] So everything that needs to be built has to get a permit. [0:01:08] And I had a call center back when I was in college, and we used to sell a permit tracking software. [0:01:16] Most of the customers would ask us for pre-permit signals. [0:01:21] and we didn't know if it was possible because human labor it would have been impossible to [0:01:26] to gather all the pre-permit agendas for all the planning commission and design review committee [0:01:32] and city councils but with ai we thought maybe it's possible so for the last year i've been [0:01:39] sitting at a desk probably 50 hours a week managing a swarm of ais and i've had a blast with it [0:01:45] I've used VS Code. Some of you might know what that is, but it's a way to basically mimic what a software engineer would do. [0:01:54] And so now we've got this library of pre-permits and we read every agenda. [0:02:02] AI extracts the signals and then people who are interested are tracking what leads up to a permit. [0:02:10] And that's my minute and 20 long intro. [0:02:15] Any questions? [0:02:28] wait can you hear us now yeah i can hear you yeah yeah i can hear you how do we get the echo off [0:02:49] that okay so this this is for land development right you're a land developer and you're finding [0:02:57] deals yes yeah it's it's i use it as an investor i've got a title company friend who uses it to [0:03:06] contact all of the adus and the one-off transactions so that they can get more [0:03:12] closings at their title company it's it's kind of a catch-all for everything that happens at the city [0:03:19] so it's a sneak peek behind the curtain [0:03:22] so are you selling this product like are you selling this as something someone can use or [0:03:29] do you use this for yourself both not to this room i'm not selling i'm i'm following the rules matt [0:03:35] no no no yeah i know i know i know i'm not trying to put you on as well it just looks professional [0:03:40] enough that you could sell it yeah and the reason i thought it would be fun to present here is uh [0:03:47] I am an English major. [0:03:50] I have no experience vibe coding. [0:03:52] So the fact that I can talk to AI the way I'm talking to you [0:03:56] and get it to construct this database [0:04:01] is what I thought was worth sharing. [0:04:03] So any cool idea that you have, [0:04:05] sit down and start talking to your AI model [0:04:08] and voila, out comes this weird thing [0:04:12] that you didn't realize you were capable of [0:04:14] because without AI, you're not. [0:04:16] And I wasn't. [0:04:17] maybe talk a little bit more unless anybody else has questions you want to let me pass it around [0:04:23] but um for those in the room that have not created their own ai what was the dot that connected for [0:04:33] you that made that possible it was um an app called visual studio code and and um we call it [0:04:43] VS, Victor, Visual Studio, VS Code. [0:04:47] And so you can plug your VS Code app on your MacBook Pro [0:04:52] to all of the AI models that are the latest and greatest. [0:04:57] And the VS Code will then write code on your computer [0:05:02] and it will help you sign up for a domain [0:05:06] and it will guide you on how to write code [0:05:10] to create the art and you know this presentation that you're looking at and [0:05:15] I'm happy to walk anybody through it who wants to try something cool like this [0:05:19] and that's what I thought would be worth sharing is it's it's not hard it does [0:05:25] take a few steps that somebody can hold your hand and teach you how to do which [0:05:29] I'm happy to do to anybody who wants to talk about it I love it any other [0:05:37] questions? Another question here? It's Matt, right? I'm Aaron. Sorry. Sorry about that. Hey, [0:05:47] Aaron. My name's Trent. I'm curious. I don't have much of a, you know, real estate's a world that's [0:05:55] new to me in many ways. Yeah. I know the basics like everybody else, but what was it like to, [0:06:01] I mean, ultimately, what's the data base like? [0:06:07] How did you get and train your model, at least specifically here, [0:06:12] for the real estate-specific details? [0:06:16] Say that one more time. I don't think I follow the question. [0:06:20] So I guess my question is, with the AI here, [0:06:24] did you look to just build this website? [0:06:27] because you mentioned something about a pre [0:06:31] I'm forgetting the word but a pre something signal [0:06:36] that you were based on prior experience and just [0:06:40] what's changed now that you said it was a lot of manual effort before [0:06:43] something like that wouldn't be possible so what's now [0:06:47] changed to enable something like that to [0:06:50] be captured what data sources are you looking at to create [0:06:55] that that signal very good question uh i'll give you one example i have a friend who likes the data [0:07:03] that i'm gathering because you know that new off-ramp in springville along the 15 he he and i [0:07:10] talked about buying some dirt along the 15 freeway like seven eight years ago and we passed on it [0:07:17] we did not buy but now the same dirt is 20x in value and so he he laments the fact that just had [0:07:25] he known that there was a a project for that off-ramp in the works which this website and this [0:07:34] data enables so my ai so in other words if i had kids like back in the day going around to all of [0:07:41] the city offices gathering permits that's a pretty brainless manual job but but what those human [0:07:50] laborers were not going to be very good at was sitting there long enough to ask the city personnel [0:07:56] to give them all of the applications the design review committees the petitions for zone changes [0:08:04] but ai can go to the public websites of those cities and so we've built code instructing the ai [0:08:14] to go to each of the websites for the municipalities and read all the PDFs [0:08:20] thousands and thousands of pages of PDFs in each city that now AI could do in a [0:08:27] moment where it would have almost been impossible to assign runners to do that [0:08:33] kind of sifting and sorting so it's AI that makes it possible got it so you're [0:08:39] You're scraping those websites for that data. [0:08:43] Is there already a commercially available data set like this? [0:08:49] Or is it pretty, I mean, municipalities are pretty, you know, one of a kind. [0:08:54] Everyone operates slightly different. [0:08:55] So I figured that that might be hard. [0:08:57] But I'm just curious because, again, it's a world I'm not familiar with. [0:09:02] Yeah, that's a really good question, too. [0:09:04] we started in November of [0:09:06] 2025 with AI and we were [0:09:09] the first to this segment [0:09:11] of the industry [0:09:11] since then two other big players [0:09:15] are in our space [0:09:16] kind of like our competition [0:09:18] but it's all because AI [0:09:20] makes it possible [0:09:21] awesome [0:09:24] thanks Aaron [0:09:25] thanks Trent [0:09:26] Aaron you're the man [0:09:29] thanks Matt [0:09:30] I'll stop sharing now [0:09:32] Give it up for Aaron. [0:09:36] Good stuff. [0:09:37] Thanks, guys. [0:09:39] All right, we're going to have to have you in person next time. [0:09:42] Yes. [0:09:43] If I wasn't out of town, I'd be there. [0:09:46] Next time. [0:09:48] Yep. [0:09:48] All right, we're up, Perry. [0:09:50] Come on up. [0:09:50] You're next. [0:09:51] While Perry's getting ready, be sure to get some treats, guys. [0:09:55] We don't want to take any of that home. [0:09:56] So help yourself the whole back wall. [0:09:59] Yeah. [0:10:00] yes but we can give it to you again [0:10:04] yeah [0:10:04] hey was that cool any cool takeaways [0:10:10] what were you thinking as you [0:10:12] watched Aaron that has an immediate [0:10:14] application to what I do I'm like I'm [0:10:16] going to be reaching out to Aaron you [0:10:17] got a sale out of that [0:10:19] no selling [0:10:24] no [0:10:28] No, it's IT. [0:10:29] Yeah. [0:10:30] Who feels like they relate with that? [0:10:32] Like no IT background, now they're like fully [0:10:34] like app developers. [0:10:38] It's okay. [0:10:39] All right, Perry. [0:10:40] Perry, does it give you an option to share screen? [0:10:42] Yes. [0:10:42] Okay. === Perry Sessions — personal apps, Codex, Chrome extensions === [0:10:56] All right. [0:10:58] Can you hear me? [0:11:00] Yep, they can hear you. [0:11:02] Okay. [0:11:02] All right. [0:11:03] Hello, everybody. [0:11:04] I'm Perry Sessions. [0:11:05] I'm related to the Sumsions through my wife, so Matt and uncles and cousins over there. [0:11:12] So I have known of them, and I read some of the notes on the last meeting, but it's good to be here on this one. [0:11:19] As far as background for myself, I do marketing, basically. [0:11:24] I'm a co-founder of a small agency that does a lot of online advertising and I'm a coding [0:11:30] enthusiast. [0:11:32] Definitely. [0:11:32] I taught myself how to code and now AI comes and I'm loving the vibe coding scene. [0:11:37] It's very fun to me and also a little bit depressing when I think about some of the [0:11:41] projects how long I took on them compared to how fast they go now, but it's all positive. [0:11:45] So I do a lot of hobby stuff. [0:11:46] I definitely use this for my work, but I just wanted to share, I think what would be fun [0:11:52] if I was in your shoes or what I'd love to hear from you guys. [0:11:54] I was talking to Jeremy beforehand, projects that I thought are fun. [0:11:58] I do a lot of these things just for self-help and I do use it for work. [0:12:02] So is it presenting? [0:12:03] Is it? [0:12:03] Okay. [0:12:04] So yeah, that's just a little bit about me. [0:12:07] I have four girls, so you'll hear me or see me show some projects and stuff that I did. [0:12:11] If you have any questions about them or what I did, I kind of want to give like projects [0:12:15] I've done and then why I use them and kind of how you use them at work. [0:12:18] Hopefully you can use similar kind of ideas and ask any questions and then tips and tricks [0:12:22] that I've found that I wish someone would told me about AI. [0:12:26] So a little bit of a fire hydrant. [0:12:28] So I probably have more than that. [0:12:29] Eight minutes, just I'll try to shorten it down. [0:12:31] So let me know. [0:12:32] Stop me in the middle. [0:12:32] Are you still sharing the screen? [0:12:35] I need to show it back. [0:12:36] Yes. [0:12:36] Yes. [0:12:37] Sorry about that. [0:12:42] Your time is paused. [0:12:43] and then could you maybe okay I'm going to stop sharing now you reshare okay there's probably [0:12:57] something someone come see me after [0:13:02] so what is it I may have taken you off so you may need to reattempt to screen share [0:13:12] It's saying sharing now. [0:13:15] Does anyone else see it? [0:13:16] Is it just how he's... [0:13:18] Yeah, the online people, we can see it. [0:13:22] Yeah, we can see Jim Buddy. [0:13:24] Yes. [0:13:26] What's the name under it? [0:13:29] Should just be Perry. [0:13:30] Perry Sessions. [0:13:34] As he's pulling it up, I can tell more notes [0:13:36] until he gets to the visuals. [0:13:38] but the [0:13:40] I just yeah [0:13:42] the first one here I pulled up I'm a small guy [0:13:44] I'm trying to do gym for like a year [0:13:46] here so I don't know you guys [0:13:48] some people are really this is really divided [0:13:50] some people take notes when they're at the gym some don't [0:13:52] so I just I love notes I had like a year [0:13:54] worth of Vasa gym notes [0:13:56] on notes app on my iPhone and [0:13:58] nothing out there on the apps there's a million [0:14:00] free apps for tracking things but nothing [0:14:02] could import it so I made this just so that I [0:14:04] could import it off of the notes I just copy [0:14:06] pasted and told AI so that was [0:14:08] fantastic for it to just pull it in. And then I wanted information shared on like, okay, where did [0:14:14] all this, you know, put it visually for me. So my PRs and my, you know, my weight gaining or my [0:14:20] workouts. And I did some correlations between different stuff. So how often I visit and I just [0:14:25] really like the analytics to see, you know, how am I improving? Like my weight goes up, obviously, [0:14:29] I'm trying to gain weight as a skinny guy. And this is helping me to say, oh, this, you know, [0:14:33] nutrition I added recently and I can add in nutrition. I actually embedded Gemini API into [0:14:38] this. This isn't something trying to sell. Again, this is a personal app. I was talking again to [0:14:42] Jeremy. This is very nice. It doesn't cost me anything. There's ways to get, if you're looking [0:14:47] for free things, AI will tell you similar stuff. But Vercel is a name of free hosting and Supabase [0:14:53] lets you do backend free authentication. So I can log into this. It's public. I can access it on my [0:14:59] phone. You can make anything that you do with Claude or I use Codex and I can tell more about [0:15:05] that in a second. You can make it mobile friendly. And so this worked really well. This was a fun [0:15:11] app and a great project that I did. And then another one was finance. I tried Mint [0:15:16] or not Mint, Mint's financing app and my wife and I could never figure anything out. [0:15:21] And again, this was another one of those cases where I was like, I'm going to import [0:15:24] a bunch of bank statements into my data here. [0:15:28] I want to know what my spammy habits are [0:15:29] and categorize them. [0:15:30] And Mint doesn't have a, [0:15:31] it didn't have a very robust way. [0:15:33] So I literally could, [0:15:34] I look at all my transactions. [0:15:35] I've put in Google API is just the best free one. [0:15:39] So I connected that again here [0:15:41] so I can ask it my finances, [0:15:42] which I don't know if you've used ChatGPT. [0:15:44] They made this now built in to the app [0:15:47] where you can have your finances. [0:15:48] You can pull your, [0:15:49] you can use a Plaid API and it can tell you, [0:15:52] oh, this is how much you spent, [0:15:53] how much you've done. [0:15:54] So basically, we made this redundant a little bit [0:15:57] in the ChatGPT app, mobile app. [0:15:59] I recommend you do that. [0:16:01] There's kind of different, [0:16:02] there's a whole different conversation [0:16:03] about privacy and things like that, [0:16:04] but I'm all about, you know, [0:16:05] I think that this is good. [0:16:06] It's a good API to connect and get your data. [0:16:09] But so this, the big things I just wanted to share [0:16:13] with these kind of apps that are very customizable to you. [0:16:17] Again, I don't have any desires to share [0:16:19] or to, you know, sell these things, [0:16:21] but they're very useful for my day-to-day life. [0:16:26] So other ones, [0:16:27] that's something that I think I've seen [0:16:29] the most people do in the marketing space. [0:16:31] One of the biggest ones has been, [0:16:33] I mean, this is just one website, [0:16:34] but what was another one? [0:16:36] I had it for a family friend that we did [0:16:38] was from website migrations and website creation [0:16:42] and then website maintenance. [0:16:44] So if you want a live website, [0:16:46] like the one I was showing you before, [0:16:48] I made a website off of a subdomain [0:16:50] of my main website and I can use again cloud Cloudflare is a free hosting [0:16:56] service and this is just vanilla HTML and JavaScript so very fast very quick I [0:17:02] think it uses a little bit of tailwind CSS for some for the for the CSS or the [0:17:08] styling and that's probably as heavy as you get right it's just and then you can [0:17:13] make yourself a website anytime you have a question you you know oh I want to [0:17:16] update this I want to make a banner for a sale I want to add another page you [0:17:19] just go to AI and say, hey, go do this. And it uses, I use two as well in the middle of this, [0:17:24] GitHub, which is again, another free place that you connect. It's kind of like a plugin that [0:17:28] Aaron was talking about. You know, you can connect plugins. He was connecting VS Code. [0:17:34] You can connect directly to GitHub. You're logging to GitHub repository. You have a new feature, [0:17:38] a new page you want done. Tell it to do it. Like I said, I want this image on the logo [0:17:42] for this website to be updated. And it's, you know, it's done. So one thing I will say, [0:17:48] I was talking with Jeremy and as far as recommendations go, I really have enjoyed Codex as far as [0:17:53] over Claude, mainly because of the image integration. Claude, I think Codex, Claude [0:17:59] is great. It's fantastic. It's smart. And I used it first, but if I was going to make [0:18:04] my decision from my experience, it's just the Codex and ChatGPT image generation has been [0:18:10] better than Nano Banana or the other AIs I've used. It's much more accurate. It's better. I use it for [0:18:16] like a lot of products, I'll have like a product sitting on a shelf and I don't have a picture here, [0:18:20] but I say, I want a new label. You can give it the new label and say, replace it. And it does a great [0:18:24] job where a lot of times images you can tell are over, you know, too many artifacts where the [0:18:29] texts and the labels are off, but it can do a really good job compared to other. And Claude [0:18:34] doesn't even have that. It only has like vectors that it can make for you. So I recommend that for [0:18:40] a lot of the, a lot of times I just need images and it can help me create those or customize those [0:18:43] or manipulate them a lot better than Claude does. [0:18:47] How am I looking at time? [0:18:48] I didn't look beforehand. [0:18:49] Bud, we should get two questions soon and wrap up. [0:18:54] Just to wrap up, I mean, I had other things that you can do with this. [0:19:00] I made an ambassador saying this is something free or something that you can't. [0:19:05] There's a lot of things you can't do. [0:19:06] Like Instagram doesn't have an AI that you can connect into that. [0:19:10] and you can't really scrape or I don't know. [0:19:14] It's very sensitive about things [0:19:15] like sending out a bunch of messages, [0:19:17] but I found out you can make Chrome extensions [0:19:20] are a really good use case to get you to work faster. [0:19:23] So instead of scrolling through all of these posts [0:19:25] and clicking into them, looking at the person [0:19:28] and seeing how many followers they have, [0:19:29] I found a way to make an extension that does that for me. [0:19:33] It looks through all of these, [0:19:34] finds all of them over 500,000 followers, [0:19:37] all of them that have something in the bio, [0:19:38] maybe they don't, you know, they're from this correct country. [0:19:41] You can just, you can do anything. [0:19:42] Sky's the limit, but this will, you push a button [0:19:44] and it will kind of do it on your computer for you. [0:19:46] Open the tabs, close them, and it just makes it so much faster. [0:19:50] So let me get a good example of here. [0:19:53] I could say resume review of people that it's found. [0:19:56] It will pull up a new tab. [0:19:57] It's like, do I want to keep this one or not? [0:19:58] I'll say, I'll keep it. [0:19:59] And it goes to the next one. [0:20:00] I can say whether I want to use this person. [0:20:02] This is for a client that does, you know, [0:20:03] sells kids cereals for babies, you know. [0:20:06] I look through to make sure they have a kid. [0:20:07] but it's just, it's Chrome extensions. [0:20:10] I just, if I could put it in a word there for that, [0:20:12] it's just a highly recommended quick and easy tool [0:20:14] that I've just been starting to get into. [0:20:16] So you're doing that to know? [0:20:19] This is for people who want, [0:20:21] companies that need brand ambassadors. [0:20:23] So they want to pay someone money to post for them [0:20:26] or to connect and do ads in their behalf. [0:20:29] And I want them to have certain following counts [0:20:32] and to have certain things in the bio. [0:20:34] This helps me go sort through all the thousands of posts. [0:20:37] Like I searched natural baby cereals. [0:20:39] And so that was just a good use case [0:20:41] as far as finding people. [0:20:43] And I have it, I can export it then to here [0:20:45] and I can have AI help me make messages [0:20:47] and then I have to go over and personally send them, et cetera. [0:20:49] But last thing before we go to questions is, [0:20:52] I just wanna, this is a random thing, [0:20:54] is the dictation tool on ChatGPT, way better. [0:20:58] I don't know if you know what that means, [0:20:59] but there's a bunch of microphones [0:21:01] when you're using Claude or ChatGPT as far as the app. [0:21:05] And I think you can get up to 10 minutes of dictation to text [0:21:08] where you can just hit this and it talks, [0:21:10] and it's just so much better. [0:21:11] Like I'm using a lot of jargon and lingo, [0:21:13] like ROAS or whatever, you know, lifetime value and things, [0:21:16] and it just doesn't pick it up a lot of times [0:21:18] with Mac text-to-speech or speech-to-text. [0:21:20] But this dictation, smart dictation from ChatGPT, [0:21:24] and a lot of them have that Google has their own too, [0:21:26] works a lot better. [0:21:27] So anyways, questions or thoughts? [0:21:29] I'm sorry, there's a lot of fire hose of information. [0:21:32] I think we have time for one or two questions. [0:21:36] Yeah. [0:21:37] How did you link your account to the chat so that it gives you all that data? [0:21:43] I'm really curious on how to get that. [0:21:46] It is using Plaid API, which is, there's a free, you can connect, I think, [0:21:51] up to five of your personal accounts for free. [0:21:54] And it's really, I think it's pretty cheap if you do more. [0:21:56] But Plaid, it's a very highly used, very trusted source. [0:22:02] So it'll connect the API of your banks and help you pull that information. [0:22:09] For the influencer, you find specific influencers in the affiliate. [0:22:14] What app are you using to pull that? [0:22:17] How do you explain it out? [0:22:18] So I used Codex and Claude to make a Chrome extension that the only API is really used in here. [0:22:28] I think I used Gemini API connected. [0:22:31] I mean, again, you can have a free amount of tokens to use with Gemini. [0:22:34] I really recommend that, too. [0:22:35] That's been really fun to do is integrate. [0:22:37] You make an app. [0:22:38] I think, Aaron, you talked about yours. [0:22:40] You can grab Gemini's API and throw it into there and have – [0:22:45] Like this, here's a good example of this one. [0:22:47] Gym Buddies chat is basically an AI inside of my app. [0:22:52] So it knows all my, I can give it goals too. [0:22:54] I told it my goals and then it can, [0:22:56] every time I go to post or ask a question, it references. [0:22:59] So that, you talk about training in AI, [0:23:01] like I like to train it on my personal goals. [0:23:03] I recommend that, but yeah, this was the answer question [0:23:06] was just a Chrome extension connected. [0:23:09] And there's not really anything else other than I told it, [0:23:13] know what i wanted as far as um preferences and things that's amazing i would probably have [0:23:20] more questions but i don't want to take a look i'll talk to you okay thank you [0:23:25] i think we're ready for the next presenter yeah thanks very thanks [0:23:32] okay all right jared you're up and then 14 years on jerry [0:23:36] I should give you a heads up. [0:23:43] Hey, while Jared is getting ready, the industry here needs, who is working on something, who [0:23:56] is stumped, who wants to get GrokBot and Claude Code together? === Open floor — needs in the room === [0:24:02] and there's like a change of what what things would be interesting to stand here [0:24:08] like there's any needs or what yeah Tesla users how they integrating [0:24:13] GrokBot with their voice saying like you know they got up to where it's yeah [0:24:18] just integrated that so grokbot yeah saying like update the spreadsheet for [0:24:23] me update my calendar I mean everybody does that with it with a chat or if [0:24:30] favorites that that with your Tesla do you drive Tesla yeah and to drive the [0:24:35] Tesla and there's using that people I've seen it I'm very sick like basically you [0:24:41] have a bunch of graph laws yeah deploying those graph on your command [0:24:48] I think you just have a login, like you have some of a GrokBot set up, make sure you have a specific set up, and then on your Tesla login and put that some music. And then when you talk to GrokBot, you can talk to GrokBot. [0:25:02] I think I can think any questions that I want to just kind of hear because after we're done [0:25:10] there's going to be some time so that you can talk to the network and it's just we got so much [0:25:15] there's so much knowledge in this room if you find someone that's got your questions or [0:25:22] don't help them yeah I've got my system set up where all my knowledge base is posted in Google [0:25:29] right and then I got GrokBot, Claude you at instinct all connected so my [0:25:35] skills are right and so they can communicate that's important so today I [0:25:39] had it I don't know what instant there but I told instinct hey go tell do it do [0:25:44] this thing and it went through and said and I can track all in people right [0:25:48] I don't have AI talking to each other. [0:25:58] All right, cool. === Jared Fenn — give your AI a brain === [0:26:03] Okay. [0:26:05] My name is Jared. [0:26:10] I've been a home inspector for 20 years, but prior to that, I spent a decade. [0:26:14] I actually managed Nu Skin's first web team. [0:26:18] So I have like way before vibe coding [0:26:22] when we had to build web pages by actually hard coding them. [0:26:25] So I love tech and AI just gave me a whole new opportunity [0:26:30] to get back into things. [0:26:31] And so in January, I started exploring a lot with agents. [0:26:38] I played around with OpenClaw for quite a while. [0:26:40] But the whole idea of building a personal assistant, [0:26:43] things have skyrocketed in the last six months if you've played around with Muse or with GrokBot [0:26:48] or things like that so so my concept kind of what I want to try to figure out how to present in [0:26:56] 90 seconds is is kind of a different concept of how to approach AI so a lot of this is going to [0:27:03] get very technical but if you're just starting if you kind of start from this foundation it might [0:27:10] give you a leg up six months from now. So everybody, we've heard a lot of, you have Gemini, [0:27:20] Grokbot, Muse, there's all these different models. Everybody's doing different models. And [0:27:24] how can you have one person building crazy things, they're 400Xing what they're able to do. [0:27:31] And then you have someone else that's using the same model and maybe they cleaned out their email [0:27:36] inbox, you know, and the whole idea is that it's, you get out of it what you put into it. And so [0:27:42] if you don't, if your AI is relying completely on your brain, then it can only grow and go as fast [0:27:49] as your brain can keep up. And when you go take a minute to go eat or sleep or you're nine to five [0:27:57] or whatever it is, it's, it's waiting for more instructions. So that's kind of the whole concept [0:28:02] behind kind of my approach to things. [0:28:07] How many have used Fiverr? [0:28:09] Or, I mean, there's a few other type that. [0:28:11] So Fiverr, basically, hey, I need to build this flyer. [0:28:14] You go and find someone on this hub. [0:28:16] They can go and pull it all in. [0:28:18] You give them the whole scope. [0:28:20] They give you a price. [0:28:21] I want you to go build it. [0:28:23] That's how 90 plus percent of people are using AI. [0:28:27] I have this thing. [0:28:28] Go and do it for me. [0:28:29] Great. [0:28:30] The next time I have a thing, [0:28:31] I'm going to go and scope it out and give it to it. [0:28:35] I've hired many employees over the last five, six years. [0:28:38] I've hired half a dozen virtual assistants in the Philippines. [0:28:42] They work great because I'm kind of getting the pricing of a Fiverr, but I'm getting a full-time staff. [0:28:49] And so I have a VA that I can train that's really a full-time employee. [0:28:53] So I took that concept earlier this year and said, okay, I'm going to build my agent from that concept. [0:28:58] I'm not giving it a task. I'm hiring an employee and I want to build that employee around not just a job, but around me. [0:29:07] So I called my agent Atlas. So I started building Atlas in January. [0:29:13] It took multiple attempts on OpenClaw. I eventually moved everything, most everything I've done in Claude. [0:29:19] But I now have an employee that takes care of everything. [0:29:22] So I use a Plaud recorder. So it's recording every conversation that I'm having. [0:29:26] All of my calls and texts, I use Quo. [0:29:29] If you haven't looked into Quo as a business system, [0:29:32] it is phenomenal. [0:29:34] Just a quick tangent. [0:29:36] The last night I had Atlas, someone gave me an idea, [0:29:39] it was an Alex Hormozi newsletter. [0:29:41] And I'm like, hey, that's a great idea. [0:29:43] Quo lets you transcribe and integrate through AI [0:29:47] all of your texts and all your phone calls. [0:29:49] And I've been on it since May. [0:29:50] So I went and pulled about 8,000 transcripts from calls, [0:29:54] everything from clients. [0:29:55] I matched that with who actually did work and who actually didn't and looked for patterns of how I could figure out what people say on the call that ends up not booking with us or people that. [0:30:08] And all of that Atlas did while I was doing other meetings. [0:30:12] And now I have a full he rewrote our quote page. [0:30:15] He rewrote all of that because I have an employee that's doing this stuff for me because he understands he has my second brain. [0:30:24] Quo. Q-U-O. [0:30:25] amazing phone service. I, yeah, I'm, yeah, whole nother tangent there. We can go on there. [0:30:33] So this is, so if you think of hiring employees, this is my org chart. This is my AI org chart. [0:30:41] It's probably hard to see from where you guys are, but Atlas is my chief of staff. [0:30:45] I've used kind of the Greek mythology. So a lot of my agents, I give them a name. So it kind of [0:30:49] helps me remember, okay, this is the scope of what they are and who they are. I recently came across [0:30:55] a bunch of stuff from Garry Tan. Go research him. He's big. He has a ton of open source stuff on [0:31:01] this second brain concept. You can go start watching a lot of videos, but how he explains [0:31:06] it is every skill is kind of like a job description. So if I have a job description for an employee [0:31:12] that does a certain thing, if anything like that needs to be done, well, you would go to that [0:31:16] employee that does that work, right? You wouldn't go and, hey, will you go and draft this up when [0:31:20] this guy that's his job is to draft things so the idea is you have these [0:31:23] skills that define certain work and you have bots or agents that understand the [0:31:30] realm or the scope of what they do the thing is is I didn't have to create all [0:31:34] these I only work through Atlas Atlas then okay hey we need to be able to have [0:31:40] someone that does most of the coding great now we need someone to check the [0:31:44] builders coding so that it's not getting messed up so and and and challenging it [0:31:50] And so over time, Atlas basically is hiring other agents to figure things out. [0:31:57] So in kind of preparing for this, I said, Atlas, I need a visual way to explain my brain. [0:32:09] I've used the Obsidian database for a long time, and Obsidian has some of those things. [0:32:13] But I'll just kind of show you. [0:32:17] so alice said okay here is a visual representation of the brain and now how can i share this screen [0:32:25] yeah [0:32:28] is that tedious sometimes to like remember like is a is a do you have it set up so that it's [0:32:38] reporting everything without you having to press uh it's i'm wearing it it's on it has a five limit [0:32:46] five-hour limit and then it starts a new recording if I remember and like okay I just got done with [0:32:51] that meaning I'm going to stop it so it kind of keeps it but Atlas doesn't care he pulls them all [0:32:55] every night parses through it sorts it and puts it into this file that if I can figure out how to [0:33:00] share my screen again it's Plaud so Plaud has like a really nice one looks like that that's the more [0:33:09] common one. This is their [0:33:11] P-L-A-U-D. [0:33:15] This is their pendant one. [0:33:17] It comes with a little clip. I wear it so [0:33:19] it looks like a Fitbit [0:33:21] and then I don't get people questioning. [0:33:26] It's on all the time. [0:33:28] Let me see if I can share [0:33:30] my brain. [0:33:32] Do you ever feel like there's a gap [0:33:33] there with your conversations [0:33:35] where you're like, [0:33:36] It has, but it's captured a lot more than it hasn't. [0:33:48] So, all right, I'm not sure why. [0:33:53] How do I choose what to share? [0:33:56] There we go. [0:33:58] Because that almost seems like it's like the key. [0:34:01] If you can just take what your outside world is experiencing and then give AI more of that, it would make its suggestions more. [0:34:09] So and that was actually where a lot of this started is I'd go to meetings and I'd talk to people and I'd be like, what was their name? [0:34:16] I can't remember. And then I'd walk into another place and I couldn't keep track of all of that. [0:34:21] And so that's actually one of the first use cases that I started building in. [0:34:26] So this is I don't know if this is going to work on. [0:34:28] it's you can't really see it the resolution but this is this is a representation there's about [0:34:35] 18 000 dots here each dot represents a file um and then they connect so just to show you an example [0:34:44] let's see i was just looking at one earlier um you have this dot right here which is called the [0:34:54] YouTube hub. So one of the things is I don't have time to watch all the cool videos and things that [0:34:59] are coming out because everything is posted on YouTube now. So I created a thing where Atlas, [0:35:03] I said, look, these are all the people I want to follow. You just watch it. And if anyone posts [0:35:07] something, I want you to pull down the video. I want you to transcribe it and then tell me if [0:35:10] there's anything interesting that we should know about. So every morning I say, Hey, here's all [0:35:13] the, it says, Hey, here's all these people that posted this. And I think we should use this. I [0:35:17] think we should look into that. So all of, let's see where to go. So that was, uh, you can kind of [0:35:23] see that dot. So that's my YouTube hub. This dot right over here is our AI best practices. So a [0:35:30] bunch of those people that I follow, they are posting on AI and things that they're doing. [0:35:34] So it's pulling out those and those are things that it said. And so everything is tied together. [0:35:39] I have another process that runs at night and says, okay, let's find connections. Well, in this [0:35:43] meeting, we're talking about this person over here. We talked about that person. It'll ask me, [0:35:47] is this the same person? And it'll start making, and that's what all these lines are. [0:35:50] the green are things about me, personal things that I'm like mulling over and trying to figure [0:35:56] out. There's all the pink is my, my family. So I each of my kids, um, anyway, we can go on and on, [0:36:04] but that's the gist of create the brain. And now I don't even, I don't even hardly ever open up [0:36:11] desktop Claude or desktop codex or any of the other. I just do everything through Atlas, [0:36:16] which I created through, used through Telegram. [0:36:21] We last month built an Android app. [0:36:24] So now on my Android, I have my own Google Assistant. [0:36:27] So it's just Atlas and I can just talk. [0:36:29] And so on the drive here, I'm, [0:36:31] hey, let's put together this presentation. [0:36:33] And I do a lot of driving in my home inspection business. [0:36:36] And so I get a lot done just because I'm talking to Atlas [0:36:39] whenever I'm driving. [0:36:42] What was that? [0:36:46] under most of it is through Claude yep but it's not just Claude built because [0:37:01] yeah I mean it's got integrations with tons of other AIs as soon as there's something [0:37:08] exactly that's kind of the whole thing is this brain is mine if all of a sudden Anthropic said [0:37:13] hey we don't like what you're doing bam your count's gone okay great i just pull this out [0:37:19] and i hook up codex because atlas is my brain all the all the logic everything of what i'm doing is [0:37:25] mine it's sitting on my server and i just plug it just reaches out to the llm to make a decision so [0:37:33] hopefully that makes sense like you can do a lot you could do probably 80 what that's i've been [0:37:39] researching grokbot probably 80 of what i built you can do with grokbot but if all of a sudden [0:37:45] like just what happened last week elon says yeah guess what we're just cutting everybody's usage [0:37:50] it's getting too hard well i mean i i hit i max out i'm on a 200 a month plan with with claude and [0:37:58] i'll often get close to the max if all of a sudden it was like oh guess what you get half the usage [0:38:03] that'd be like an employee calling in sick for a couple days every week so I [0:38:08] I need the flexibility that I can plug in whatever LLM I want no that was the [0:38:16] other thing a requirement I had from day one all of this is built on a VPS a [0:38:20] virtual private server just because I needed to be mobile I I can't have my [0:38:25] computer on all the time a VPS just a virtual cloud server so cost me like 200 [0:38:35] bucks a year and it runs now a bunch of my websites a bunch of apps it's just a [0:38:42] it's a virtual cloud desktop yeah yeah it's through Hostinger [0:38:55] Maybe very technical, but knowledge graphing, RAG architecture is what I've been working on for like two years. [0:39:03] I'm curious, how are you handling semantic drip and long-term database or brain ontologies like this? [0:39:12] So that's something that's kind of unique with what I've done. [0:39:15] Going through, oh man, that's a really long answer. [0:39:20] but yeah but basically it comes down to in a family friend that passed away in [0:39:27] the course of going through all of his notes he actually used to work for [0:39:31] Novell and they had a bunch of patents that expired about four years ago which [0:39:35] 20 years ago were the precursor for what is now the RAG system and it has some [0:39:41] indexing things that we've implemented and it doubled my file size but the [0:39:45] retrieval is so much faster. So yeah. [0:39:54] So think of it this way is you have a library with a ton of books and you have a, [0:40:02] and you want to find the book, [0:40:03] you go to a card catalog or you go and look it up and it says, okay, [0:40:06] you need to go to this shelf with this. But if you also had within that, okay, [0:40:11] not only this is the book, but here's the table of contents. [0:40:14] And within the table of contents, you said, here's the page and here's every word on this page. [0:40:19] Now, since every word is basically indexed, if I need to find a certain word, it now goes through like three layers of indexing rather than a single. [0:40:29] So retrieval is really quick. [0:40:31] The only difference is your file size doubles, triples, but retrieval is very. [0:40:38] Yeah, I found it because his website, he died about four years ago and his website had a search engine for searching. [0:40:43] he had about 600 articles that he'd written over the course of his lifetime. And, and I cloned his [0:40:50] website because his family wasn't really maintaining it as well. And so I'm like, I'm just going to [0:40:53] clone it and I'm going to build it myself. And I would try and to test it out. I'm like, okay, [0:40:57] go find this file. And they kept saying, I can't find it. I can't find it. And I'm like, why? [0:41:01] And basically it went through, we found these old patents that he had filed on there. And [0:41:06] basically we built his same search engine for his website. And it was 10 times faster than [0:41:13] than what AI could build me and that led to me being able to use a lot of it. [0:41:22] On a VPS, a hosting or server. [0:41:27] It saves you money as far as like on increased price because that rain you can access without using tokens for cloud. [0:41:35] Is that right? [0:41:36] Kind of because everything is there. [0:41:38] I don't need to keep huge session. [0:41:40] I can like every after every. [0:41:43] everything is stored there. So as long as it knows how to retrieve that information, [0:41:49] every time you're making a call, it has to take everything that you're thinking about and talking [0:41:53] about and send it over. And that's what uses up a lot of tokens. And so if you can keep those [0:41:58] sessions really small, you use less tokens because all the data is here. So it just knows, [0:42:02] oh, I need more context about this. Okay. Go grab it from this file. [0:42:05] Hostinger. [0:42:09] Hosting. [0:42:11] H-O-S-T-I-N-G-E-R. [0:42:14] Hostinger. [0:42:16] There you go, man. [0:42:18] Give it up for Jared. [0:42:19] Thanks, guys. [0:42:23] Yeah, you're going to have the things going to be after. === Korey Walker & Jeremy Dye — Epic Engineering's AI hub === [0:42:27] I think you also solved some of my problems. [0:42:31] We're recording this event, and what I plan to do [0:42:35] is take that put into my apf uh ai and say take all these ideas and tell me what i can implement [0:42:42] because it's a good stuff sorry sorry corey take it away matt love that you put this together it's [0:42:54] so cool to be able to rub shoulders with everybody that's trying to learn new things um [0:42:59] A year ago, we set some goals to try to increase our efficiency by 20%. [0:43:07] We're a full-service engineering company, Epic Engineering, and Jeremy took on that load. [0:43:15] Our goal was trying to bring in all of our financial information on projects, all of our project management information on projects, [0:43:22] and then link all our staff to that so that tasks and hours and goals on projects [0:43:29] were all linked together in one spot so they're easy to see and easy to follow [0:43:33] and AI-driven. [0:43:36] So essentially we have like an administrator for every person that lines out [0:43:41] what they need to do today, where they're at on it, and then uploads that [0:43:45] so everybody is completely transparent with relation to where we are on projects. [0:43:52] And so Jeremy has done a lot of work [0:43:55] over the last four months with some help [0:43:58] to be able to put this together. [0:43:59] So I thought it'd be fun to be able [0:44:00] to walk through that with you. [0:44:02] Yeah. [0:44:03] All right, so the gist of it is we used AI [0:44:07] to build a project management software [0:44:10] that we can then use AI to run. [0:44:13] So we built all the like structured data and everything [0:44:18] so that when we ask for AI to do certain tasks for us, [0:44:22] It has that structure to put everything in. [0:44:25] So we built a custom software. [0:44:28] It has everyone's dashboard. [0:44:29] It has task management. [0:44:31] So it goes through everyone's tasks. [0:44:32] We have employee performance metrics and scoring. [0:44:38] We have timesheets for professional services. [0:44:40] So we build by the hour, right? [0:44:42] And we have to fill out what we do [0:44:43] every 15 minutes of every day. [0:44:45] And it's like a huge tedious thing. [0:44:47] So we can use AI to kind of automate some of that, [0:44:49] what we're doing. [0:44:50] We also rebuilt, we have, we rebuilt our CRM, [0:44:55] client relationship management software. [0:44:58] So it integrates with the other stuff we've got, [0:45:01] does all of our business development stuff. [0:45:03] So we've got like our salespeople, [0:45:05] they've got dashboards for who to call next, [0:45:10] what to follow up on, what we've, you know, [0:45:12] who we've reached out to already. [0:45:14] There's like some gamification there for earning points [0:45:18] for doing certain tasks and things. [0:45:20] One of the things that we have spent a lot of time on is generating proposals. [0:45:25] So like in construction, every project is different. [0:45:28] We have a client, they call us, they say, how much is it going to cost to do this engineering project? [0:45:33] And that takes a lot of time because we have to like run it through a lot of things. [0:45:37] So now we have some steps that we've built into it. [0:45:42] It ingests emails, it takes photos, it recommends, looks at past projects, [0:45:48] compares all of our existing projects to new projects [0:45:53] and helps us fill out proposals. [0:45:56] Then it rolls right in, sets up our projects. [0:45:58] It creates PDF contracts that we can then send to clients. [0:46:02] So all of our information from our proposals [0:46:04] get pushed right into contracts. [0:46:07] We send those out for signature [0:46:09] and then it pushes into our project delivery. [0:46:12] So we can track every single project, [0:46:14] what status we're on, how we're doing on budget, [0:46:17] who's spent how many hours working on whatever. [0:46:20] We do quality control, [0:46:22] checking that our drawings don't have mistakes on them [0:46:24] before we send them to clients. [0:46:26] We have meeting ingestion. [0:46:29] So it records meetings, adds assignments for everybody, [0:46:32] tracks those decisions and things. [0:46:34] And then it checks our design. [0:46:35] So a client says, hey, I want you to do this sort of thing [0:46:38] on this project. [0:46:40] It like double checks our work and says, [0:46:42] hey, you forgot to do this thing [0:46:43] that the client told you to do on this date. [0:46:45] So it helps us keep track of that. [0:46:47] Yeah. [0:46:48] Yeah. [0:46:49] So does it take me a long time to go? [0:46:52] Five months right now? [0:46:54] Five months, basically full time. [0:46:57] 30 hours a week. [0:47:00] If you were starting now, what would you do [0:47:04] to expedite you being able to do this faster? [0:47:07] Is there a way that you could expedite that? [0:47:10] If there is, I would like to know it. [0:47:14] We actually got quotes from other people that were talking 18 months. [0:47:18] We actually are looking at some people to come in and help us do some security. [0:47:22] One of the guys I went to originally for this proposal showed him what we had done in four [0:47:27] months. [0:47:28] He was flabbergasted at what we had accomplished in four months. [0:47:31] Couldn't believe it. [0:47:32] And it was so much bigger than he ever thought. [0:47:35] He understood he was going to do. [0:47:37] Yeah, question. [0:47:38] You can do it faster on my platform. [0:47:39] On yours? [0:47:40] Talk to me after. [0:47:42] Cool. [0:47:45] So we're figuring it out, right? [0:47:47] Everybody's exploring. [0:47:50] We've got dashboards. [0:47:51] We've got meeting agendas. [0:47:53] We've got directories. [0:47:55] We're tracking licenses. [0:47:57] And then all of this has the MCP connection. [0:48:01] So you're opening up your favorite agent and you're saying, hey, I just got off the phone with so-and-so. [0:48:06] I talked through this. [0:48:08] We need to add this lead, add this whatever, add this whatever. [0:48:11] it tracks it scores it it kind of generates everything um and helps us through it so we're [0:48:18] got a lot of like back-end tables that makes everything work i have a feature request from [0:48:24] employees so as employees are using it we're still in active development and people are requesting [0:48:30] features and then i'm using ai to kind of triage those features and automatically build those in [0:48:36] for me as we go. [0:48:39] So kind of getting feedback from employees [0:48:41] and doing that. [0:48:42] So that's the overview. [0:48:44] If there's faster ways, I'd love to know it. [0:48:47] But maybe like what things are like automated [0:48:49] that are making this? [0:48:52] So like a proposal used to take us four hours. [0:48:56] Now it takes us 15 minutes. [0:49:00] You can just speak into Claude, [0:49:02] tell it what you want it to do, [0:49:04] tell it to upload it. [0:49:05] It creates it. [0:49:06] It knows our billing rates for every person. [0:49:08] It knows what each person is good at. [0:49:09] It knows how many hours it's going to take each person to do each thing. [0:49:13] What I find myself being fascinated in is you have something built out like this, [0:49:18] but then you make AI smarter on top of itself. [0:49:24] Because I get so mind numb being like, no, you idiot. [0:49:29] I don't like the button right there. [0:49:31] I want it over here. [0:49:32] Yeah. [0:49:32] Yeah, and that was one of the things we were running into also is every employee has their own agent and they've all kind of taught their own agent how to do it their way. [0:49:42] And we were getting proposals from each person and everything looked different. [0:49:47] And so, you know, Claude will create a proposal and it'll make it look this way. [0:49:50] But by building kind of the structured data, we have our letterhead that gets used for everything. [0:49:56] We have the sections. Everything is divided up into assumptions, limitations, exclusions. [0:50:01] and so no matter which employee does it, [0:50:04] it all comes out looking the same. [0:50:07] And they're all using their individual thing, [0:50:08] they've all got their own preferences, [0:50:10] but by building that kind of that framework [0:50:13] for it to fit into, it really helps with [0:50:16] 20, 30 project managers. [0:50:19] So yeah, that's kind of what we've been up to. [0:50:23] We're super excited because it's helping speed us. [0:50:26] Before we were just, basically everyone was just using [0:50:28] their own email to track what they were supposed [0:50:30] to be doing. [0:50:33] We had finance stuff that would keep track of ours, [0:50:35] but not really like a together place for everything. [0:50:40] So questions? [0:50:42] What database is it on? [0:50:44] What database? [0:50:46] Back-end database? [0:50:50] I don't know. [0:50:54] Testing everything CPs, and it's all [0:50:56] stolen for your other systems? [0:50:58] So we have, I don't know. [0:51:00] You have your payroll system. [0:51:02] So we have, we use Vantage Point for our finance system [0:51:06] and we have APIs. [0:51:08] So this hub, this hub keeps a local copy [0:51:13] of everything in our finance, [0:51:15] keeps like a local copy of each project, [0:51:17] but then it adds all the extra things [0:51:19] that we wanna track on top of, [0:51:22] because like Vantage Point doesn't have a field [0:51:24] for contingency. [0:51:26] There's no way to add contingency to a project. [0:51:28] So we have contract value. That's it. That's the only thing our finance software lets us do. [0:51:32] So now we can add contingency and then that lets us track our planned hours and we can like [0:51:38] basically add a whole bunch of extra tables to that without losing our old data. [0:51:44] Good question. Just for the room. [0:51:47] More like overview here. You're the boss, right? [0:51:50] Yeah. [0:51:51] On the company Epic, right? And then you just said, hey, I want to build my efficiency, right? [0:51:57] so you hire him no I'm just he was working with us okay I'm just looking at from my perspective [0:52:03] I'm a builder yeah and so like the overview is I have no desire to decode it you know yeah yeah [0:52:13] I got enough shit I gotta do myself right um and but I'm interested in that part is like hey I'm [0:52:21] willing to spend money if I can hire somebody to do what I wanted to do. [0:52:26] Yeah. And there's a lot of people in this room that have all learned skills, right? [0:52:33] That's where mine is like, how did that process? [0:52:36] But like the barrier to entry is pretty low, right? As long as you understand what you're [0:52:39] trying to get, you can get it. It's just a matter of kind of spending some time on it. [0:52:44] I'm just, that's like, I don't want to learn about something. [0:52:47] Yeah, yeah. Or hire a 20 year old, right? [0:52:50] hey i i see a need i see a problem i need somebody to do that problem based off of what's in my head [0:52:57] or include a ton of legs yeah i mean the fact is as we bootstrapped right jeremy is incredibly [0:53:02] intelligent and incredibly capable he's an it guy but he's a voice that was incredibly smart [0:53:08] they said yeah i think we've figured this out there's a number of people in this room but so [0:53:13] we've got five months of experience [0:53:20] I think, like, as business owners, there's a lot of us that are like... [0:53:24] But with AI coding, you don't really need to know code. [0:53:26] You just need to know how it should function. [0:53:27] Yeah. [0:53:28] And you can make it work. [0:53:30] Same thing, too. [0:53:31] It's like, at that part, I got enough stuff to do. [0:53:34] I need somebody else to do that part. [0:53:36] Yeah, yeah. [0:53:37] So you hire a part-time guy, and they just follow you around, and they do it. [0:53:41] I'm interested in how that came about. [0:53:43] Yeah, so I'm a structural engineer. [0:53:45] That's my training. [0:53:47] So I've done project management, and so I'm just kind of, like, interested in it. [0:53:50] and learned it. [0:53:51] That's really cool. [0:53:53] And where do you host the system, this particular software? [0:53:57] Is it on your local system? [0:53:59] It's on a VPS. [0:54:01] Yeah, it's on a VPS. [0:54:02] We use our Microsoft login, and then we just [0:54:06] have only our employees are allowed to log in. [0:54:11] Yeah, so we just use the Microsoft as the gate for that. [0:54:14] Do you have any other more examples [0:54:16] off the top of your head like that you [0:54:18] had for taking it from four hours down to 15 minutes? [0:54:21] Are there any more? [0:54:23] Like cost savings? [0:54:25] Okay, so right now, what used to happen, [0:54:27] we would set up, we would send a proposal, [0:54:30] and then we would get that, [0:54:31] but we'd get the go-ahead from the client, right? [0:54:33] And then we would have to set up our project. [0:54:35] So we'd go in our software, we would create this, [0:54:39] enter all the client information, set up all the stuff, [0:54:42] and then we would paste all of that into a Word file [0:54:45] that was our contract, [0:54:46] and then we would send that to the client. [0:54:49] And then we would paste all of that information [0:54:51] into monday.com so that we could track it. [0:54:53] And then every week we had to go in [0:54:55] and look at our budgets that we had spent [0:54:56] and paste it in there so that our users [0:54:58] could see how much they'd spent on every project. [0:55:00] And then we also have folders. [0:55:02] So this is, we use a file server called Ignite. [0:55:06] It's a cloud storage, kind of like Google Drive [0:55:09] or OneDrive or whatever. [0:55:12] But it's on the cloud, it has APIs that we can use. [0:55:16] So I actually, I have report templates [0:55:19] and it will open that report template. [0:55:21] It'll paste in the client information. [0:55:23] It'll save a PDF. [0:55:24] It'll create a folder for me for that project. [0:55:27] It'll populate it with all of the design, [0:55:30] construction, contracts, [0:55:31] like all of my subfolders that I need. [0:55:35] I've got a thing in here that I built yesterday, [0:55:38] which is for reports. [0:55:40] So I go to a site visit. [0:55:42] There's a park city, water came through the roof. [0:55:45] All their artwork fell off the walls. [0:55:48] It's a big art museum. [0:55:49] They're down. [0:55:50] They're like, hey, engineer, what's going on? [0:55:51] So I walk up, I take a video, [0:55:54] I walk through and I narrate the video. [0:55:56] I upload that video to our hub. [0:55:58] It will take screenshots of different pictures as I go. [0:56:03] So it'll extract still pictures. [0:56:05] It will type up my entire report. [0:56:08] It will put all the pictures into that report. [0:56:10] It uses my letterhead. [0:56:11] It puts figure captions on everything. [0:56:13] I got a nine-page report in less, you know, one paragraph prompt. [0:56:21] So that's the kind of thing that we're setting up there and trying to streamline. [0:56:26] All right. [0:56:29] All right, Alex, you're out. === Open floor — what changes now === [0:56:39] Who can help on the ground? [0:56:41] We got people in the rear. [0:56:46] While this is being set up, this whole conversation about AI has just got me thinking more and [0:56:53] more about what the future has in store. [0:56:56] I mean, I'd love to hear your comments on it, but it feels like information is not that [0:57:06] valuable. [0:57:07] less valuable. And so it's like, what's the only right answer to the top? If we make money [0:57:14] faster, if we get the things that we want in our jobs, it seems like the relationships [0:57:24] that we're building are going to be the things that, at least to me, are going to be like, [0:57:29] that Trump, that king, that's not going to be taken away, you're going to wish you had [0:57:33] more up that you can't really do as well later because it might be harder to [0:57:40] get up things kind of develop and air goes yeah I know this the the time from [0:57:56] idea that like effective product is completely impressed I was gonna ask [0:58:02] how long you guys went from idea to that product and I bet it was crazy short amount of time what's [0:58:08] not going to change is people still have really stupid ideas a lot of the time and even trying [0:58:12] to turn a bad idea into a product it's still in that I think most people have great ideas and it [0:58:18] used to be where enterprise SAS was most of a good idea for some people but you have 40 other menu [0:58:27] items on your SAS product that you don't need because they're trying to put a lot of good ideas [0:58:31] for everybody. Where now, just like you guys, you build your own custom in-house [0:58:35] system that doesn't rely on anybody else and it's just doing exactly what you want. [0:58:39] And that's the exciting future I see the day out. Because regardless of what your [0:58:43] idea is, it's not that expensive because it's not going to take up much time to prove [0:58:46] what's a good idea or not. And then you get updated on the fly and build it your way. [0:58:50] It's a beautiful day. [0:58:51] You have eight different programs you're working in. [0:58:53] Yes, right. [0:58:54] So far. [0:58:55] So for those that are wondering, my platform is called ByteWork, and it's basically like [0:59:03] an enterprise platform, so you don't have to think about where to host and how to set [0:59:07] up security, and all kinds of others. [0:59:10] Just build the apps you want, the data, the databases, that's all there in the platform [0:59:15] at a highly secure level. [0:59:18] And that's exactly what it's built to do. [0:59:21] I still need to figure out all these other things. [0:59:24] And it's like, just build apps that you want to do [0:59:26] and stuff to do what they did. [0:59:27] It's like, build things you want to build. [0:59:29] And they're all AI-nated, AI-connected, [0:59:32] and you can automate anything. [0:59:34] It's all there. [0:59:35] One place. [0:59:36] What's the problem with the sideboard? [0:59:38] I work. [0:59:40] Our next site, if you go to our website, [0:59:42] it's very construction-focused, because that's been our thing. [0:59:45] But it can do, it works in the industry. [0:59:51] um okay a little background i grew up with matt um i'm a little different a lot of people in this === Alex Sisk — a manager-led team of Claude sessions === [1:00:02] room i'm actually in grad school right now i'm getting my mba at byu um so i don't have a business [1:00:07] to apply all the things to all the time so a lot of times that's me applying it to my life and then [1:00:12] kind of working and getting into like maybe some um kind of ai application consulting [1:00:21] stuff like that while I'm looking for a job [1:00:23] and doing everything else. [1:00:23] So what I wanted to present on today [1:00:26] is kind of like what Jared talked about [1:00:29] is kind of like what I aspire to, [1:00:31] what I'm like working towards [1:00:32] is kind of building this like whole literal second version [1:00:35] of myself and everything like that [1:00:37] is what I've been slowly working towards. [1:00:41] But as I'm doing that, I have a VPS that I use [1:00:45] for just some like personal apps that help me [1:00:49] with various areas of my life. [1:00:53] And so all I'm doing is just like one at a time, [1:00:55] adding those in. [1:00:56] Like this week I built a big recipe book, [1:00:59] like digital book that my family can all share recipes to, [1:01:02] that I can build grocery lists off of [1:01:04] and then I'll just integrate it into like Walmart online [1:01:07] and just have it literally order my food [1:01:09] and have like it just get delivered. [1:01:10] And so I'm basically trying to offload all the crap [1:01:13] I don't wanna do. [1:01:14] But I'll show you my Claude, [1:01:16] but this is kind of the workflow [1:01:17] of how I've set up my Claude code to work, [1:01:19] which is mainly what I wanted to show tonight, [1:01:22] is I have one manager that a lot of people [1:01:23] call chief of staff or whatever. [1:01:25] I just literally call it, it's called manager. [1:01:27] I have names for all my other stuff. [1:01:29] I got Rocky and everybody else. [1:01:30] But manager manages leads and each project [1:01:35] kind of has its own lead. [1:01:36] And then those leads can just kind of spawn up sub-agents [1:01:40] who then, I have a reviewer who reviews back to it [1:01:43] and feeds back to the manager who can kind of, [1:01:45] so it creates this really nice loop. [1:01:47] And as long as I've correctly set my goal from the beginning, [1:01:52] it's helped me really kind of keep things moving [1:01:55] in the same direction. [1:01:56] Because I have the tendency to like, [1:01:58] halfway through a build, I have this new idea [1:02:00] and I just start running with it. [1:02:01] And it's kind of, I can get a little bit lost with it. [1:02:06] And so how I, you can see it over here, [1:02:08] how I currently have it set up, [1:02:10] where I have these leads for different areas [1:02:13] that I'm working on in different apps [1:02:14] that all kind of just feed back into themselves, right? [1:02:22] And so this is probably a little bit of a simpler way [1:02:26] to do what I think several people have kind of talked about, [1:02:30] like managing all these various different things and stuff, [1:02:32] like within Claude. [1:02:33] Like I still have my duet, my instinct linked to this, [1:02:36] and I can talk to them and help with things like that. [1:02:40] But I just live in this screen right here. [1:02:42] actually literally just today found out I could open up two Claude screens and so I have can have [1:02:47] my duet on half screen I can have my manager on half screen and go back and forth and that's been [1:02:51] that's been awesome so anyway these are just kind of some fun Claude mods that I've played around with [1:02:57] to kind of get my stuff set up the way that I like it and like to use it but yeah that's basically [1:03:05] like I'm happy to kind of answer any questions or even help kind of show people how to get some of [1:03:09] this stuff set up if it's something that's like, maybe not quite as advanced as what others have [1:03:15] already presented on, but like maybe a little bit more tangible or doable, or if you just have kind [1:03:19] of some individual apps that you're wanting to build for your own life. I believe that this is [1:03:24] like a really nice way to do that. Yeah. Yeah, my favorite is this button right here. So I built an [1:03:35] enhance button because I like to just talk into into the prompt I actually have [1:03:41] it on my github so I'm happy to send you a link if you just want to go set it up [1:03:46] so these are you can kind of last few weeks Claude mods so you can literally [1:03:52] tell it what you want it to do and it can create this space so I have this is [1:03:55] my builder board so like these are my next five tasks that I need to work on [1:03:58] and then right here are all of my projects that are currently being worked [1:04:02] on and then at the bottom I have my idea shelf where I just plug ideas into and I can quickly [1:04:07] like click brief it and it'll pull and create a prompt over into my prompt box but this is what [1:04:14] I just call this my button bar but the enhance button is my favorite so I'll just talk into it [1:04:20] there's idea dump and then I'll click enhance and it will go through and rewrite my prompt in a way [1:04:25] that Claude can way more easily ingest with criteria, goal, kind of all the all the various [1:04:34] things that Claude codes looking for that most of the time like I'm not taking the time to type out [1:04:40] and like organize for it. I like it better because I just rip it. I don't spend a lot of time in plan [1:04:50] mode to be honest with you. But similar, I think like, as you're kind of honing in, [1:04:59] but it'll be more even like mid build. As I'm just saying like, hey, like, let's work [1:05:05] on this, change this changes, whatever, this still will help it write better than what [1:05:12] I could do. I actually I mean, this is a very simple, I was working on a LinkedIn post, [1:05:20] So I took, yeah, I took this video today [1:05:23] that kind of show you like just very simple. [1:05:26] So like that was, I need an app that reads my golf scorecard [1:05:29] and quietly turns every bogey into a par for morale. [1:05:32] Nobody needs to know. [1:05:34] And so then I clicked the enhance button. [1:05:37] This will get posted at 10 AM on LinkedIn tomorrow [1:05:38] if anyone wants to go, you know, find it. [1:05:41] And then you can see here where it'll pull up [1:05:44] based on just that little simple thing. [1:05:47] It's gonna build your constraints, your done wins. [1:05:49] so it kind of knows how far to take it, [1:05:51] how far to run with the idea that you gave it. [1:05:55] None of this, I mean, sure, [1:05:56] Claude could have maybe come up with some of this [1:05:59] if I just submitted the very first thing, [1:06:01] but now I've kind of built it in a way [1:06:03] where it's gonna be a little bit more seamless. [1:06:06] And obviously, most of the time, [1:06:07] it's gonna be more than two sentences [1:06:08] that go in before I enhance it, but. [1:06:10] Have you noticed that improved your token usage? [1:06:13] um yeah i to be honest i don't pay a lot of attention to my token users most of the time [1:06:21] like i have these i have these circles here to show me like if i'm ever getting close on filling up [1:06:26] um but most of the time i'm not too worried about it [1:06:30] i'm gonna explain like i'm five fan as well do you kind of also use the mod with like [1:06:38] quick access to skills yeah I actually built a skill mod at one point that I had open in a second [1:06:45] tab up here I haven't been using it but but yes it was kind of I called it my skill drawer so I [1:06:52] could just open that up and quickly like if I wanted to deploy it but I was also getting and [1:06:57] I've still been dealing with a lot of like context below in my context window it's just compressing [1:07:01] way too often. And so are compacting and so I did that to turn off a lot of the skills that were [1:07:08] being fed in all the time. But yeah, the mods I think are awesome. Because even if you just go [1:07:14] just Google like Claude mods, like people, there's a lot of like, blog posts and people or YouTube [1:07:21] videos where people are just showing them. And I went and watched like, I probably spent two hours [1:07:25] one night just like reading stuff and watching YouTube videos. I just picked a few of the stuff [1:07:28] like the, I kind of ran with one, someone had this idea thing and someone had the build board. [1:07:34] And I was like, well, I only have limited space. So let's just kind of like fit those into one. [1:07:38] And I wanted this, I have a buddy who has this thing down here, he it's filled up like really [1:07:42] high. It drives me like when I look at his screen, I'm like, that drives me crazy. Cause [1:07:46] I don't like being able to only see a little bit of up here. So I try to keep this really thin, [1:07:50] but you know, use the space well. So yeah, that's been, um, super fun for me to use. So [1:07:56] So any other questions? [1:08:00] Yeah. [1:08:00] A little bit of a question earlier, [1:08:02] we're talking to someone else. [1:08:03] So we're creating like just tasks [1:08:06] that you're giving to cloud so you can run them all parallel [1:08:09] in the background. [1:08:10] Are you using, are these just singular agent sessions [1:08:12] or do you have multiple agents that review the code, [1:08:15] check out, like run a whole bunch of different processes [1:08:18] or is it just- [1:08:19] Yeah, it kind of depends. [1:08:22] So yeah, I mean like technically [1:08:23] these are all individual agents that can each kind of like create their own as needed. And then I [1:08:31] have a reviewer, I have like really strict rules around like my reviewer agent that will go in and [1:08:38] see what's been done and then feed back into the manager. And then that's kind of what comes back [1:08:42] to me. I'm not seeing, I'm not getting into the weeds. Sometimes my control freak takes over and [1:08:47] I like get into these and I start typing in them and I'm like, no, no, no, hold on. I got to get [1:08:51] it back to the manager and like let the process manage itself. But like, for example, like LinkedIn [1:08:56] has been something I'm building like a LinkedIn hub. And so I actually built the thing where this [1:09:01] lead will, will spawn up 100 sub agents that will all read the post draft that it made for a LinkedIn [1:09:08] post. And they all will take on the role of some LinkedIn user. And they'll react to like whether [1:09:14] they like that post or not. And then they kind of feed their input back to the reviewer who brings [1:09:18] it back to the manager. And so then there's kind of the criteria of like it has to at least pass [1:09:23] a certain point with the reviewer or the manager will just send it back through before [1:09:27] coming back to me. And so I try to give them quite all of the different leads that I call them [1:09:33] quite a bit of autonomy to do what they need to do. [1:09:42] Right now they're all running on this. So I spent a lot of time with my on my Mac. I spent a lot of [1:09:47] time with my Mac caffeinated and open it all night and things like that. But the plan is to [1:09:54] build, I want to build my own machine like at home, like a server, just because I think [1:10:02] looking at financials of all these LLM companies, like I just feel like they can only offer it [1:10:09] at the amount of usage they're giving at these prices for so long before like they're going to [1:10:15] eventually have to raise some of those prices, [1:10:18] unless they all compete. [1:10:19] I don't know, I've heard mixed opinions. [1:10:22] But I'm a big believer in eventually having some local, [1:10:27] I just wanna have a server at home [1:10:28] where I can have a lot of stuff, [1:10:29] have a local model that's just automating [1:10:31] the little things that I don't. [1:10:33] I mean, I use Opus for almost everything [1:10:34] that probably overkill for a lot of what I'm doing. [1:10:37] We have a really similar setup [1:10:38] with all the pages data that holds everything on our platform. [1:10:40] So having this podcast sometimes [1:10:43] we mostly just write a task and we just draw these. [1:10:45] But they all have, like each developer [1:10:47] just has their own Mac Mini as a server [1:10:49] that's running the fleet. [1:10:51] And the fleet is a whole series of agents [1:10:53] depending on what it's going. [1:10:55] It's kind of a similar setup, it's very cool. [1:10:57] Yeah, yeah, I've considered the Mac Mini option. [1:10:59] I've had one in my shopping cart for a couple months now. [1:11:05] I've gone back and forth on whether I go that route [1:11:07] or try to build my own, but. [1:11:11] Yeah, any other questions? [1:11:15] community yeah networks and cards and all your stuff yeah and that community yeah that's that's [1:11:21] what i hear so okay cool thank you yeah [1:11:32] all right [1:11:37] okay and that will conclude our presenters will feel free to hang around uh network [1:11:44] Yeah, the mocks, I have not heard about that. [1:11:55] Yeah, I think just last few weeks. [1:11:57] It was on... [1:11:58] It was on... [1:11:59] It was on Claude. [1:12:01] Oh, I was on Claude. [1:12:03] Yeah. [1:12:04] I felt pretty, like, consistent with, like, apps for it. [1:12:10] I was like, this really what I've been like waiting for that forever. [1:12:25] I've been using instincts like crazy. [1:12:28] And I've told everybody in my life, like the day clock comes out with this. [1:12:33] I'm all in and it's going to be, you know, and I'm like this. [1:12:36] And then I feel like they're already improving. [1:12:38] but yeah, I'm not, you're not on the list, [1:12:42] get on the list or get it. [1:12:43] Cause I'm, it's good stuff. [1:12:45] All right, Jeff. [1:12:46] Wait, so it's really blurry. === Chad Mckinney — skills, memory and integration scoping === [1:12:50] Not sure why. [1:12:51] I mean, I don't know if it's my connection. [1:12:54] I can maybe stop sharing and then share again, [1:12:56] but real quick about me, I'm a software engineering manager. [1:13:00] So I manage a team of about 30 software engineers [1:13:03] over six different teams. [1:13:04] And then I, on the side, we'll do integration work for a lot of clients. [1:13:10] Let me stop sharing and see if I can get it to look better. [1:13:18] But because I do a lot of work on the side, yeah, I'm not sure why it's so blurry. [1:13:26] I was just going to show you kind of one of the workflows. [1:13:30] I also got my master's in cybersecurity management, [1:13:32] So I'm a little bit leery on just giving free reign to AI. [1:13:38] I've seen there's going to be a lot of legal repercussions for a lot of people who let their AI run free. [1:13:43] And I don't want to be one of those. [1:13:45] So I have definitely trained my AI. [1:13:48] One of the biggest things for me is memory. [1:13:50] So I'm constantly creating memory files and nesting memories. [1:13:55] So that way it's very applicable to a specific client or specific project I'm doing. [1:14:00] This here, I also use a lot of skills. [1:14:02] So I create a lot of skills based upon very specific needs. [1:14:06] So a lot of the client work I do is I work with someone who's really not tech savvy on usually an HR management team. [1:14:18] And I work with a sales rep who doesn't have the time or desire to make these systems work. [1:14:25] So I've got a lot of friends who work at like BambooHR or various SaaS companies around Utah. [1:14:30] they'll say hey I've got someone who's looking to have BambooHR speak to in [1:14:36] this case it was Paylocity they use Paylocity for their payroll they use [1:14:40] BambooHR for for their they're looking to use BambooHR for their HR management [1:14:46] system is it possible that's usually the question I get is is it possible like [1:14:51] well I mean probably so I created a skill this one here is the integration [1:14:57] viability, it literally just goes through and it I have it run [1:15:02] through all publicly available API documentation. So it'll go [1:15:07] through and prioritize the documentation based upon the [1:15:11] source, right? So it'll compile all of that information, it'll [1:15:15] get all the endpoints that I'll be needing to work with. For [1:15:20] typically, one of the nice things too, is I also use Granola for [1:15:23] all my meetings, it just sits there and records [1:15:28] and gets a whole transcript of the meeting. [1:15:30] And that transcript is then, I use the MCP [1:15:33] to connect it into Claude and I say, [1:15:35] all right, based on this conversation with this client, [1:15:37] I'm looking to build out this integration [1:15:40] and they're looking to have these data sets [1:15:42] moved from this place to this place. [1:15:44] And I feed all of that in along with [1:15:47] the other portions of the skill. [1:15:50] And it comes back and it says, all right, let me, [1:15:52] It just builds, it's a little bit verbose, [1:15:54] but it's great for me, [1:15:55] because it gives me all the ammunition I need [1:15:58] to say what it's gonna take, [1:16:00] and then to build out a pricing. [1:16:02] So this skill here, and then once this skill runs, [1:16:06] this gives me a good kind of initial, yes, it's viable. [1:16:09] So then they'll go and talk to the client, [1:16:11] because I'm not, the nice thing is I'm not selling, [1:16:13] they're selling. [1:16:14] So I'm typically kind of the after sell, right? [1:16:17] So they'll sell to the client, [1:16:19] and then they'll say, all right, [1:16:21] now how much is it gonna cost? [1:16:22] I'm like, oh, well, let's build out a statement of work. [1:16:24] So I'll take all the information from here, [1:16:26] which is basically, it has an entire scorecard [1:16:30] that I had it build. [1:16:31] And like I said, you can't read it. [1:16:32] I don't know why it's so blurry, [1:16:33] but it's all the requirements, the authentication process, [1:16:37] the specs that are provided, a model of the data, [1:16:42] how often the APIs change, [1:16:45] because that impacts the overall cost [1:16:47] of managing this integration long-term, [1:16:52] whether it has a sandbox, if there's custom fields [1:16:54] or if there's not custom fields, [1:16:56] if I can do things in bulk [1:16:57] or if I'm gonna be doing things via webhook, [1:16:59] so it's at the time the event happens, [1:17:02] it just automatically triggers my system. [1:17:04] If there's rate limits, so I can't go beyond a certain amount. [1:17:07] And then it goes through the domain coverage. [1:17:11] You can't really read it, I apologize, [1:17:12] but it's talking about pay statements, hiring, [1:17:16] new hires, terminations, all the various things [1:17:18] that someone in an HR department cares about. [1:17:22] That's what this specific integration is for. [1:17:23] That's why I gave this as an example. [1:17:26] And then it goes through and it tells me what kind of API it is. [1:17:29] And then I'm able to build out and take all these things and determine the risks, [1:17:35] determine how complex the build is. [1:17:37] And then I also have it give me a list of open questions like, [1:17:40] what should I ask this client before I agree to take this job on? [1:17:44] So it gives me a full scope. [1:17:47] Yeah. [1:17:47] yeah so like i said i'm a software engineering manager uh by day and then i i build these [1:17:59] integrations in the evening as just a side hustle just because um just my own just ride creative [1:18:07] it's just my own thing it's just me alone yeah me myself and i um it's uh i i do i i handle all the [1:18:15] hosting, build out the infrastructure, handle all of the, you know, the security measures, [1:18:21] I make sure to lock down all the ports, keep everything really, really succinct. [1:18:25] And the entire integration portion runs completely AI free. I, everything is built to, to be [1:18:32] self hosted. So that way, it just is self maintaining. And I have logs and measures [1:18:39] and things that will notify me if something goes wrong. I don't want to rely on AI at all for any [1:18:44] my client work because I don't want to deal with those extra costs so I take [1:18:48] the AI for the build-out process and and then I I host the whole thing once [1:18:54] again hosting or is what I use for for a lot of my client work or AWS depending [1:18:59] on what they're looking to what what kind of data they were looking to do I [1:19:03] don't store anything so I just I'm like I can host it anywhere because it's just [1:19:07] all encrypted in transit so yeah so a lot of clients that I work with right now are there's [1:19:18] there's several construction companies out in East Canada that I'm building an integration for so one [1:19:25] of them uses plexus is the name of their ERP system and the other one uses a company called [1:19:33] CMIC for their ERP system it's where they manage all the stuff that was being [1:19:38] talked about earlier right their entire all of their employees stuff they're all [1:19:41] of their their quotes and all their build out and every they're you know all [1:19:46] their projects and everything so what I do is I get those projects pulled into [1:19:50] BambooHR so they can do time management as well as project allocation and stuff [1:19:57] like that so so then what it'll do from here I have a statement of work handoff [1:20:03] so I say hey everything's ready to go I've signed off with the client generate [1:20:08] a statement of work and it generates this beautiful statement of work for me [1:20:11] I've designed the whole thing I went through and painstakingly worked with [1:20:15] Claude to make it look beautiful but I don't want to take up too much more time [1:20:18] but anyway I apologize it was all blurry I don't know why but yeah any any [1:20:23] questions a lot of my stuff like I said is based on skills and and memory so [1:20:28] that way everything I use very little tokens just because every time I do [1:20:34] something I store it and then I start a new session I never stay in sessions [1:20:39] long so it's just constantly and so I have a ton of memory files just stored [1:20:43] on my computer just a bunch of markdowns yeah is it very specific to [1:20:53] like the niche that you're selling to? [1:20:56] No, so I just put it, I created it generically enough [1:21:00] that it's just two SaaS companies, right? [1:21:01] It's like, hey, can I connect these two SaaS companies? [1:21:04] And I give it the use case, right? [1:21:06] Most commonly, because I've got a lot of friends [1:21:09] at BambooHR, that's typically the clients [1:21:11] that I'm working with, but they'll have, [1:21:13] I'll get referrals that are like, hey, can you, [1:21:15] like a recent one is Ramp and what's it called? [1:21:21] And Productive.io. [1:21:23] neither of which are BambooHR, right? [1:21:25] Ramp and productive IO, productive IO is for expense management [1:21:30] and ramp is for employee expense accounts, right? [1:21:35] So now it's, hey, can you get these connect to each other? [1:21:38] Yeah, I can. [1:21:39] So I use that same skill and put those in together [1:21:42] and built out a quote for the client. [1:21:44] Like if the company were to reach out to you [1:21:47] whatever means they construct you [1:21:49] and they explain to you like their processes [1:21:51] what they want automated or whatever they want you build essentially and what they want you to [1:21:57] connect would you be able to essentially take that recording of the meeting and like the bullet points [1:22:03] put it into that using the skill and it come back saying what you can yeah tell me whether it's [1:22:09] viable right it tells me tells me the complexity of the integration whether they have publicly [1:22:14] available apis if they don't have publicly available apis it'll go and kind of slew through [1:22:19] and try and find, whether that's a swagger document [1:22:23] or a postman record, it'll try and find stuff [1:22:27] and I have a rating system for how valid [1:22:30] that API documentation is. [1:22:33] It's like, hey, this came from the source, [1:22:35] that's a 10 out of 10. [1:22:36] This one just kind of came from an API, [1:22:41] like, yeah, exactly, someone says that it's possible, [1:22:45] that's gonna be lower on my ranking [1:22:47] And so the viability score is based upon kind of that weighting. [1:22:51] And I keep refining the skill the more I do it. [1:22:53] And then the nice thing is I also create a bunch of environment configuration files locally. [1:23:00] So that way the AI doesn't need to know any of the personal information, [1:23:05] like the API tokens and stuff, [1:23:07] and can still actually query against live data [1:23:09] to then give me a better mapping of data as well. [1:23:13] So that way, that's stuff I was doing by hand, [1:23:15] going through and running the Postman scripts [1:23:17] and pulling all of the all of the the data that was coming from those endpoints and now AI is [1:23:21] able to pull that and create a markdown file for me yeah no the ones are in Nova Scotia [1:23:36] surprisingly the two that I'm working with yeah okay yeah no I I've thought about going [1:23:47] onto their marketplace but it's like I don't really need to it's kind of like [1:23:50] I get pulled in for these kind of one-off jobs no it's actually the so the [1:23:58] sales guys I was just friends with several of the sales guys and they're [1:24:01] like dude I need to close this deal they need this you know for you a QuickBooks [1:24:06] was probably one that was like they use QuickBooks it's like dude that's an easy [1:24:10] integration for me to build and they're like perfect can I get you hooked up with [1:24:13] it? Zach Janik is like my my number one guy. That guy's a beast. He's awesome. Yeah, he's awesome. [1:24:22] Anyway, so yeah, like, so that's the kind of stuff a lot of times it's they're just trying to, like, [1:24:27] from my side, I'm like, I don't need the extra money. It's just it's nice. But it's I, I had [1:24:32] some friends who asked for help. And I was like, let's do it, man. And so now I'm just using AI a [1:24:36] ton to just like, have some fun keeping my skills because as a manager, I don't really code a whole [1:24:42] lot anymore so I'm like this is my way of staying in the code and keeping up to [1:24:46] speed on on coding practices so anyway anything else === Wrap-up — how often should we meet? === [1:25:12] What like at what pace, what tempo? [1:25:16] Do you do this monthly? [1:25:18] It's just changing so fast and I love this atmosphere. [1:25:22] You know, just like, bring our all the best ideas together. [1:25:26] Bring your buddy, whoever wants to come. [1:25:28] I'm going to mind the workshop. [1:25:30] Yeah. [1:25:32] Where we actually like... [1:25:34] We actually work and try to integrate it up. [1:25:36] Get at least the basics down. [1:25:40] I think twice monthly and then workshops. [1:25:43] I overthink things up. [1:25:46] So I'll make this short and sweet. === Jeff Davis — training courses from textbooks === [1:26:04] My name is Jeff Davis. [1:26:05] I work for a pest control company based out of Texas. [1:26:07] uh i do all their compliance and licensing stuff um i know the Sumsions because uh i bought a [1:26:14] house recently and they helped me out with that so um i actually on april fool's day sent a picture [1:26:20] of the driveway of the new house and it was graffitied which i just used chat to put in [1:26:26] some fake graffiti and send it to them and the builder and they were freaking out my phone [1:26:30] started ringing and I'm like, dude, it's just, it's just a prank. But anyway, um, this is, uh, [1:26:37] definitely way lower on the intelligence scale than what everybody's been speaking about so far. [1:26:42] But I think it is, uh, some stuff that, that some of you may run into. So in my job, I'm responsible [1:26:48] for making sure everybody's licensed, uh, to perform pest control, which comes along with a [1:26:54] lot of testing, a lot of training. Um, and there's a lot of costs that goes into that. So [1:26:59] recently found, uh, been working in Claude to develop our own training. So we don't have to [1:27:06] purchase them through third parties. Um, and it's actually saved us this year alone, $100,000 [1:27:12] in just training fees through these third parties. So, uh, making the boss happy, [1:27:17] getting a little bit of kickback on the bonus on that as well. So basically all I've done [1:27:22] is reached out. We operate in 13 States. Every state has different requirements for licensing. [1:27:27] I got a hold of their training materials. Sometimes it's a PDF that's 400 pages long [1:27:33] of how to do pest control. Sometimes it's a physical book that I have to order and have [1:27:38] shipped to my house. So what I've done is basically taken those physical books, I unbind them, [1:27:45] take them to a freaking copy center and have them scan in all those into a PDF. And then I upload it [1:27:51] to Claude and tell Claude, hey, build me a SCORM file. [1:27:54] So SCORM file, for those of you that don't know, [1:27:58] which most of you probably do, [1:27:59] it's what LMS runs off of. [1:28:02] You can drop a SCORM file into your learning management system, [1:28:05] and it's just this nice package that has everything in there, [1:28:08] the quizzes, the way you want everything to work, [1:28:11] the timing, all that stuff. [1:28:14] So essentially, I've gone through all 13 states we operate in, [1:28:17] built all these SCORM files, dropped them into our, we use Paylocity for like our HR system [1:28:24] and it can read, it has an LMS tool in there to be able to read SCORM files. [1:28:30] And so I can assign these trainings to all of our employees, our new hires, and then it tells me [1:28:36] when they've completed them and then they go take the test and our pass rate on the tests have gone [1:28:40] way up. Instead of being, you know, like 50% pass rate on these exams that cost 100 bucks each time, [1:28:46] it's like at 90% pass right now. [1:28:48] So essentially I've just used AI to read these textbooks [1:28:53] for me and extract the info, create these small quizzes. [1:28:57] And then I send that off to the new hires [1:29:00] and that's how they prepare for exams [1:29:03] and meet the state requirements. [1:29:04] So kind of some low hanging fruit there, [1:29:08] but any questions on any of that? [1:29:11] Okay, cool. [1:29:12] Well, Alta pest control. [1:29:16] It actually was started by two brothers here at BYU. [1:29:20] They were my college roommates. [1:29:22] So, yep. [1:29:23] So these SCORM files, have you found that like, [1:29:29] so I thought about building like a personal LMS of sorts, right, [1:29:34] to just learn different things. [1:29:35] Mm-hmm. [1:29:37] Have you found that it's like pretty good or is it like, [1:29:42] does it feel like very AI? [1:29:44] it's it's good enough to meet the state's requirements it's this is one right here okay [1:29:52] so it basically i can set you know how long they have to be on this screen for most of these the [1:29:57] screen will read to them what the words are and then you know as they go through them at the end [1:30:02] there's a quiz they have to do i set it so they have to get 70 to pass the quiz so it's it's very [1:30:09] 90s looking in my opinion but uh the third party that we're using is 200 bucks a pop per employee [1:30:15] we had 400 employees in texas this year alone so i mean we're doing this the same thing that [1:30:22] they're doing the next step is we're going to package this hopefully into a website and start [1:30:27] selling it for half the price that all these third parties are and i mean you can get into a lot of [1:30:32] different industries this way make a lot of money real quick uh because there's a lot of companies [1:30:36] that don't want to go down the AI route and build it themselves. [1:30:39] So, yep. [1:30:41] You're saying you paid $200 per employee for a different software. [1:30:45] The same idea still had quizzing and study goals for them. [1:30:48] They were passing 50% of the time. [1:30:51] And now you've made your own and improved it and got nice. [1:30:55] Yeah. [1:30:55] So basically just back and forth with Claude to boil down the data that's in those textbooks. [1:31:02] Those textbooks are 400 pages long. [1:31:03] and those third-party trainings talk about a lot of categories and pest control that we don't [1:31:09] operate in so our employee it's just a waste of time for them to have to do those third-party [1:31:14] trainings but it's required by the state of texas for them to do it so uh essentially we just boil [1:31:20] down our own trainings it meets the state's requirements and it helps our guys pass the [1:31:26] exams a lot quicker. So yeah. Okay. Awesome. That's all I got. Big bonus. [1:31:33] All right guys, that concludes tonight. [1:31:39] uh the food hang around uh eat all the drinks that's their own job thanks for all that songs [1:31:47] thanks for coming [1:31:54] and [1:32:09] Thank you.