Latest / 4kd.ai / Ep 12 | From Experimentation to Implementation: Andrew Kuhn's AI Playbook

Transcript

Ep 12 | From Experimentation to Implementation: Andrew Kuhn's AI Playbook · 4kd.ai

• Jack BeVier (00:00) Hey everybody, welcome to another episode of forked.ai. this is Jack BeVier with Dominion, the Dominion Group. I'm joined as always by my co-host David Moses of Avenue Group Real Estate. David, great to see you, sir. David Moses (00:08) I'm here with the group. I'm joined as always by my co host David Moses with Family Control State. David, Jake, do you play? Great to see you, and we've Jack BeVier (00:18) I feel like the third wheel. But I'm looking forward to a lot of entertainment, Andrew is a good friend, known Andrew for a long time. he's got a couple companies, Sunrise Communities and the Kuhn Investment Group. And I did not realize that Andrew was kind of like on a similar AI path or or some kind of parallel AI path. we've like connected a little bit over the past couple years, once or twice. And then David was we were talking a couple weeks ago, and David was just like, Hey, you know who we should have on? Andrew Kuhn. And I was just like, like, what's he been up to with it? So he's like, No, we gotta get him in. So luckily both of these guys are in Detroit. And so Andrew was able to to go out to David's office and they're there in Detroit. And I'm hanging out and really looking forward to the conversation. So David, I'll hand the mic to you. What what do we want to dig into here? David Moses (01:07) So my conversations with Andrew, obviously, Andrew and I have known each other for a very long time. You know, fift fifteen, sixteen. Yeah, fifteen plus years. Yep. And actually I met Jack through you. Yep. Because you got me into the RIR group and and that's how I met Jack. but we got to talking, you had built this actually at that time you were just building this this monstrous server and system. And the whole point was to start to implement AI in your business. And you developed a what you called your chief of staff who who is an AI agent named Jarvis. Yep. And so I kinda wanted to first of all start with today. And where is Jarvis today? What does he do? what does he, you know How how do you use him in the entire ecosystem of your companies? Yeah. And then then I want to kind of go back and do a a post mortem of, you know, kind of like how you got to this point. Like how how how did it come, how did it evolve? How did you get started? feel free to be very technical and and just just kind of kind of let us know because this is a this is we we do a lot of kind of how to on here. because we're all kind of playing with it. I think we all at some point came to the realization that, you know, we all looked around for experts who can help us do this right. Yep. And to varying degrees, we found some people who would be helpful and then realized they're just figuring it out too. No one really knows. Yep. And the people who do know really tell us all the time, like that. Even the engineers at Anthropic and at OpenAI are basically just telling us, hey, we're just figuring this out too. We don't know what it's gonna do. And then we have to figure it out and pivot. Yeah. so with that, fire away. No, absolutely. So thanks for having me here, David. Good to see you again, Jack. so for everybody out there listening and watching, my name is Andrew Kuhn. my primary business, I own a company, Sunrise Communities. we acquire, own, and operate multifamily department communities here in Michigan and growing to the Midwest. so my background, and I'll zoom out from Jarvis as a whole to technology and AI, just as a fundamental thesis and a background. So growing up, definitely had a nerdy side of me, you know, love technology, used to build my own computers, that type of thing. Always had a fascination with that. also coupled with the fact that I, in any business, and David, you'd known me long enough to know that like part of being successful, especially in a very physical real world. business like real estate is being able to identify either inefficiencies or advantages into take advantage of whether it be a creative deal structuring or whatever, right? So, anyways, with a couple years ago and I've been a student, I mentioned him before we went on air, gentleman by the name of Peter Diamondis. I've been a student of his work for 10 plus years. And Peter's best known for Moonshots and the X Prize, which he founded 20 plus years ago. which was the first prize to put private people in space, right? which he's been a mentor and friend of Elon Musk for 25 plus years. And and now, if you pay attention to his world is Abundance A360 group, which I'm a member of, it's the nexus of everything: AI, robotics, longevity. one of the coolest new moon shots, which I'm excited about, is education or you know, evolving education through technology. But really, the top of the top of the world. Cutting edge, what's going on on the bleeding edge of AI. And followed his stuff for a long time. And a couple of years ago, he wrote a book talking about the exponential future and said that, you know, for all of humanity, we have been on a linear trajectory of growth. And if you go back to math, we are now, with the advent of AI and evolving technologies, getting into what's kind of an exponential growth curve. And it's really hard for humans to understand, but things, the pro. Speed of progress is accelerating so rapidly that humans are actually having a hard time understanding it. The amount of change is so funny. I was just listening to a podcast this morning. and they basically said that we are on a trajectory to experience a hundred years of innovation in the next 10 years. And so, as you and I both know, and Jack as well, AI as a technology as a whole is one of We will look back in humanity as one of the great inflection points of our species and everything. And so being a techie nerd guy a couple of years ago, I was like, this is awesome. How do we use this in a physical world business? Because I think one of the challenges is with AI is that, you know, everybody hears about it. And then we have, you know, we have a whole bunch of societal things going on between data centers and younger generations and, you know, all the Doomers versus the Optimists. In regards to technology, but most of the people talking about AI talk about it in the phys in like the technical world businesses, right? So it's the model companies, whether it be a frontier lab, it is someone creating software or whatever. Robotics. Yeah, robot. I mean, anything, right? Like and quite honestly, robotics and self-driving cars are two very good examples of taking AI as a technology or SI, I guess as we're calling it now, superintelligence, right? And and putting it in very physical world businesses. and the three of us, all being in real estate owning real estate businesses, know that real estate as a whole is a very slow to move physical world business, right? We literally provide housing as a service for humans in the real world. We're not in the metaverse, we're not, you know, online or anything. So knowing that being able to embrace this technology would be able to help. was really the first impetus to try and figure out like what can I do with it, right? So to catch you up to today to answer. So I'm a big Marvel movie fan nerd and you know finally going regoing through the whole series right now with my kids, which is awesome. always loved Tony Stark and Iron Man. And if you remember his first creation was Jarvis, right? And Jarvis was the all-knowing technology that really helped build his superpowers. and was something that Tony Stark created. Now I know that it's being widely adopted and it's so funny. I was sharing it, like I was just on a flight back with Peter. He named his Skippy. he originally actually named it Jarvis, but that changed it due to just he knew a lot of people to be embracing that and also the fact that Skippy comes from like a sci-fi book that he read that was a kind of a just a unique character. and so I've I played with different names, but Jarvis is stock, and you know, Jarvis is his own person. So within our system, I will zoom out from just Jarvis to say, what are we at today? So my goal as an organization is to completely replace our third party tech stack with all custom software programming that is built to be a AI native or SI native. organization that gives my people, my physical world employees, we call them team members, our teammates, superpowers, right? So I want to remove the hard work, the technique, not the hard work, but really the work that humans either don't like doing or could be do done better with technology so that the people that work at our organization can do the things that humans are genuinely great at, connecting with our residents, bringing value, you know, and really helping run the day to day of the operations. And so that was the thesis that Jarvis began. And so today I now have Jarvis as a chief of staff, but he's really much more than just a chief of staff. It's kind of a misnomer at this point. It's almost I there's so many things he does, right? So to give you an idea, we have over 200 automations in our organization. We have a completely interconnected via vector database system where every single data vault or vertical of information of our software has all been connected in-house. we have multiple layers of redundancies and automations. My entire team has a Microsoft native agent called Sunny, which is our internal digital employer agent that is not only connected with all the information, but has the proper permissionings for the various roles in our organization and is operating in a secure environment to do Tasks, information recall, advice, that type of thing for our our our internal team members. Jarvis is a chief of staff, not only has knowledge and oversight and permissioning of all of our software systems, direct API accesses, to be able to do anything, but also do quite a bit of work that you would not even expect or understand for a modern day organization. I can share some examples of that, but that's kind of an overview right now. Let me let me just ask a couple questions there. So when it comes to the permissioning. Yep. because I think a lot of people are, you know, security's like you know the biggest reason I think the biggest obstacle especially for for those of you who use culture index high D folks. So which you are. Yeah I'm a culture index person by the way. And and it So so in in that vein, do you use the Microsoft ecosystem to maintain the those permission levels? So essentially when Sunny, when Sunny looks at what can this person see, what can this person do, are they basically just looking at Microsoft and saying, what has have they been given permission to see in Microsoft? So good question. How do you drive permissioning? So and that's one of the real challenges, right? So I think the two things that people fundamentally misunderstand or don't pay enough attention to is number one security. So both cybersecurity externally and internal protections, but also governance and the legality of it and the ever changing laws. And you know my legislative background, so we can talk about that too, because that's changing every day. But for permissioning, there are really two main factors. So number one, I chose Microsoft as an ecosystem. One because it is the native business environment operating system for lack of a better term. term that we've always I grew up in between Outlook and Excel and Word. And quite honestly, for large enterprise businesses, Microsoft has the best secure environment with infinitely customizable pro, you know, customizing programming for any organization. Now the problem is is most people only know Microsoft for like its email program, which is Outlook, Excel because of spreadsheets, Microsoft Word. But if you ever take the time, yeah, and yeah, and that right. But there's really there's a ton of different apps and functionality and permissioning that can scale up to the biggest companies in the world to the smallest. And so we knew my thesis from day one, and we can talk about local models and why I build a server on site and all of that. but the whole perm the whole thing was I needed a container to be able to have levels of control, but also permissions. So Microsoft is one. So however, people are categorized within your Microsoft, ie SharePoint. So SharePoint is where a lot of permissioning is driven from. Like what silos of information are people able to have access to? But the other one is actually our property management software. So we're st we're an app folio customer. been happy with it over, I think we're on year six now, and tried all different levels, all different customizations. And really those are the two things that drive permissioning levels on the individual within an organization. That's awesome. I mean, that's Jack BeVier (12:38) What's David Moses (12:39) and go ahead. No. Jack BeVier (12:40) what what's your system of record? David Moses (12:44) for in which way? Jack BeVier (12:46) I guess what's like your company's core data that you are pulling from? Is it are are are you pulling from like, hey, we I don't know, we we keep some data in Appfolio, we keep some data in this other thing, we keep some data in this other thing, we bring it all together in a Postgres database, like which David Moses (13:00) Yeah. Jack BeVier (13:01) is very similar to what data what David has done. or yeah, I I guess that's my question. David Moses (13:06) Yeah, no, no, great question. So basically, yes. So currently until we can replace software, and I don't know if Microsoft honestly would ever be replaced, but what I will say is I do believe software is no longer a moat, right? Because people can literally vibe code. You can literally have your agent. I've done this multiple times where I was building one of the versions of my underwriting model and we took the top five out there. I scenarioed them out and I had Jarvis and I go through them all and and basically mapped out the whole functionality, then we build it ourselves with all the best features, right? And so it's just a simple use case. But what we do is because App Folio is a good data store for all our PII, and that's another thing is PII, right? So for personal information and what can be exposed and that, because we all deal in sensitive information, we leave our Appfolio information there. There are two APIs within App Folio. There is a reports only API and there's a developer's API. And so we connect our system to both so that we can not only pull the pre-drafted reports, but we can also go into the thousands of endpoints that it has. And then also same thing with Microsoft. SharePoint is our online file of record for all the information. So the templates, the forms, all that stuff. And then it's pulled to our local server that we keep on site. And I have multiple redundancies. We can talk about that on-site, off-site, cybersecurity, all that other stuff. But for processing and an analysis of information. Yeah, that's I Jack BeVier (14:33) Sh is Sharefile what's what's document repository? Sharefile? David Moses (14:37) SharePoint. SharePoint. SharePoint. And yeah. Jack BeVier (14:39) Share point. okay, gotcha. David Moses (14:40) Yeah. Yeah. It's called SharePoint. We we we s similarly, we left Sharefile mostly just 'cause of its you know i it's clunky. but but the what what I will say is that if you'd asked me three months ago the same question Jack just asked, I would have a very strong hey, we have These Mongo like we use Mongo. It's a it's a relational database that's n no SQL. So basically it's just in each individual record, it only maintains the data that you actually have. It doesn't have a bunch of blank fields for things you don't have. that's really the you know, so it makes it very, very fast to write to it. So the and which is mostly we're writing data. We're very you know, I mean we we write a hell of a lot more data than we ever Yeah. You know, pull out. But but what I will say today, different from three months ago, is that I don't really know. I I now give the agents, yes, it's within the Mongo ecosystem, but I give the agents the agency to make the decisions on how they want to organize the data. Is this a new collection? Is this within an existing collection? how do the collections kind of relate to each other? because I found that how it stores the data, how it logically kind of goes through how to store the data. is less meaningful today than it used to be. It is. it it's important that it's organized and that all the data is there. But you can, you know, AI is just so good at taking data structured in whatever way, reasonable, logical way it was, you know, it that it was that the agent in implemented it in that particular moment. And it can be extracted later in an in a a myriad of different forms and still fully understood without latency and without and and without hallucination. Yeah. So yeah. That that it's a I just almost don't care how they're building it. Yeah. And quite honestly, no, honestly, and that's one thing I think people don't understand is that the frontier models are evolving so fast and are getting so good. The average IQ is is above 150, 160 now that the reality is is that pretty soon we're gonna cross the threshold. And we a lot of people, a lot of the front end leaders in frontier model land already saying we have or we've crossed AGI where honestly, so the question, the answer is always within the question. So if you have a question, ask AI and it will give you the answer. So learning that, learning prompting, but also not being so rigid in structure within how the human brain processes answers and letting the algorithm and the frontier model figure out the best way. Cause I promise you it's going to figure out an infinitely better way to solve your problem. And so really your thoughts, your questions, your prompts all need to be goal oriented. this is what I'm trying to achieve without many restrictions on how you do it, because it is ultimately going to find out a better way than you could. That's right. You can devise it. And it's just like it's you know, and and it'll show you, it'll prove to you that that you know, the way and you're so certain that the way you design it that you designed it was was the right way and then it it As long as it has the right context, as long as it knows what the goal is, that was a very important point that you made. If it doesn't understand what the goal is, then it really won't build it the way that, you know, forward thinking, the way that it really is going to be implemented and used. So it needs to know the goal. It needs to know it does how this is what good looks like, right? Yeah. Exactly. Much like give it examples. Much like it much like a new team member you hire, like you need to be able to show them what good looks like and and be clear. And same thing with kids, like in parenting. Like I want my kids to understand what done looks like, what good looks like, you know, and what the expected outcome is. And if you give them that and let them figure out the way and use its creativity, it will it will do a better job than you really could only. And it's so counterintuitive to how we think. It's like if you if you're learning to golf, right, and you've never shanked a ball in your entire life, or you've never sliced a ball in your entire life, you've never had to address it and figure out how to correct that. Yeah. And so that's how our brains think. But AI is not like that, right? AI doesn't shank the ball unless you give it parameters that allow it to even understand that that's an option. Yeah. Right. It it's just give it good examples. Don't tell it what not to do. Tell it what success looks like. And it will it is wildly more accurate. It is. And and useful than to, you know, do what our brains do, which is worry about. all all of the potential pitfalls and all of the potential you know, where am I gonna step in today? Like that's just not how AI thinks. It's like this is what the goal is, this is what success looks like, give it examples and let it do it. Yep. Yeah. Absolutely. Jack BeVier (19:23) So have you been what's the harness behind Jarvis? David Moses (19:28) Yeah, good question. So currently I am so as my daily driver, and it has been this way for a while. Claude Code from Anthropic has been really how I've set everything up for the majority. and we can get back into the beginning and what the progression has looked like because this is this is thousand plus. I'm probably closing on two thousand hours worth of work now on this. Which is which is really interesting. But to say on top of running a company and you know, running the apartment owners association with the state of Michigan, all this other stuff. But anyways, so harness is primarily my main or from a harness standpoint, anthropic with clawed code into various models. Right now, you know, we're doing five point five, hope is five point five, fables, okay, but quite honestly, as of today, that is the best model for our daily driver now. I have built in many automations and when I build, I have a very evolved process for when it comes to coding or doing build projects, where I do not only a pre-mortem where I'll go out and ask it what should we have designed six months from now. I also do cross model reviews on everything. So basically I will run a code 5.5. I I need to build something. It will then do a cross model check with ChatGPT on codex on Astra or whatever the latest model is. And then it will also run the Groc 4.7 to do a third model. And the reason being is the separate model families have all different blind spots. And so not only will it be validation of a poor build quality, there are minimum build thresholds. There's, you know, software programming master techniques that you should always implement as a basic of good coding. and so it's amazing how deep and thorough you can get. So to increase your likelihood of one shotting. when you're running through this. And so you know, there's a couple other things to talk about with this, right? So not only on models, but how do you set it up within an organization? So the difference between API and and flat rate accounts, right? So for me personally, now this is a little bit overkill, but I have three separate 20x max accounts for Claud Code. I've got a Codecs 200 Pro, I think we call it Grok 200 a month code, and then my company driving Sunny. because of the way that the app is set up or the you know the digital agent set up had to be a direct API. And so you couldn't like use an individual, you know, flat rate that's gonna get access to a whole bunch of people. and so that's a that's an API setup. And so is that okay, so you don't have like a harness like like Ermes or means or anything. No. And the reason being within that traditional sense is because I have made the conscious decision that Jarvis is an internal facing permissions-based tool, really, that I have varying levels of approvals for for autonomous actions. And so most of it is driven or automations because I'm not just gonna give Jarvis my credit card and say, go do this. But like where the use cases come in without having to use a third party harness is it it allows me to be in control of driving the efficiency and the automations that I want. Cause my whole goal is how do I make myself one of the what they call the the thousand X engineers, right? How do I make myself so wildly productive and then make the other people in our organization so wildly productive? And you can only do that when, or at least in my my current experience has been it's best happens when I'm still driving the majority of the direction. So what's the UI? Like like are you in are you in cloud code? Yeah. Jack BeVier (23:12) Yeah, is it VS Code or terminal or? David Moses (23:15) Yeah. So so I'm I'm a I'm an old school terminal guy. Okay. So I'll run, you know, anywhere between so I you have a dual screen setup. I have a going back to the server thing. So right now if you look at the current state of the market. The frontier models in the companies, right? We have a couple of things going on. So we have a ton of VC money that's pumping up these valuations and they're at a race. So token usage cost is being subsidized on all these flat rate accounts, right? And we're getting, I mean, I probably spend $40,000, $50,000 a month of compute on my flat rate accounts and everything. Cause I max them out. I mean, I I I'll burn through them, you know, not even five hour limits, but I'll burn through week limits in a day or two sometimes. when I'm doing heavy and I'll run six to eight. 10 terminals at a time, all on separate projects. I have a basically we build an on-site server, and currently I've got a a Z7 processor, a four terabyte hard drive, an Nvidia, RTX Pro 5000 with 48 gigabytes of VRAM, 512 megabytes of RAM. I mean, I build a you know 25,000, $30,000 server. and the reason being is because I wanted to be able to drive local models if this if that gets there. Yeah, if the switch gets there that we can drive and I don't have to have this exponential compute because it does show up in my GNA costs, right? And so like I I'm not gonna start running super high GNA costs f because of all this software tech and everything and and not have a you know reasonable return on them. So and it'll burn tokens. on things that are wasteful. Yes. That's that's the that's the main issue that I've you have run into. I I also I use Hermes as my harness. Yep. And I run because you can't use you you can't use anthropic correct. You know, but Grok now Groc is allowing you to and and and open AI. Yep. So I've got one Grok Pro account and three open AI codex accounts. And then we use Telegram and Discord. That's our that's our way of interacting with the system. But I love the idea. You said that that your users, when they interact with Sunny, they're interacting through a Microsoft. Through Teams. Yeah. So through Teams. Yeah. So we basically built a Microsoft agent that is basically an app, an app based cloud based agent. How do you build that in Azure or in Azure, yeah. So everything's held up in Azure. And also that's the nice thing is like, so when I asked Jarvis early on, I said, Hey, we need to up our cybersecurity, go through and he has full permissions. So basically he has graph-based full permissions to read, write, and a lot of different things in the Microsoft ecosystem. And so basically he helped me design Sunny as an app that is an interface, lives in Microsoft Teams, and she's just another person. She can join other conversations with people. They can do one-on-one. She can put things on people's calendars. She can draft emails for them. And so my goal, because my team currently has everyone does have a 20, well, I guess it's $25 a month now, Chat GPT account. my goal with Sunny was to replace that with a more user-friendly interface that they like. And so that was my attempt with Sunny. Now, the friction there is that chat-based agents. don't have the functionality of a website. But you got to remember every frontier model company out there, much like Facebook and the other social media companies, wants you living in their world, right? So so they're making it easier for you to use their app on your phone. And I still use Claude to drive Codecs and Jarvis a lot, but I've built, I built custom stuff and we can talk about that later. They can use Teams on their phone. But you can use Teams on your phone. You can talk through Teams. One of the things we did is I also run a Mac Mini with multiple accounts. So Jarvis has his own iCloud account. I have my iCloud account. So Jarvis can read my text messages, my emails, you know, proactively do responses and that type of thing. but one of the things there's a company out there called Instinct, which is backed by Sierra Labs, which is really cool. And the interface is just Messenger. And so it's it but it connects through MCPs to all your information. So basically I had Jarvis, we we trialed it after Moonshots. And I was like, hey, check this out. And then actually I had Jarvis texting Instinct and my agent talking to their agent testing him. Cause he was like, I want to see if you know he can do this and if he has reasoning and if he has memory. And so then we build it within iMessage and then iTunes account account. And now I use i messenger or you know, iMessages as well on my phone. That's awesome. That's right. Yeah. I mean so it it's a very yeah, I mean it's just a whole world like I there's a whole world rabble holes that we can do. Like every single Jack BeVier (27:51) Dude, it's David Moses (27:52) thing you said, like like I I'm jump one thing I jumped down the rabbit hole yesterday on was that my my team. So I've given almost everybody on my team now our our our rock for for Q well I g it's Q three but it ends like wrapping Q three at us tomorrow so yeah our quarterly like is beginning in November. So it we we will everybody will have an agent that they can interact with. Love it. I have one like an agent per person because I use it's easy to use Microsoft to just have them, they just click and then all their permissions are their their permissions. Very much true. So the but what I have them doing is I'm building a very kind of simple be because over time it will know what emails were written by them, what emails were edited by them versus what the AI wrote. And I want to take their writing samples before everything is written by AI. Yeah. And really bring their voice into right. So so it and and it can't be, to be honest, it doesn't like you could give it a 10,000 tokens of of sample writing in an email and it's still gonna come through kind of like generic AI So I think what because it it may know how to kind of use your lingo, but it's not responding to things specifically the way you would respond to them. you know, it's too vague, it's too generic, it's too high level. So we're actually gonna go through and say, okay, we're gonna have maybe it's a million tokens of emails that they've written over time, you know, actually categorized in Mongo. Mongo has a vector database essentially where it's like any any Mongo collection can have one field that is auto embedded. As a vector database. So then it can use those as okay. It gets context if you're writing. This person asked a question about this thing. This is how it was responded to all these times. And it can really start to understand not just how you speak, but what direction are you supposed to go in to answer this question or to respond to this inquiry? and and I I'm hoping I haven't implemented it yet, but it it's conceptually that's that's what we're building and f Fable is the. I always go to Fable first just for when I'm trying to develop something new. If if it's like abstract creativity than the Opus series of models is currently Jack BeVier (30:24) Yeah, novel stuff, yeah. David Moses (30:26) Exactly. so so I always go there first to kind of like give me the give me the game plan. Love it. and then I I give it to, you know, my my agent is Rufus from Rufus, yeah. From from Bill and Ted's excellent adventure, the George Carlins character that that guides Bill and Ted through time. And so Rufus just will take that plan and implement it. So so that's that's the plan. One thing thanks for bringing up about Fable. One thing you guys can do too that I found well, okay, so there's a little bit of background story to this. first off, 90 plus percent of AI initiatives that most companies launch fail. Second, there the most of the recent studies that have been coming out have proven that if you just give people Chat GPT or whatever, There's good A ways to use AI, there's bad ways to use AI. And really the thing that companies are struggling with now is figuring out how do I productive like get my company to be more productive and efficient and actually get real dollar gains out of this productivity and not just give them more tokens to spend because what happens is if you give people an unlimited amount of of tokens or bandwidth or whatever, they'll just be on that all day. Right. And so and you really won't drive your initiatives and your goals and your rocks. It's diminishing. Yeah. It's very diminishing returns. And honestly, it can even be detrimental. And so it tangential to that, but also related is the fact that it does help, especially when you're setting up your model systems for people, to have in your case, Rufus or whoever your main, you know, your frontier model is, to go through and ask the system to categorize the tasks or automations that you have and then match them with the appropriate model to maximize token efficiency, but still have a success rate high enough. And so what it does is all 200 of my automations don't just run on one model. They're actually structured and tiered based on the type of work, how complicated it is. And then also we've set up an automation, or I have, anyways, so that every every couple of weeks, I think is currently set up, it checks to see what the latest frontier models are, reruns the analysis so that I don't have to do things. Right. So one of the one of the keys in AI and technology is like you don't want to have to do more of everything. Like you want to be able to build. the automations so that they're future proof so that it doesn't get dry steal contacts and context is everything as we know. Right. Information is everything. So so the in in in that vein, we are now going back through all of our automations and we are scoping API key for for automation. So we can track cost at the, you know, and I wish, honestly, I wish and I don't know why you can't have with chat or with or with any of the you know the the frontier models you can't have where an AI can actually call for a an API key. Yeah. Like it should be able to create a new API. You should have some kind of master level to create a you an API key because this is like really tedious work. We have to we will essentially have a workflow and that workflow will have a scoped API key. But what happens when you side that workflow really is going to work best on Gemini for or work is gonna work best on which I only use for video and and pictures picture analysis because it's it it it'll take a big video and and analyze it. But but it what do you do when when you when you're switching model family? but again it's like API key for specific and then you can track costs at a granular level and then You're looking at it from, okay, let's pull this out. There's also another thing that I'm looking into, which is the dis the the new decision API that open AI's I there it's either out or it's coming out, where it basically just takes a binary yes, no or a simple, like it's it it's a super cheap API and it's not for giving like how many times do you ask a question and you're expecting a just an answer and you get like three paragraphs and it's like no. I this is a quick decision I need to make. Do I do it this way or that way? Or you know, just what's the answer to my question? Like I if if I ask you know, w what pests are an issue in whatever city, yeah, like I don't want like you know the history of the city, yeah, everything else. Like I don't need everything. I just need to know, like, yeah, you know, am I gonna buy a house with a crawl space in this city? Or are there rat and you know, and other type of of rodent issues that are gonna, you know, create create huge issues for me. Yeah. Yeah. Jack BeVier (34:47) Make that help. Yeah. David Moses (34:49) But anyway, so so I I I digress, but that's getting getting it to the point where it's individual workflow has an individual API key. I'm tracking costs down to that automation. So when I wind up with a big bill for tokens, I can look at it and say, okay, I know exactly, you know, this this token has this much spend. You know, in this time frame and that alerts before I wind up spending a bunch of money. Makes total sense. Like one of the things that people in general think that AI is going to replace everybody's jobs and everybody's gonna be jobless and all this stuff. The interesting thing that we found though is that the more we lean into AI, we don't do less work, we end up doing more. But what I found is that we can do the work and productivity of a much bigger team now. Right. With the use of technology. So I haven't been letting people go. Honestly, I've been bringing more people on. And really, what it's done is allowed smaller companies to compete with the biggest of the bigs. Right. Because the advantages that they used to have because of the extra bodies or the extra revenue to have the additional, you know, the whole acquisitions team of eight people. And, you know, I got my head and I got two directors and I got three analysts and I got two outbound call guys. Like I can all do that, you know, with technology now. To the nth degree. Like we're high, it's funny, I'm hiring an acquisitions guy right now. And I can't tell you how many people are applying as an acquisitions analyst. And I keep sending these kids these responses. I'm like, look, I'm just telling you, technology is making, I don't want to say it's making it obsolete, but like that is a skill set that will not be in demand. What will be in demand is the ability to have conversations with humans and be able to get a deal over the finish line, come up with creative solutions, work with the bank, you know what I mean? Build trust and rapport and than that type of thing. And so like the real world human skills are becoming more important and they will continue to become more and more important as this moves on. Yeah, and and and and increasing the value of the humans that have that gift for human to human interaction. Because these people are generally, you know, you you take the good with the bad. You have somebody who's got who really will emphasize the relationship over anything else with your company. but maybe they're not the best at returning phone calls or or returning emails. And oftentimes it's just using AI to be that reminder for them. Yeah. Right. To be that squeaky wheel. Because unfortunately, you have this squeaky wheel problem in every business. You have customers, you have people and and you wind up focusing 80% of your time on 20% of the customers. That's just normal because some people overcommunicate, some people make a mountain out of a molehill, some people whatever, right? But you want you want these things to be somehow, you know, you want some objectivity inserted in that. You do. So you don't have your best people who are relationship people spending time on relationships that you know, doing things that are just to satisfy somebody who's really just a time suck. Yep. And neglecting somebody who's a good relationship but isn't going to be that pest. Yeah. Right. And so this creates this, you know. And it's not just only about money, right? It's about reputation. It's about you know, and that's oftentimes why we spend time with the people who are the loudest. but but that's that's a big thing. It's it's it's making sure injecting some productivity gains in people who are really that valuable. The thing that you mentioned, those people are going to be valuable where the engineer and the analyst and the insert your Yeah. You know. Yep. You so so one thing's gonna h one or the other. Either the engineers are going to have to learn the people skills necessary to be valuable in a world in which we don't need people crunching numbers, or we need the people who are, you know, have that gift for communication and use the tools to fill in where they are where where their natural shortcomings are. Yeah. and so I you know, I I feel like you could do a lot with culture index and with AI in terms of just understanding how people function and how they don't function so that you could have an agent understand this is this per like this person is a low D, right? Like this this email from this one guy on this big deal that didn't get answered for three days is n you know, it's not gonna be important until that guy the cause that guy's not he's not bugging you. Right. He's just gonna leave. He's just gonna go with somebody else. Yep. And you're never gonna know he's a quiet quitter, you know, and and that's it. And by the way, Jack, are you a culture index company? Jack BeVier (39:29) yeah. Yeah. David Moses (39:30) Okay, awesome. All right. So I all right, my consultant's probably gonna kill me for this. Sumit, love you, man. But so so culture index gr probably the best statistically accurate, right? Of someone's actual hardwiring. Yep. And we have implemented so Jarvis and really the entire ecosystem understands what I call the sunrise operating system, for lack of a better term, or the sunrise communities operating system. understands and everybody's culture index is in there. We also gave it as much context and background and understanding on reading culture index, the images, as well as everything else for our people, so that it does not only as part of our interviewing process. But also as our individuals themselves, when we're having meetings, level tens, that type of thing, it's like, hey, look, just so you know, this person is this, their D's obviously above the line, they're, you know, this type of thing to help make our communication better and more effective so that it's better received by the individuals. And so it is ripe for utilization within an organization, like in you know what I mean? And really integrating culture index within your technology because it is again the best. Predicator. So funny. So our acquisition analysis has only been up for a week and a half. We've had 70 something applications. And Nicole, who sits in our people in culture seat, she's she, you know, makes all the higher fire decisions. And it's so funny because I get a lot of young guys that are like reaching out to me and they want to interview. And then I'm like, well, you gotta fill out the application, take the culture index. And I can't tell you how many times I'm like, no, I want to interview them. And then I look and they're like a philosopher in a 7% match. And I'm like, no, or like a craftsman. And it kills me because our minimum threshold is that like for me to even interview, right, you have to be at least a sixty percent match and close to our C job. And so, like, and that's that's to my lack of experience and trying to get that. But I know that, you know, eight months in and it's so funny because I so I don't know about you guys, but I Jarvis for me, because I've given him So much context of the last 25 years of my life, both personal and professional. I have created so many coaching personalities through the whole arc of someone's work professionally that I admire out there. Will then run that as a filter, right? They he knows our core values. And I will say, Jarvis, I'm really struggling with this. I want to talk to these guys, but they're, you know, you can look at the results. Go, go pull up the results and look at the. spreadsheet and you know I give him access to every site so he controls Chrome and he'll log into sites and you know he'll build spreadsheets of who's taken it, who hasn't, who needs reminders, all this stuff. And I'm like, how do I handle this? He's like, Andrew. And the amazing thing is he'll be like, hey, on this call with your culture index consultant two months ago, he said, if it's not above a 60, I'm not even bothering my time with him. He's like, Do you really need to go through this again? And so like he will actually be a Jack BeVier (42:26) Ha ha. David Moses (42:26) great coach and push back. And that comes to the soul file That comes to designator. I absolutely that yeah co-creator. So I consider him as a co-creator whose singular goal in life is to help us achieve our goals within the organization, right? Yeah. And so he has the autonomy to push back on my ideas, my beliefs, to bring other viewpoints. And then I've actually, this is another. So if you guys want to talk to use cases, I got a million use cases for you. But I, you know, I I have a big library at my house and I love books. Always have. Ever since when I was like 20 years old and I heard Jim Rohn on a cassette tape say there's a reason people with and this is in the nineties and early two thousands, like there's a reason people with homes over $250,000 have a library in them. And at that time, apparently $250,000 was a lot for a house, right? Now it's like a not even a starter home. But anyways, so yeah, this was this was early on. This was early this year, but I went through my house and I I have, I don't know, at least a thousand physical books in our in our house plus. And I went through and I took pictures of our library and everywhere I have stacks of books. There's all of our house. And then I uploaded the photos of Jarvis and I said, Jarvis, I want you to organize every picture, every title you see in here. And then, because most of them are business, I'm a very heavy nonfiction reader, and go through, pull the frameworks from every single book you have, and then apply them to Sunrise on how they could be applicable. So now he has a knowledge base of all this business context. And same thing for business coaches, that I that I like, you know what I mean? Yeah. Go online, build the profile, and then coach through that, through that mindset for me of where I'm struggling with this right now. And so honestly, as a thought partner, I think it's as valuable as a resource as of anything as a CO leading organization. Do you think culture index can be modernized? improved upon. So I think the I think the so I love culture index and like I said, love the company and making sure I'm staying within their terms of service as a as a user of it. I do believe that they could evolve the technology and the user interface and the online platform to be better and more efficient. Maybe have an API. Yeah, maybe have an API or something that makes it easier to use. but it is still a wildly productive tool that we use every day. No, I love it too. I like there Jack BeVier (44:41) I bought it David Moses (44:42) there is there is a small part of me that looks at Looks at it and says, you know, man, the more a the better AI gets, the more the socializer just looks like the perfect graph for everything. Jack BeVier (44:59) Yeah, it's really interesting. David Moses (45:04) Because it's like th those are the the feedback we get from our socializers is always, we love this person. Like even when they drop the ball, we would never leave them because. They care about us and that matters. Jack, what profile are you by the way? Jack BeVier (45:17) Yeah, but but it but in the context of in the in the context of an organization, my experience is that socializers, if you give them a list of seven things to do, they will not. They they will do four and a half and then like and David Moses (45:27) correct. Yes. Jack BeVier (45:29) then squirrel and you know and you know they make a friend, you know, with the squirrel. Like David Moses (45:33) Well wait, but but which four and a half? Because that's what's important. Because Jack BeVier (45:36) Yeah. David Moses (45:37) because if you can use AI to augment, you know, if they're only gonna get four and a half done, you Jack BeVier (45:43) Yeah. To augment. Mm-hmm. David Moses (45:45) right, it to make sure that they understand the people and the needs of those people, because that's what motivates them, right? So Jack BeVier (45:55) Mm-hmm. David Moses (45:55) if it if they understand that they're gonna let someone down or that somebody is you know, really does need this thing to be done. That thing's gonna be on that list of things that get done. Well, David, Jack BeVier (46:03) Yeah, that will motivate them. Yeah. David Moses (46:06) you're a rainmaker. I am, yes. Yeah, you're a rainmaker. And then Jack, what's your profile? Jack BeVier (46:13) I'm a very extreme architect. David Moses (46:15) Okay, so s welcome to the club, friend. So I was architect two, like three deviations to the right on my A as well as my D and then a ten logic and like an eight I with an EU of fifty Jack BeVier (46:26) Yeah, that's it. David Moses (46:27) six, I think. So like like I am the ultimate system buildings, right? And this is why I lean to this, and this is something to keep in mind in in the whole context of AI S I S G I, I guess. Anyways, super general intelligence. basically the fact that not everybody is a tech nerd and knows AI and is automatically gonna trust it and use it. Everybody in your organization to be a digital native company needs to have an aha moment where it makes their life better in a measurable way to them, right? Does or does something that's really hard or monotonous for them. So I'll give you a real world example. Actually, two real world examples. So Nicole being chief of people and culture. Also, my wife, like I love working with my wife. It's hard for a lot of people, but it works great for us because she's a coordinator and we have completely opposite things. She's a great utility player. But, anyways, we had to let go an employee two weeks ago. And we walked in, we did the exit interview, and then two minutes later, I Hey Jarvis, we just let go so-and-so, run the offboarding procedure. In two to three minutes, he had gone through. shut down permissions on everything, taken a Microsoft profile, moved it to legal liability backup, had searched, pulled everything for downloads, had frozen everything, gave next steps to every individual in our organization for check truck out, clean this, do that, all within maybe a five minute span in its entirety, right? another that's an automation. That was an automation that he helped build for an offboarding. And we had used him previously and he gets better every time and it adds something to his checklist, but that's the new standard. same thing. One of my there's a department rock actually where and this gets into the technical minutia, but my team was having problems tracking recurring work orders, right? For preventative maintenance specifically. So they wanted to know which vendors and they were trying to come up with all these solutions, and they were gonna create fake users with just a blank email and all this stuff as a workaround. And actually, I had the team come into my office yesterday. And I said, Hey, Jarvis, read everybody. Because I asked them, I said, give me some context on what you're trying to solve. I had Jarvis read my emails on it and pull the people who were in the room. And we logged into App Folio, we logged into Microsoft. it pulled all of the documentation, it pulled all the endpoints in the API and ended up coming up with a solution because Appfolio natively will not allow you to track recurring preventative maintenance work orders in a standard output, you know, format. And we came up with a solution that solves their goal, does not require any technical or customization, and we'll be redundant in in a solve it. So like we solved their rock in 20 minutes working together with technology, right? And something that we could have never done outside of it. And so, and at that point she goes, my gosh. She goes, I finally understand what Andrew's doing in his office talking to himself all day. Right. That's right. Yeah. So those Jack BeVier (49:19) Yeah. David Moses (49:21) aha moments come and I think they're I mean, yes, they're important. and I think the the biggest frustration that I have once you get people's buy in, once people say, my god, like this is amazing. I love whoever they know. Yeah, whatever. Yeah. Right. there's this kind of subconscious you know, reward s something in the reward center of the brain. yeah. That basically devalues things that are easy. And and I and I think that like there's this kind of inherent bias towards hard work, you know, being more valuable than than easy work. Don't do it easy. And and so there's no way I'm gonna fight against that wiring. That's not possible. But what I can do is allow them to take something that is has now become very, very easy. And ha and challenge them to make it better. So if you're going to do something that is going that used to take you fifteen minutes and now takes you thirty seconds or a minute, how can you make it, how can you spend some time making it so that you take take the whole fifteen minutes, but but make it so that every time it's done. Yeah, the iterations. Right? It's more robust. It accounts for more, you know, potential things that could fall by the wayside. Yeah. and and yeah, I mean it really is we communicate so often and so much that we can't remember all you know, but a small percentage of it. But AI can read all of it. So all those not forget it. And not well, if if you breed if you create contextual memory, so we started with an obsidian vault and markdown files and, you know, we talked about that is built being a early version way almost a year ago now. Of creating contextual memory, right? And when you do that, it does give you the ability to do that. And the frontier models are getting better at it. Again, everybody's realizing that's a fight for data. Yep. to have this persistent memory. But you really hit on something that I think people don't put enough effort in is that asking so, and this is this is just leadership 101 that I think actually it's actually not one on one, it's actually more advanced skill, but like when you're guiding people. making if you really want a high quality of work, usually the first draft of everything is not going to be the highest possible answer. And so being able to give feedback back to the system and be like, is this really the best system? Is this, you know, really world class? And then do it again in asking for multiple iterations till you evolve it. People don't understand that it is a back and forth interactive thing. Yeah, it's an interaction shouldn't be one shotted and there really does need to be evolution and constant iteration and generations of stuff to improve. Much like a sculpture, right? Like again, when you look at David and you know, carving the sculpture out of a block of marble, right? He was in there all the time, but it took iteration after iteration after iteration. Take a little piece away, take a little piece away, evolve it. And so that is truly what world class work looks like is the thousand iterations. Yep. So what's missing? go ahead. No. Jack BeVier (52:32) I have a there the in terms of like you you you guys made a point about like using the extra time to like make the process better, make and you know, make the make you know, improve the quality of the work product, improve the quality of the of the business, right? And like the on the culture index side of things, I'm trying to figure out like at what point can I slip in a sev stealing a customer's seven minutes of a customer's time to get them to do a culture index survey so that I can start to build. To add the CIs into my CRM of all David Moses (53:03) Yeah, can go around. Jack BeVier (53:04) of those humans. So that when my craftsman loan processor writes an email to an architect customer, which for those David Moses (53:12) Yeah. Jack BeVier (53:14) who don't understand CI, it doesn't really matter. All you need to know is that those people are different. and they like often can talk past each other because of those personality David Moses (53:17) you did also Jack BeVier (53:22) differences. That like, like I wanna, you know, like I wanna be able to like translate to customer, right? Like what you said. David Moses (53:28) Yeah. Jack BeVier (53:29) And like in in the context of the entire email, right? You like you have a persuader, a low, a low D salesman, a low detail salesman talking to a technical expert, cu a technical expert customer. The customer asks David Moses (53:41) Yeah. Jack BeVier (53:41) four questions the persuader answers to, like that the David Moses (53:45) Ha ha. Jack BeVier (53:46) that the that they could flag that, David Moses (53:47) Yeah. Jack BeVier (53:48) like, hey, hey, before you, you know, on this draft, like, you know, you know, when the dra when the draft's ready to send, you know, before it leaves the outbox, review it. And flag it if you didn't answer two of the questions and translate what you said into something that a technical expert wants to consume would be just like all of a sudden you would David Moses (53:57) and flag it if you didn't answer to the question and translate what you said to whatever it's functional. Yeah. Jack BeVier (54:07) just all all of the, you know, I wouldn't need a socializer in every chair. I would just you know, I would have a translator, right? Like a personality translator function David Moses (54:10) I would be a good advisor in every chair. I would have a transfer. Jack BeVier (54:17) as you're interacting with the outside world is something that like I wanna figure out how to get that in place. Like David Moses (54:22) That that's the microcosm of what I said. It's like do you take do you do you take the engineer type people and do you get them to the point where they can, you know, really interact well with humans? Or do you take the the people who are really good with humans and and fill in those technical you know, limitations that that that arise. But but I I Jack BeVier (54:42) Yeah. And I think both work, right? Both both work. David Moses (54:45) I think i I don't know that you would need the culture index of the person on the other end. Because I think look, mirror neurons are a are are a real and important thing. And I think that because AI is so good at understanding and reading the tone of a request, I think that your go-to should be respond to this person, right? Respond to this person and meet them at the level of this email that came in. Right. And Jack BeVier (55:15) Yeah, yeah. The way they are. Yeah. Yeah, yeah. David Moses (55:18) and I think if you can do that They'll appreciate it because look, sometimes they'll write an email and they'll be annoyed. Sometimes they just don't have time. Sometimes they send it before they're completely done with it. There's a lot of different things that happen when somebody when you get an email, you don't know what was on the other end going on on the other end. But if you meet them at that level, then it you know, you meet them at the level of their email and respond to them, you know, people and it's a sub it's subconscious. They just they get it, they they, this person really read my email and is addressing the things that I That I put in it and when they look at it and say, What about this and what about that? Like they can look at their own email and say, well, I guess I never asked that. So I guess I can't fault them for it. You know, it brings up a really so two things. Number one, it brings up a very interesting data privacy thing. Now, this is just my personal opinion, but data privacy is dead for the most part. I hate to say For sure it's dead. Privacy I know people don't understand this. Because here's why. Context and information is king. Everything that's going on in the frontier models right now is how do we get more context, right? And they're really starting to push up against human biology and the regard for what's truly private anymore. And what I mean by that is is like, so the recording of conversations, a lot of our communication is verbal, right? Until we have telepathy and can read each other's mind. we still have to communicate. Yeah, right. And so if you remember, you know, 92% of communication is nonverbal, but there's still a lot of conversations that happen. So now what happens? Okay, then you open up recording calls or having meetings and making sure that you're following the state laws for recording and proper permissioning, right? But that's rich context if you're doing that in a legal manner and that everybody is on board with it. also the technology jack, by the way, to like communicate, I mean, that's definitely already there. I don't so a little bit of background. My wife is adopted. she is she was born in Seoul in South Korea. she came over to America when she was two, she does not speak Korean. Her mom still lives in South Korea. We met her a couple years ago. She does not speak English. AirPods, I don't know if you guys know this, but AirPods have live time, and I've it's so funny. I've been buying Kickstarter versions of this for 10 years now, trying to get it to work. Yeah, AirPods, yeah, right. But AirPods now have live time translation for you. You literally put your AirPods in. Open up the translation app, they talk in their language, you hear it in your language. And same thing if they're wearing AirPods. And so I truly believe we're gonna get to a level or I in some ways we already are, but you still have to have context on that person and how they're hardwired or how they communicate and make a best guess because it can go wrong too, right? Like you just like humans can read things wrong, AI can read things wrong too. And so and make assumptions because A lot of the bottles are based on efficiency and still they still make a lot of assumptions and things that obviously we all know by outputs aren't correct. Yeah. Yeah. That's Jack BeVier (58:11) Dude, my my I've I have a droid, my I got a Google up I got an update, system update today, and one of the features is live there's live translation now in Google Translate. And I'm like, dude, like like standard David Moses (58:20) Awesome. See, I love it. I love it. Jack BeVier (58:22) for free? That's amazing. Like so David Moses (58:24) So I I I wonder what else iPhones have now that you'll get in six months or a year. Jack BeVier (58:28) Okay. David Moses (58:30) Wow. Sorry, it's a little little Android dig there. Yeah. Jack BeVier (58:32) Brutal. Brutal. David Moses (58:34) You know, look, like I said, things are evolving so fast that like it's hard for humans to understand the pace of it. But what I will say is that like and this is another experience because here I am in LA last Friday at this moonshots event. So Peter Diamantis is running it, his whole Moonshots pot crew is there, everybody in. Palmer Lucky from Andrew was there. Ben Lamb from Colossus was there, who's the guy who's like taking woolly mammoths and extinct like real world Jurassic Park. Yeah, the direwolf and the woolly mammoth and all these like wild Astro Teller, the head of Google's moonshots, right? So, what you don't realize is how deep in you are and how far ahead you are if you're having these conversations right now, relative to the general populace. Like a lot of people are still just figuring out like. What is ChatGPT or they've heard of AI or and they're they're still so even if they're terrifying utilized in their life, yeah, because they don't know, right? Like the natural human emotion or reaction for something that they don't know is to, you know, pull away from it. but I truly believe that this is going to be one of the greatest tools and like I said, inflection points in our species history, quite honestly. So Yeah, I like. Jack BeVier (59:42) I I'm becoming I'm becoming more and more convinced that like I I I went on this rant, I was talking to somebody at a university the other day, and I'm like, dude, I am convinced that like the the unit cost of creating code has come down a hundred X, a thousand X, right? Like five years ago, I was aware that there were, you know, that our technology as a company was not like state of the art, but then I looked at the cost of creating that technology and I was just like, I just don't think that it's actually worth a squeeze. And so I don't invest in that side. And now, given the unicosts like that have come down so much, I'm like, you would be insane as a company not to develop a core expertise here. And there are, and like, so it it occurs to me that, or I think that, like introduction to programming and software architecture should be core curriculum at business schools. Like at business schools, that's gotta be core curriculum. And what I'm David Moses (1:00:38) Yeah. Jack BeVier (1:00:39) doing right now over the past year is I'm pret I'm learning how to be a CTO. Yeah, seriously. But but I'm learning David Moses (1:00:42) I I I think at high schools it should be Jack BeVier (1:00:47) how to be like every company that has five employees or more, like it would be it's negligent to not understand how to create code and have a mental model for creating code and using that as a tool in your organization. And so, hey, if you took a you got a business degree or you just self taught yourself business. And you have no technical background, you well suck it up, Buttercup. You're gonna have to go learn David Moses (1:01:12) Yeah. Jack BeVier (1:01:12) how to be a CTO. Like, and David Moses (1:01:14) Yep. Jack BeVier (1:01:16) and and it's so interesting because I'm here, like you know, hearing you guys talk about the different tech stacks that each of us have put together. They're like pretty di they're like rather different, but the fundamental David Moses (1:01:24) Yeah. Jack BeVier (1:01:25) concepts are you know, the abstract concepts are all the same. It's just like I use this tool because of this, I use that tool because of that. So the like there's an abstraction layer which is that we have all built. and are all but but the the the way that we execute it is completely different and they're all fine. There's nothing wrong with it. It was just like this one was better for me and that one was better David Moses (1:01:43) Correct. Jack BeVier (1:01:43) for you and that one was better for you. But not understanding David Moses (1:01:45) Yeah. Jack BeVier (1:01:46) the concepts to be able to make those decisions for your company at this point to me feels like negligence as for business owners. David Moses (1:01:53) I agree. It does. I think the regrets I have are are the only th regrets I have about any of my stack are the things I pay for. Yeah. Yeah. I mean, like l like it Jack BeVier (1:02:02) Okay. David Moses (1:02:03) really, it's it's true. Like it there's really almost nothing like besides the Microsoft ecosystem, there's really no software. I'm I'm happy with N A N. I'm glad I don't pay for it. I'm happy with Mongo, but it pisses me off that I pay for it 'cause I don't have to. I could've I could have just loaded it and yeah, you know, managed it myself. But yeah sorry, you were gonna say No, there's the it brings up a couple things that I think our younger listeners would really do well to heed. Okay. Number one, taste and judgment is going to be an ever increasing valuable skill, knowing what good is, knowing what the correct things are. So taste and judgment, human taste and judgment in outputs. The second thing is that the super who's of tomorrow are going to be the people who have a technical understanding. But also a deep operational understanding of the thing that you're doing. And so the people that are making the most gains. Now, the reason I'm able to create so much in my space is because I have 20 years of real estate experience. You know, I mean, we've all been doing this game, whether it be buying houses, renovating houses, the finance, the structuring, the management, all this stuff, right? So we have a deep operational expertise in our area on top of a technical expertise, right? And then three, Also understanding that the future skill sets are going to be a basic fun and you don't need to learn programming, right? Because the a the system's gonna be better at programming than you are, but understanding the fundamental concepts of it as a skill set as early as possible. And I think it's honestly gonna even for the creatives, like let's remove business from it. Quite honestly, like there's a whole big discussion out there about like what is art and what is real and What's human created and not created? And is it still art and you know all these philosophical discussions, which are a lot of fun to have? Sure. But the reality is is that like you still need to understand that they want you need a fundamental understanding of how it's built, why it's built, that type of thing. Right. And so yeah. I think those are all gonna be wildly important for people. And the younger they can learn them the better. And that goes into a whole education disruption topic, which I'll go down. Well, for sure. I I I and I think if we you know, kinda couple those things because I think all those are learnable and and doable things if people focus on them. And you have a growth mindset. Yeah. Right? Like you've got to have a growth mindset. Mindset is so important. And that's it kind of exactly segues into what I was gonna say is that like the parts that you can't teach, the parts that are really, you know, you have to really develop you really have to develop resilience. You do like like that is I think the biggest the biggest issue with young people. Like yes, there are so many young people who have, you know, the kind of like the built-in safety nets or like a giant university, you know, to incubate them or, you know, other just like they're just wildly ahead of everybody in that in that regard. But those kids haven't failed yet. Right. None of them have failed at anything. So they they're going to have to inevitably, if they're gonna achieve success, they're gonna have to have some failures and they're gonna have to bounce back from those things. And I think that's, you know, it's difficult for younger people. I kind of like, I'm very excited for that 18 to 25 year old, you know, kind of like that, you know, that kind of age range for what they're gonna do for the world. but that like 25, you know, to thirty five, you know, that that generation of kids Who are you know, there's there is a lack of resilience there. I mean, there's a reason why we see entrepreneurs starting businesses. I think the average age of an entrepreneur starting a business is like in the 40s. Yeah. So which is a just a wild statistic, but it's really it's the it's the super young and it's it's people with experience and a passion to solve a specific problem. Yep. And and they have they have that that thing where they just, yeah, I there's a there's a good chance I'm gonna fail at this, but I'm gonna do it anyway. Right. Because I'll figure it out and I'll learn from my failures and I will use those failures to get to success. And I I think there's a gap there. And I think that for people who are you know, young folks. Starting out, understanding this technology, absorbing it, implementing into their daily lives, making it native to them the way we had to do it with the internet. you know, is is something that it will pay dividends dividends for them later, but they need to add that mindset. They need to add that resilience into their into their skill set. I guess it is a skill. Yeah, I mean I it but but it really is just you know over overcoming adversity. It Yeah it's really difficult. Have you read so Angela it's so funny, I just I shared with you, I just came from parent teacher conferences this morning. I got a first and a fourth grader. And two of the things that we really stress in our household, and as you know, being a father and a dad and husband is really important to me, is the ability to have grit. Right. So Angela Duckworth wrote a book called Grit. It's kind of like a business one on one book. And it basically says that the people who are the most successful in life learn the characteristic and skill set of grit. Right. So persistence, sticking through things when they get hard. And I would agree that this last generation has been observed to be kind of babied and things have been done for them. And so quite honestly, there isn't a lot of stickiness or stick tuativeness, right? And persistence. And so that is a skill set we are working hard to instill in our children. so they're successful in life. The other is the internal locus control or a growth mindset, right? Like I can control my environment, I don't know it all. Failure is actually a good thing. And that's actually a framing thing that in America we've done a really poor thing of doing is that like we we perceive failure as bad. So we we actually shy away from failure when failure is just feedback, right? So you either win or you learn, and that every time I fail, I want to fail fast and I want to fail forward. And so I'm constantly testing things, I'm constantly learning things, I'm leaning into growth and I'm leaning into persistence and grit so that I can continue because ultimately we will win if we keep at it, right? And I think honestly, I wished every school in America taught that as a class because we would we would we would have a wildly different situation right now. Well, we wouldn't have I don't know, eighty-five percent of folks with a negative outlook about AI. I mean, and and this is very it's actually somewhat unique. to Western society. Like this is not this is not how it is in the Middle East. This is not how it is in Asia, you know, where where they have a a more positive view towards technology. so I I think that that's I yeah, I think I think that goes back to grid and and just we fear what we don't understand. But it's that's okay. I do too. But it's just leaning into that versus running away from it. That's the difference. Yep. Yeah. Jack BeVier (1:09:17) I think I think that what you guys are referencing there is also like fundamental to this like change management process that we're all going through, right? Like I feel like this is like very much a big capex cycle where we need to rebuild our businesses or we think that there is an advantage, a business advantage to rebuilding our businesses with this new technology, which is a fundamentally iterative intelligence machine, right? And so like To your point about 90% of the AI projects in corporate America are don't David Moses (1:09:48) Mm-hmm. Jack BeVier (1:09:49) go anywhere. They just fail, like they're just wasted tokens. Because it's real fun to go from zero to 80% over the weekend, right? Like you get an idea in your head and you vibe code David Moses (1:09:56) Yes. Jack BeVier (1:09:58) something and you get a bunch of dopamine and you and it's 80% David Moses (1:10:01) Yeah. Jack BeVier (1:10:02) good and you show it off on Monday morning, they're like, That's great. And you show it to your boss, and your boss is like, This is fantastic. Can you give me this, this, this, this, this, this, and this? And 80 David Moses (1:10:09) Yeah. Jack BeVier (1:10:10) to 90 is a little bit of a grind. David Moses (1:10:12) It is. Jack BeVier (1:10:12) And 90 is not good enough. And then you know, and then you have to iteratively grind it up to 98 before it is enterprise worthy, right? Before you can put it in front of David Moses (1:10:22) It is. Jack BeVier (1:10:23) a customer. And 90 to 98 just sucks. Like 90 to 98 is just hard. And especially especially David Moses (1:10:27) It does. It's very painful. Jack BeVier (1:10:30) if you don't start as a business person with this like software architecture framework in your head, like the best practices of building software, which is what you are doing when you're vibe coding, when you Don't understand the mental model that you're supposed to be operating in, that's an even more painful process, right? So like learning David Moses (1:10:48) Yeah. Jack BeVier (1:10:48) that helps accelerate the learning curve. But I think there's just a big difference between the leaders and practitioners of AI, the leaders who are trying to implement this technology and the practitioners that they're trying to get to use it, having the grit to just iterate, like keep at it and just it it'll get there. Just don't, just don't get distracted. David Moses (1:11:06) Keep at it. Yeah. Don't give up. Yeah. That's right. No, he Jack BeVier (1:11:11) Yeah, just don't give up on it. David Moses (1:11:12) nailed it, right? Like, cause there is a wild difference between a vibe coded piece of software and an institutional grade enterprise ready to roll out, even just internally without bugs. Cause what happens is so one of the big things is not only launching software, but also adoption. And so we found that if people lose trust in the software, the AI technology roll out, then there's no use and no adoption and all this waste of time, right? So that's been but getting through to making it durable, sticky, doesn't have bugs, that type of thing. And when you do have bugs, it's instant, you know, fix and and feedback loop and that type of thing. And when do you launch? Wild wildly, yeah. When do you launch? I mean, it's wildly important, right? The other interesting thing. So I was talking to any go ahead. Jack BeVier (1:11:52) Dude, Andrew, to to to to that point exactly. So like my everyone who have got vibe coding, like the FDEs who are, you I'm calling them FDEs who are vibe coding, business people who are vibe coding stuff, I part I I partnered each of them with David Moses (1:12:03) Jackson and Palantir over here. Jack BeVier (1:12:04) an integrator, and that's all they do. They all they do is make sure that what this thing guy built gets used by the managers and gets pushed down to staff. And when they don't like something, yeah, because if you have the vibe vibe coding it handed David Moses (1:12:11) Who so that's a great yeah. You need to designate somebody for that. You have to Jack BeVier (1:12:19) to the manager, the manager doesn't have time. frankly, to like dig into that. And they're just not going to do it. They're going to go back to doing it the way they want to because they want to go see their families. Right. And the vibe coder's so in love with his ideas that you know you know, he just David Moses (1:12:28) Yep, it's easy. Jack BeVier (1:12:33) doesn't want to hear that like what he built wasn't, you know, wasn't the mo you know, the Sistine Chapel. So like I David Moses (1:12:38) Yeah. Yep. Jack BeVier (1:12:40) I I've got literally am paying p that your only job is to be an integrator of the stuff we're building with the operation to make sure that this is not just wasted tokens. David Moses (1:12:49) Well what I was gonna say though is that Jack brings up a really good point is that for the next probably three to five years, there's gonna be a whole subset, there's gonna be a new role and it's kind of starting. But basically, people who are AI implementers or AI integrators as a service, I don't know if it's gonna be a company or as it feels like to me as could be more of a consultant role, but you're gonna have to get deep within an operations company like this, understand where all the data stores are and figure out ways to sync them together and then make them into usable integrations for organizations. I think people will make really good money doing that for people who aren't AI driven because I was just having a call with a EOS implementer the other day, and he was calling me for some advice because he's like, How deep are you in AI? I'm like, huh, how long do you have? And he said, in his latest, in his quarterlies, he's running into some organizations where it almost always is driven by the CEO as a leader, right? And and I agree with that. But also he goes, I'm actually running into some weird situations where one guy is like, all his favorite friends are now. virtual and all he wants to do is code and he doesn't want to be a CEO anymore. And it's you know what I mean? And so like I I tell people like AI is like basically crack cocaine for entrepreneurs because it gives you a superpower. Well in no time before in history has it been so easy for an entrepreneur who's a high I and has a whole bunch of ideas every single day to hit that dopamine button just on command to manifest all these ideas in the reality that he has. Right. Yeah. And so they literally want to do that all day rather than the hard work of everything else. And so it's kind of interesting what it's kind of spawned out in a lot of different ways. I I think look, EOS, I could you get you don't run EOS, right, Jack? Jack BeVier (1:14:36) I run I run EOS Lite. I I like yeah. I yeah. Yeah. David Moses (1:14:38) Okay. Yeah. You pull from it what you like. Yeah. So I I'm gonna but we use Stretti. I'm gonna make a call to Stready just to, you know, kinda explain to them, hey, if my agents can't interact with your software, I'm replacing your software. It's the simplest software that you can imagine and it's like completely unnecessary. but anyway, to just I digress, but the the reality is that EOS is a good system and AI can give it superpowers. And the idea that you have every L ten recorded, like the whole idea of EOS in a larger organization, an organization of more than five people, is you have multiple meetings that go on, but they're all feeding a common goal set of goals, right? A common common rocks, company rock, like everything kind of flows Jack BeVier (1:15:29) Mm. David Moses (1:15:31) up. to what the goal is. And AI does a great job of analyzing the individual meetings that are had to make sure that things flow up and down between those meetings. Things that are issues that really should be issues from, you know, lower level meetings make it up. And issues that one group is having, that another group is having get, you know, you know, get shared and elevated even if they seem rather innocuous in and of themselves. But when you get enough people having the same problem, even if it's a small problem, it could be a big problem. So, or at least a an opportunity to fix something. Yeah. But I think that that like EOS provides that framework and AI can make it you know, can really supercharge it. And so we're kind of exploring that a little bit now, making sure that there are just some simple flows that pull out of okay, pull out of this meeting, you know, What did they spend time on? What did they spend time talking about? Yep. And how aligned is it with the rocks of that particular L10 and the rocks that the company has? How aligned really are we? How fragmented are we? and and really give it a metric as it relates to the time being spent on those issues, right? And are they the even the right people to be solving those issues? Has another group already solved that issue? And it can be something that just needs to be communicated. So those things, you know, I've, you know, kind of been banging the drum since the beginning of AI is my vision is a full, is AI having a full 360 degree view of the company. That means every phone call, every text message, every email, every meeting is memorialized, distilled. Technology is not there to utilize all of that context, but it'll get there. And it it's just the constant process of taking large amounts of data, distilling them down into what's important as accurately as possible, and allowing future systems to be able to to be able to it work both ways, right? Extract from that meeting what wisdom can be imparted on the whole and impart into that meeting and those people the what what the whole ecosystem has has learned and also using it as a kick communication tool because we're all guilty of sending you know emails our companies, not just us personally, but of sending out emails that give the part of the truth that they know and understand, but not a full picture that they really need. And there's people that interact with our companies who have to go to this person and that person and that person to be able to eliminate all that. And basically when you're gonna have a meeting, a conversation, an email to somebody, and there's some pressing issue that they have to have addressed and you're calling them about some other issue that it's communicated. Again, this is it goes back to the socializer, right? Like that person cares what you care about. and so Before you hop on that call with so and so, have your agent, I tell this to my people all the time, have your agent review all the calls they've had with everybody. Yep, all the emails and text messages they've had with everybody at the organization recently. Jack BeVier (1:18:57) I can't think of it David Moses (1:18:59) What are they going to ask about? You know, what what's top of mind to them so that you're meeting them at a level, it's like, yeah, I need something from you, but the best way for me to get something from you is to give you something that I know you want and need. Mm-hmm. And that way you're gonna take it seriously because I care about your needs. I care about what's important to you. So I need you, you know, to then care what's important to me. And provided you have a company that has enough resources to do that, to actually, you know, intentionally, you know, gather that information and you're you know, you're you're well organized, which I can't say we are, then you'll get there. But I I also don't think the technology is there. I I really don't. I don't think the technology is there to contextualize Jack BeVier (1:19:40) Hey, let me let me David Moses (1:19:42) enough. Jack BeVier (1:19:43) let me let me are David Moses (1:19:43) No, please. Jack BeVier (1:19:44) let me make an argument. I think that this is a recent idea or recent idea that was introduced to me by one of our people. and so like I think that the technology's there. It's just that we have not yet, as business owners, spent enough time building out what Palantir got famous for calling their ontology layer. So like David Moses (1:20:05) Their ontology layer. Yep. Jack BeVier (1:20:07) P Palantir's whole business model is putting even pre-AI was putting forward deployed engineers in identifying what the systems are, bringing them all together. And they're really kind of like enterprise value. I the I this is a paraphrase, and someone correct correct me because they know better. I'm a real basic learner here, you know, intro to the this. But like the ontology layer is the thing that like Alex Carp David Moses (1:20:27) Like we have to think of like Jack BeVier (1:20:30) like pounds the table talking about. And it's the David Moses (1:20:32) Yes. Jack BeVier (1:20:33) mapping of the business ideas to the Object within the software architecture and the data fields and mapping what all of those business ideas actually mean in the context of what's been built here. And so, like, I think that the best models right now do a great job of taking a stab at it, but there are differences between there, there David Moses (1:20:36) And I think that almost right now does a great job of taking that out of the current. Jack BeVier (1:20:55) are still like there's misunderstandings still on behalf of the AI because. It goes in and and what the software architecture is, is its context, right? And then it like tries to fill in the gaps with emails and other soft context, phone calls, email stuff. But if we spend time actually documenting what we meant, like what the business intent was, what the relationships between each of these entities and feel and concepts are within our business, and we document that in the ontology layer, and then the agents. Have that as a starting point to interpret all the data in our systems and the inputs coming through phone calls and emails and outputs. I think that it gets there. I just don't think we've built our ontology layers yet. And I think that that might be the unlock for the beautiful, the the elegant machine that you're thj thinking of there, David. David Moses (1:21:38) It isn't there. I just don't think we've built our uncommon. I think that that might be a lot for some data. Yeah, it's it's but it starts with data, right? and and this is why it's important. Like we we've made it a point to say that any data that goes into any of our systems should go in by AI. I don't want you going in and even coding a receipt. I I don't want that, right? You're gonna tell your agent this is what it is. It's gonna grab all the context it could possibly need. And then it gets gonna give a a decent volume of data when it enters that transaction. And that becomes your ontology layer because there's context in every transaction. So there's there's redundancy. There is, okay, I want to look up every time this type of transaction has happened. I'm trying to code it from an accounting standpoint. I want to look up every time this type of thing has happened because this thing just happened. And I want to know how we dealt with it in the past and who did it and how happy was everybody with them. But as a, you know, we don't have the ability to have local models that get post-trained. Right. So the point where like you're you're asking a question and it's not going and grabbing a bunch of context. It just it's post-trained. It knows what your policies are, what your procedures are, it knows what your it knows it it knows how your company generally operates. It may need context in terms of specifics, but it doesn't need it doesn't need to know what the rules are for entering a transaction into QuickBooks. It just has that as part of its post training. We don't have that ability yet. and we don't have the ability to extract an unlimited amount of context before giving it to the agent to make a decision. So I can't like a larger organization like yours, Jack, it would be difficult to, you know, gather enough context to Do something without without breaching the limits of what that, you know, what those context limits are. So you really need to have everything so well categorized that the agent can distill it almost instantly down to the things it needs to know from the a voluminous amount of context that you've created to act on it. And I just Jack BeVier (1:23:57) I think that the David Moses (1:23:57) don't know. I don't know if the context windows are there. Jack BeVier (1:24:00) I think that the ontology layer though is the Cliff Notes that gets it so that it doesn't have to ingest this massive amount of context. It it it in it grabs the grabs the data that comes in, and because the ontology layer is documented that like s that, hey, when an email comes in, the email is right here. all of a sudden I understand that's a contact. That contact is is linked with that account, that account is linked with these three opportunities. So when it says David Moses (1:24:21) Yes. Jack BeVier (1:24:22) division street. It's like, it's this opportunity that it's talking about. And look, here's the records in that opportunity that show me all the context of the communications that we've had about Division Street. And all of a sudden we're not using very many tokens for it to have like really good context. But then it also David Moses (1:24:38) understanding what's going on. And y you've you've done that very well. I was just gonna say I've been sitting on my hands over he's Andrew has done this very, very well. So so I've been a fan of Alex and Sham, who is a CTO's work for a while. And if you guys read Shame's most recent books, I've been a fan of Palantir. So about six months ago, because the point was from the very beginning of this is I wanted to create the knowledge graph that would drive the connections in the system. So six months ago I went back with Jarvis and my system and I actually built out a ontology layer system within our entire company. I had to do all the research on Palantir. I every single I honed my X feed to pull up posts and papers. I uploaded them all. And then I basically had to go through my entire knowledge graph or my information base and then build the ontology layers with the neural connections, kind of like a brain. And so now, so yes, it does work. Jack BeVier (1:25:33) Would Obsidian? W would you use Obsidian or Yeah, okay. David Moses (1:25:37) obsidian, yeah. So I have 80 So it's not a vector database. It's no, this isn't markdown files. MD files. These are MD files Jack BeVier (1:25:44) Mark them files. Yeah, yeah. David Moses (1:25:45) that then the frontier model has then organized and built hyperlinking between the individual files and names, right? Names, relationships, properties, owners. a lot of my acquisitions has been, you know, finally built into a custom CRM. And basically now the ontology layer and the neural connections are built into every single conversation, right? For the map. So, yes, it does work. It is a wildly helpful piece of it of technology to implement in your system. And the frontier models are there today to do a really good job. I would recommend building weekly or monthly health checks on it to make sure that. Dead links are screened, that you are relinking things because links break, code gets jumbled and stuff like that, and run regular health and diagnostics on it. But it has been wildly helpful for contextual knowledge and just you guys danced around it earlier, but if you're a business owner and you aren't recording your meetings. through your system, like you definitely need to be, you know, same thing for calls. So that's that's another reason why we like Microsoft. Yeah, that's as fundamental as putting your shoes on in the morning. Yeah. Well it's all about shoes. At the end of the day, and this is why the the frontier models are paying so much inform so much money for specialized information. And why even in our industries we're seeing all the bigs merge because they realize how valuable the information they have, right? What do you use for recording by the way? so so for internal meetings we use Teams. So Microsoft Teams has a permissioning with their Microsoft Teams Plus account where you can set the auto-transcribe and auto-record every meeting. So every recurring meeting is set in by whoever the organizer is, that's recorded to their OneDrive and then it's auto-transcribed to a VTT file. Okay. for external meetings, I'll use if I'm in person, I'm using Plod. I've played with granola and fire fireflies is kind of the main one. If someone else is doing it, Although Plaud's really coming back with Jack BeVier (1:27:39) Matter. David Moses (1:27:40) the physical device. and then obviously Zoom and the others have it's surprising to me that Fireflies doesn't have a a physical device the way Plod does. And it's surprising to me that Plod doesn't have I know, like the the digital application. Especially surprising with Plot, because I mean how difficult is it to build something that's just the the genius of Fireflies is simply that it reads your calendar and it knows every meeting, regardless of whether it's a teams meeting, a meeting, a go whatever, go meet, whatever it is. Yep. I just Wild. And then for VoIP as well, right? Because we do a lot of phone calls, like all of us here. We're a dial pad company. Yeah. We used to be ring central. We like the tech of dial pad better. So about two years ago we made the switch to dial pad and we use dial pad. Yeah. And that's just for our call management. So yeah. Jack BeVier (1:28:19) What do you so what are you how exactly are you building the those markdown files? Are you writing? Like are you writing? Are you typing? Or David Moses (1:28:26) No, no, I don't write anything. I I only use I only use Obsidian to launch basically my terminals into the proper vault, just because it's a quick shortcut for me. I very rarely go in Obsidian. It's funny though, but my vault has gotten so big that it if there's actually a feature in the Obsidian app where you can actually click and see kind of like an ontology visual connection of all your files as little dots with connections between them. And it's been wild Jack BeVier (1:28:52) Yeah, it looks like a brain. David Moses (1:28:53) to Huh, it's like a brain. Yeah, it's like a Jack BeVier (1:28:53) Yeah, it looks cool. Yeah. David Moses (1:28:54) brain. It's really cool. It takes it did it takes like twenty minutes for mine to load though because there's so many connections between all the files. And so it's it's been fascinating, but super helpful. So these MD files, they are Kind of like it'll have, you know, policy and procedures, best practices, you know, examples of the right way something is supposed to be done, whatever it is. Is that is that generally what the Yeah. So Markdown is one of the simplest file formats, both size wise and complexity wise. It can hyperlink in and between files. So they're structured and I've really let the system design the best structure, but everything from CRM profiles, people's names, contextual history, frameworks from books, culture data. It's got like hard data. It's all it's all got hard data, right? And so and it it hyperlinks together. And then my brain or my memory is is within this data vault of all markdown files currently. Now I do back it up. So I run if you guys want to get into the the backup side of things, because the biggest so you want a hundred percent uptime. So for Sunny and our team that runs through Azure and everything's in the cloud and everything's backed up through Microsoft Azure for our cloud services and powering through the API. Everything, there's also a privacy port where you don't want to run things through model or put things in the cloud. So not only do we have the local server, I have a local external drive backup every night, I have a third layer of a third party off site, and we use Backblaze for redundancy. And then also because if power ever goes down in our office and I want access to that data vault locally rather than the cloud, I then just recently, a couple of weeks ago, put a secondary server back up running at home in my basement, connected to my system so that I can log into that intranet and then be able to have it go up and run as a redundancy server. Because your house won't ever lose power. Well, my house has a generator too and and and on top of on top of the business 'cause the business location and stuff. So same set I mean it's like s the same. I've got a I've got the server at my house. Yeah. It's got external drive connected. Yeah. Backs up every night. Battery backup. Yeah. Backblaze I've got running. It's like six dollars a terabyte. Yeah. It's like super cheap. Super cheap. And it's like winds up being Actually, no, I'm sorry. It's unlimited. I use the unlimited and it's per device. It's like you know, just because I've got a lot. but that is running that is running probably every twelve hours, eight to twelve hours. Yep. and that's yeah. But yeah, the same s same kind of you want redundancy, right? So yeah, is is you want you want a hundred percent uptime. Of your full capabilities within your system. Right. And so, and that's really traversing user interfaces, traversing because so and what happens is every layer, continual layer gets built because something breaks, right? And this goes back to the thousand iterations. we're on a trip over spring break in Mexico, and something happened to my system and it shut down and I had issues, and then I needed, I was like, okay, don't want this ever to happen again. And then You add a layer, then you add a layer, you know what I mean? And so and then you try to follow best practices to keep hundred percent uptime, hundred percent accessibility for you and your team, but also be able to manage control and access in and of the databases and stuff. So Yeah, but the databases that we ha Yeah. Go ahead. Jack BeVier (1:32:22) Hey, I have a stupid I have a st I have a stupid fundamental question about yeah, you building your obsidian set of markdown or your set of mark your your array of markdown files David Moses (1:32:31) Bolt. Yeah, the ball. Jack BeVier (1:32:34) that are linked together, your ontology layer. Are you feeding everything? Or like how how exactly like bore me with the details. Like you open up obsidian and what? Like you you like everything like it's David Moses (1:32:44) Yeah, good question. So Obsidian, think of Obsidian just as like a hard drive. And Jack BeVier (1:32:49) Okay. David Moses (1:32:50) right, it's a hard drive. And so every time that I open up for my personal Jarvis, he is opened within that file directory, which is my operating system that holds the vault of every markdown file. So you have the power of the frontier models, you have the persistent memory of a local data vault. Jack BeVier (1:33:11) Got it? David Moses (1:33:11) in everything is d you know, written in markdown files and it knows to write everything to it, to connect it, to use it as its, you know, local brain for research and and pulling everything together. So every phone call? Yeah. Jack BeVier (1:33:23) so as as it as as it ingests as you point it to a source of data, it ingests that data and turns it into a markdown file that then becomes persistent memory. Okay. Okay. David Moses (1:33:29) Yes. And and it'll write markdown files. Yep. Yes. Wait, so is that like like phone call transcripts, text messages, emails, the meeting notes, like everything goes into Obsidian? So yes. but it has to process them. And so you have to put up automations to be able to process, transcribe, and then document and pull out the usable information from it. Okay, so it's not it's not it's not it's not saving just like like raw no no no it's extracting from the doing the processing yeah that's why I said I've I've used so many billions of tokens it's not even funny but and that's why that's why Jack BeVier (1:34:08) Yeah. David Moses (1:34:09) flat accounts make sense. But what I was gonna say though is that also, but it does not limit you from still going out to your third party API. So like for example, once you go to developer access on AP on for the developer API with App Folio, there's I think there's like 13,000 endpoints or something. I we were just messing with it yesterday, trying to solve this custom field question. And you can't just extract everything and save it. So what you have to do is design the endpoints or the reporting that you need. It does a nightly pull from Atfolio every night to then generate the reports that it then delivers to my team every day. So every day, my team, my leadership team get specific reports from Atfolio Lifetime. my staff member, my team members get a specific report on their KPIs that's important to them. so that everything is live time updated. So they don't even have to go and SL. Just Sunny sends it to them at nine a.m. when they get in the office and it's there. Yeah. I so I mirror Appfolio to Mongo. Okay. Every transaction that happens that is accessible through the API in through the the database API, not the reports API. that's accessible is ingested and mirrored into Mongo. Into Mongo. But it's not a So like we we deal with transactions, conversations, and so far it's just dial pad, our our voice conversations and our meetings. we we need to add email, but it's just it's a daunting task. So do you guys use Microsoft or is it a whole company? All all Microsoft. so we pull, you know, the ev every phone call gets ingested into Mongo and you have every phone call gets processed. Yep. And when it processes, it basically takes the AI transcript that Dialpad does, the full transcript, and then it essentially looks through all of our systems to relate everything that we need to know about this call. So this person may be calling. Well, they're part of this, you know, they're a tenant that lives at this specific location. they have other co-tenants. they may be calling about a specific work order, they may be calling about something, a specific payment they may be called, whatever it is, right? So it tries to does a pretty good job of kind of enhancing that call so we know everything else. We know all the context that went into that call. And then it creates, you know, kind of the output for us is it creates a call recap. And that recap is, you know, this is what we talked about. Here's the promises I made to you. Next step. Here's the promises you made to me. and allows people to really understand it. you can have external recaps, internal recaps. You know, they click a button, it goes into a folio, they click another button, it goes great, you know, it sends them a a recap. So either look, th those things are I think key, but it's just that's the data that I was talking about when I'm saying like like get a real full understanding of what's going on. but then it's getting people to use it and understand how like that it's even there. Like they're asking questions and it's not magic that, you know, these that they say, Hey, what's im I'm meeting with this board member. We have a community management company as well. It's like, hey, I meeting with this board member, what's important, what's going on? It'll look through the phone calls and emails. that you know, with that with that board member, with that board, all the work orders. It looks through everything and says, Well, you know, here's what's going on right now. and and the idea is to deliver to people, you know, to to essentially Mm make us look perfect, right? Make us look like so there's no reason you shouldn't be prepared anymore. Right. No. You Jack BeVier (1:37:43) Mm-hmm. David Moses (1:37:44) ha you should be prepared for for for every every meeting. And it's just a it's just a matter of when that meeting's gonna be. Right. It's like, you know, if that meeting's gonna be next week, then you know, set an agenda and make sure that you ingest enough context to understand and make that a meaningful meeting. but yeah, we we waste a lot of time in meetings. You know, I think we do we do. We waste a lot of time in in meetings. But if you if you can figure out a way to have the meetings because they're touch points, there's a they're they're they're still like they're necessary from even if you're sitting there talking about nothing. Yeah. It it's an important touch point to the people who have those meetings, right? They they they're making connections, they're talking to other humans. So if you can take that opportunity to make it also productive and make it as impactful for the other f for for the folks in that meeting as as it possibly can be. You know, they're coming out of that meeting, you know, rating it at ten and and being very excited. They're coming off that phone call with that customer, really feeling like they impress the out of that person. Yep. and and that's that's really what it's like that's what these tools are for. Right? They're not to replace you. They're to supercharge you. Yes. Give you superpowers. So that that is but yeah, the ontology layer is something I'm going like literally when we're done. I'm gonna go r you know and say, Hey, you know, how can I either use Mongo to, you know, build this obsidian vault that, you know or just ask what's the best way to g to create an ontology layer, right? Exactly or can I just use Mongo and create an ontology layer? Figure out the best way. Again, I try to tell people. AI may be the question, but AI is also the answer. Yes. So like you gotta get comfortable asking and trusting, you know, that it's gonna come up with the right answer. Yeah. And I I look, I think that it will it's also it so like we talk about Big Brother. We talk about like, hey, you know, that thing that's the the biggest fear, right? People don't want all their calls recorded and whatever. Because it they their their fear is that they're gonna say the wrong thing, they're gonna be exposed for, you know, something that they fear that there is a shortcoming of theirs, whatever it is. But I think that it can be big brother. It can point out every mistake everybody makes, but it can also point out the wins. And it can also understand the importance of the balance. Right. So when somebody is doing 90% of the things correctly and they're doing 10% of the things incorrectly, you have a general mid level manager persona. in America, in every small to mid sized business that doesn't get it. There they they think that their job is to point out where everybody's up. And that's not their job. Like their job is it their job is to call balls and strikes. Their job is to point out the wins and they're if somebody's doing ninety percent right and ten percent they're they're falling short, you want to your feedback should be ninety percent. Yep. And they gotta catch all they gotta catch the winds, right? Like you gotta call out the winds as well. Right. So so that and and I think that, you know, that mid level management is dead for that reason because it and it's not it's it's it's not because they're going the necessity of that human is going to be obsolete. It is the it it is what that position has become that will become obsolete. Yep. because it's just not a it's just not a great way to run a company. Jack BeVier (1:41:25) All right, gentlemen. This has been a super interesting conversation. Man, Andrew is a real absolute pleasure to to catch up with you, man. It's really impressive what you built too. So got David Moses (1:41:35) Likewise. Jack BeVier (1:41:35) me reinvigorated once again to dig back in. So looking forward to it. David Moses (1:41:39) Love it. Love it. Again, it's just it's stay flexible and understand that you know, change is the new normal and just embrace it because nothing is gonna stay stagnant any longer, you know. So life moves pretty Jack BeVier (1:41:52) Yeah, that's true. David Moses (1:41:53) fast. Life moves pretty fast. Enjoy the ride. Hang on and enjoy the ride. So awesome guys, but thank you. This is awesome. thank you. Thank you. Jack BeVier (1:41:59) All right, everyone. Hope that's another episode of Fork.ai. Hope everyone got a lot of productive use out of that time. And we'll see you next week. Have a great one.