4kd.ai

Craig (00:00)
Hey, welcome back. It's Forked AI and we have a full cast of characters today, Mr. Bavier. Dave is going to be joining. look, there he comes right now. Exciting. Good to see you, Mr. Moses.

David Moses (00:10)
I'm very sorry about the delay, gentlemen.

Craig (00:12)
Sorry, you just missed my amazing opening. Jack, good to see you. Jack, why don't you go ahead and introduce our guest today and we'll just jump into a lively conversation. I had a wild night until 2am last night with Claude Cote and Codex. So can't wait to discuss that.

Jack BeVier (00:27)
Nice. Yeah, it is my pleasure to have Kris Guerin of Riparian Capital Partners on the podcast. Kris is someone who we have talked to on the Real Estate Investor, Real Investor Radio podcast previously, as he is a very prolific renovator, community developer, residential investment expert.

and ⁓ is active in Baltimore as well as a number of other cities and has a much very, very interesting platform approach to building a rental portfolio. And I have had the pleasure of knowing Kris for a number of years now and working with him in, as we're both very active in Baltimore. And just another thing that we happened to also share is that Kris and his team have always been very tech forward in their approach to the business. And so as AI kind of hit the scene,

⁓ Not surprisingly, those guys have been on the front edge of that curve the entire time. So a secret, a secret, you know, desire of mine this for even, you know, getting involved with this podcast was David to get you and Kris talking together. And I just want to be a fly on the wall and watch you two just like go at it. And because I think you guys are two of the most innovative people that I've met in terms of the real estate investment space as it relates to to.

David Moses (01:34)
No pressure.

Jack BeVier (01:39)
AI tools. So Kris, thank you so much for joining us this morning. We've been looking forward to the conversation and thanks for your time.

David Moses (01:46)
It's nice to meet you, Kris.

Craig (01:47)
So guys, last, I'll just, I'll start us off and maybe spark some conversation here. Yesterday was a real unlock. Dave, I really just wanted to call you to go like, man, something happened today that was just wild. you know, I'm, McClellan, our AI God here at Dominion developed this kind of starter kit, if you will, that basically

For lack of a better word, Jack, I really look at it like as like it's almost like cursor in that it's just got a really beautiful interface. the starter kit basically gives you all the connectors that you need. The Salesforce API, Alloc API, Ring Central API, all the enterprise tools that we use here. And then it allows you to build on top of that.

And so yesterday I was playing around with it and well not play, was working and I decided to basically build like a phone burner app for LOs so that we could basically select a list of leads, press a button and that starts a phone burner session where you can quick dial people and it has, and so I talked to it for about an hour, Dave. I was just sitting there kind of explaining the problem, explaining what I wanted to do.

And I've done that before, but I think as you get better at prompting, you speak in a much more sort of structured way. And for the first time, Dave, I was able to walk away from the computer for an hour, an hour and a half. it, well, basically I told it right before I set it free. said, build, build me a PRD with very, very specific tasks that can be implemented and then verified and then move on to the next. But I want a team of agents.

that are doing this in parallel and in sequential. And it wrote me a PRD that was probably, I think it was 830 lines and so, and like nine phases. And so it one-shotted a really incredible sort of mock-up GUI because I told it, was like, just make it so that all of the data works and so that we structure it properly so it can fit into a bigger

gooey that I've already built. And that's kind of like phase nine. So it was it was really incredible what happened, man, that it just it just one shotted foam burner for me, essentially connected to Salesforce connected to ring central connected to outlook dispositions. And it it works beautifully. So really, really exciting. Then at about two in the morning last night, I was telling Jack I

Jack BeVier (04:02)
connected to Salesforce, connected to Ray Central, connected to...

Craig (04:13)
I decided to do a battle royale with Codex and Codex five five and Claude. And I said, take a look at the PRD, take a look at the entire project that I'm building and, the PRD, not just for the phone burner part, but for the bigger project. And I said, let me know what you think Codex of this. And it gave me some things to think about. And one of the things that we talked about last week on the podcast, Dave, I don't know if you remember it up, but Jack was talking about the product that he was building and

Jack BeVier (04:21)
Thank

Craig (04:40)
it sort of fell over on itself because you keep bolting on things like more functionality and more functionality. And so that's exactly what Codex told me I was doing. It was like, you're doing a lot, but like you're doing like you're putting it all in this starter kit rather than building modularity like tools that you can just sort of plug into the starter kit. So that's, that was a real unlock and I'll probably be working on making it more modular today. So crazy day yesterday. Get your guys thoughts on that.

Jack BeVier (04:51)
Like you're doing, like you're putting it all in.

I'm today.

Kris Garin (05:07)
⁓ You're reminding me of the phrasing, God object, which is what happens when you build a gigantic piece of code that's supposed to do everything. And what I've found is as I'm trying to come up with stuff, I do that and the tools and I usually am working in cloud code.

Craig (05:19)
Yeah.

Kris Garin (05:28)
It just, that's the sort of force of gravity. And, you know, I just, every so often you kind of have to stop and just go through a refactoring exercise and say, break this into four or five different things. Then you spend some time talking about how to do that. And then just does that and it's pretty quick. And then you kind of keep going. But,

Jack BeVier (05:35)
way in refactoring exercise is they create incentive for countering things and you spend some time on that and just do some cool work and take it back.

Craig (05:44)
Yeah, it's pretty quick.

Kris Garin (05:46)
I think that's exactly been my experience. You build these things and they just get big and top heavy. You have this great idea and it's amazing how easy it is to play. I also think there's no other way to learn that than to do it and the cost of that lesson.

Jack BeVier (05:52)
Thanks

Kris Garin (06:05)
you know, now versus what that education would have taken, you know, months ago, certainly years ago. You know, it's just like, that's the thing. It's like those mistakes are cheaper than they've ever been, you know, but there's still the mistakes that I think we have to go through in order to, to build something that, you know, has any, any, any real durability to it.

Craig (06:24)
Yeah, Kris, I want to jump into like how you started and, know, like sort of the evolution and like where you are today. Yeah, I mean, we can jump off on that if you like. I'd to hear about like, you know, everybody started working with chat GPT trying to figure out how to write with it. But when did you really start taking taking advantage of the tools?

Jack BeVier (06:24)
Hey Kris.

Kris Garin (06:43)
Yeah.

Jack BeVier (06:43)
Also,

would you mind just giving for those who didn't hear the Real Investor Radio episode, just give folks a background on your real estate and property management business so they understand the scope of what you're trying to tackle from an ops perspective.

Kris Garin (06:56)
Sure, sure. So, Repairians started in 2019. That was kind of act three for me professionally. First decade or so of my career was in journalism. I actually worked in newspapers, magazines. I've read a few books about business. Interviewed one too many people who was having more fun than I was in business. Got myself to grad school unexpectedly. Fell in love with real estate, which I've kind of come to think about as like...

three dimensional checkers, you know, like there's no one part of it that's all that complex, but somehow when it all comes together, it's just completely insane. And I spent most of the next decade in investment banking, raising equity for public and private real estate operators and really a lot of emerging.

managers, emerging asset classes, more kind of unusual structures, and got to see a lot of both the emergence of workforce housing as an institutional asset class in apartments and the emergence of single family rental as an institutional asset class full stop. And so I kind of got more and more convinced that the workforce...

segment of scattered site, the naturally occurring affordable housing, you know, was really, it was aging, but critical segment of our housing infrastructure is under invested a lot of aging ownership that needed a succession plan and a transition and

that would be an interesting place to focus. And so we started repairing in 2019, got into Baltimore in 2020 with a portfolio acquisition, thought we were going to do kind of third party property management with hands-on asset management. We ended up taking over property management fairly early. And today we've got a...

about 800 units in the city of Baltimore, overwhelmingly row houses. We manage in our fund and own about 2200 units total. We're also in Pittsburgh and Detroit, Cleveland, few other places in Ohio mostly. a lot of what we've done has been through...

some interesting transactions where owners who've got, know, they've owned these assets for a long time. They're looking for an exit. They've got a tax issue where they're able to actually come into our fund by contributing the assets and taking back equity. And so that's been an interesting structure too. And so.

you know, operationally intensive, never expected to be a property manager. You know, we've actually, you know, gone back and forth on third party in-house and we've done it, you know, different ways in different markets. And I think there's probably more than one right answer to that question. But the whole thing about scattered site is everybody struggles with it because there's just so much information, right? And there's so many small decisions and so many handoffs. And so, you know, beating our head against

that wall from the beginning. It's easy to say we're going to go in and kind of like institutionalize this. You you buy something from the mom and pop owner and you realize like, you know, you're worse at it than they were. And you're like, wait a minute, you can't make that up in volume. You you got to, you got to actually like figure out, you know, things that work. And so that's kind of where we were in that's our business. And then kind of segueing to the, to the tech and the AI stuff.

That's kind of where we were in, you know, I'd say, you know, kind of like, you know, early 2023 when this kid from Stanford was like looking for a property management sandbox. He wanted to try some stuff with utilities and sort of like one degree of separation. And he came in and he built something that I think today you could probably figure out, you know, you know, with Claude code and, you know,

bunch of time and it wouldn't be as good as what he built. at that time, it felt like magic. And utilities, as I think most of your listeners know, in our space, particularly in jurisdictions where the water utilities, municipal or quasi-municipal, are always a real pain point. And so he built something for us that...

went into the utility portal, pulled the bills, kind of integrated with our tenant ledgers, handled the payments, all that stuff. And it wasn't perfect, but it worked really, really well. And it was something that we really weren't able to manage effectively, almost at any cost in-house, because it was just a lot of moving parts and it was hard to get people to focus on it when all the other kind of, you know.

bullets or wizened over your head, right? Trying to get through the day and property management. then, and so this was kind of when like ChatGTP was like amazing, but it was kind of like, hey, you know, read me the Bill of Rights in the style of Jay Z, you know? And like, that was like a kind of amazing party trick, you know? And like, that was amazing party trick, but like, you know, and we saw, you know, that these tools were already replacing like actual work.

and that that wasn't like a future state kind of thing. And so made a decision in early 2024 that like whatever we did, we were going to try to be organized about it and started a journey of developing some of this stuff in-house. have, we've definitely done some interesting things. We've learned a bunch along the way. And I think our...

settling into a middle path on how much infrastructure, how much ad hoc point solution, all that sort of stuff. But that's how we got to today. And I'd say a lot of if the AI is calculus,

Right? know, a lot of, ⁓ I'd say like the vast majority even, you know, of the, of the, of, of the journeys has been algebra, you know, and it's just been, just been the basics of organizing information and, and, and, and particularly, and this has sort of become my obsession, perhaps, you know, beyond the point where it has like a lot of business utility, but you know, the, the, the stuff that never gets written down.

And how do you start to kind of find ways to fill in the gaps there where in a traditional way running an organization, you would never document these things because there's no value to it. You get somebody 85 % up the curve, them in a bullpen, and then the rest kind of gets solved through osmosis and common sense. But you can't leverage AI that way. And so thinking about, you know,

how much of the challenge and the opportunity here is technical. Sure, like obviously, but like if you just froze the technical development, even six months ago, like you could do stuff that looks like science fiction if you could perfectly organize your data, but nobody can perfectly organize their data.

Jack BeVier (13:13)
I challenge of being an opportunity here is technically difficult. Sure, obviously, if you just follow the most technical rule, and you did six months ago, and you did seven months ago, it's fast fiction. And you can prove a paper as you did it. And nobody can prove a

paper as you did it. And so I think the journey for

Kris Garin (13:33)
And so I think the journey for us is

like, we keep getting painfully pulled short and brought back to questions of like context and schema and like fundamental stuff. And we do that because we build these cathedrals on foundations of sand. And then we find their limitations and then we kind of come back. And so we've been able to put in place some cool stuff, which I'm happy to talk about, but I'll

Jack BeVier (13:56)
Thank you.

Kris Garin (14:00)
I'll stop there.

Jack BeVier (14:01)
So what's the team look like from an operations perspective, like from a property management and real estate perspective? And then what's the team look like from the tech side? what's your balance?

Kris Garin (14:11)
Yeah.

So we've really got two people running our kind of in-house tech effort. And then they'll leverage some outside engineering when needed, although less and less of that.

except for like really, really high value stuff. So, you know, on the property management front, you know, we've got property manager, you know, running, running the market, you know, you know, APMs, we have somebody, you know, overseeing compliance. We have somebody who, you know, manages sort of the, you know, physical, you know, property services, repair, maintenance, turns, stuff like that. You know, so that team, you know, in Baltimore is, you know,

plus or minus 15, 20 people, right? It kind of adds and flows a little bit. We've actually gotten out of property management in Pittsburgh and in Detroit. We still do it in Cleveland because we found local operators we felt were actually just.

more ensconced with some of the local relationships, particularly compliance and inspections and things like that. There's just a real hometown dynamic there that is tough to overcome. I think the idea has always been, and we've explored it with greater or lesser success at various points, but the idea has always been we want to really embed

the sort of product perspective in the operations as deeply as possible and create the conversations that never organically happen. Because I think what's different is not that the engineering genius is gonna be more interested in really understanding your business.

Jack BeVier (15:48)
Mm-hmm

Kris Garin (15:54)
Like

that's never really gonna change. It's not gonna be where that person's interest is. It's also not gonna be the best use of those really expensive hours, right? What's changed is how much closer the business people can get to understanding what they need to provide.

in order to make good engineering solutions possible. And so trying to figure out how do we embed this product literacy in the operating team and then the operations literacy with whoever is anchoring that product effort. That's...

That's sort of the magic. And again, it comes down to this, it's like the context, it's the test of knowledge. mean, if you sit a bunch of property management employees around a table and you bring in a couple of engineers to do a product discovery session, there's going to be a lot of like, it sounds like you're doing this, right? And then people are going to just nod their head because they're going to be uncomfortable and they're going to want to get out of that room as quickly as possible.

Jack BeVier (16:48)
Ha ha ha ha.

Kris Garin (16:49)
And then like, you know, you're going to get something that's technically correct, but like mediocre, right? Because they're building what they were asked to build. so it's like, how do you create that conversation? Like that's the hardest, hardest, hardest part. And that's what's so interesting about, you know, stuff like, you know, what Craig was talking about, right? Is you can, like, he knows exactly what he wants.

Jack BeVier (16:54)
Yeah.

Kris Garin (17:07)
And so the time he would have spent explaining it to an engineer who then goes and he could just spend getting it wrong and getting it wrong and getting it wrong until he gets it right. And it's like, when he gets it right, it's going to be amazing. And it's actually not more time than it would have taken in the old way.

Jack BeVier (17:08)
Yep.

Kris Garin (17:23)
Right. And, you know, so I think it's, I think it's all it's people stuff. Like it's all more than ever, you know, which is a little ironic, but you know, I think it's, and I think the gap between groups that are literate enough in the tech and, but also like thoughtful and sensitive enough about the people, you know, and, and, and then the ability to articulate like the real operating need. think that's like where the magic is going to happen.

David Moses (17:47)
I, sorry, go ahead. I mean, I definitely completely understand and agree with what you're saying. What I kind of connected really well with is when you have an engineer sitting with a property manager. And I think that one thing we've talked a lot about on this podcast is just, you know, the engineer has been the technical expert.

Jack BeVier (17:47)
So just go ahead Dave.

David Moses (18:07)
who essentially is a glorified translator, right? They're taking, you know, this is what the property manager does, this is what their pain points are, this is what they're trying to solve, and they just translate that to code and something happens. But what, you you miss a lot there because, first of all, you miss all the opportunities of what, how could this be done differently? Not just differently, but better. And I think you also...

You know, like you said, just... There's so much lost in that translation that if you just had that person, that property manager, able to just, you know, screw up, create something, waste some time, then I think they would be able to fix it and the end product would be significantly better. But that whole system hasn't existed yet. Like, it exists now.

But most of the property managers, most of people actually on the ground doing the thing, they don't understand that these tools even exist. And the ones that do, think are, I think I'm very excited over the next six months to a year at how that's gonna evolve. Because I think that what Craig and Jack built with this toolkit, it's really just like, that's really just handing over.

the keys to the car and saying, you know, figure out how to drive, you know, like here's, I'm gonna show you where the shifter is. I'm gonna show you where the steering wheel is. I'm gonna generally tell you the rules at a very high level, but just go ahead and drive and figure, see if you can get where you're going. See if you can get where you want.

Craig (19:35)
Skip the learner's permit, go straight to driving, yes.

Kris Garin (19:39)
So I think that's right. I have also been humbled and not in like a humble to accept this award like sense, by the realization that like there are a lot of people who like just do not give a about any of this and like that's actually fine.

Jack BeVier (19:39)
You know, you're my favorite.

Kris Garin (19:58)
You know, and so I think yes, but also I think there's still something, a special ingredient. There's something about, there's curiosity. Like you have to be curious and like a little perverse and ambitious in a certain kind of way, like intellectually ambitious, you know, in a way that anybody at any level could be.

But many people, if not most people, are not, which doesn't make them bad at their jobs and doesn't make them. And I think it has taken me some time to realize that just because I think this is the most exciting thing in the world doesn't mean I can infect everybody around me with that enthusiasm. And so I think, yeah, 100%, making that accessible, but also figuring out who becomes

you know, who on who of the five people doing the same job is excited by that? And how do you empower them, you know, rather than trying to turn all five of them into something that like delicate because, you know, I've tried it a few different ways and it's that that you know, that curiosity is like, you know, that that's actually like the one ingredient that like you can't do any of this without like if you don't actually find it really interesting.

you know, then you don't power through all the mistakes and all the, all the everything. And that's, you know, it's like something that, you know, that generally people aren't hired for or screened for, you know, in these, in, in, in, in these roles. And by the way, don't necessarily need to do because using the tool once it's built is fine. And that's good enough, right? Like, you know, not everybody has to be a builder and not everybody has to be like a forward deployed engineer. And, you know,

figuring out how to strike that balance or something I still haven't totally nailed. we've gotten it wrong in both directions is all I can tell you as we've been sort of. But that's because I'm one of the people who's like, I might give me the keys. No, no, I don't need to read the manual. I just want to grind on the gears. it's fine. But not everybody's happy to fail in public and do all the stuff that we do in order to learn that stuff.

there is always going to be a barrier there that's not technical, but it's again, it's like it's just like a different kind of person, you know, is going to be empowered by like this constellation of opportunities, then is the kind of person who would have been empowered by the dominant constellation five, 10 years ago, and figuring exactly who that is and how they rise and how you identify them and how you support them is going to be, it's going to be really interesting.

Jack BeVier (22:28)
So my favorite experiment that I have going on right now, there's this like, know, Americans are afraid of losing their jobs to AI when in fact, I think they're going to lose their jobs to other Americans who are using AI. And a lot of the commentary has been that, well, the AI is not replacing, you know, higher skilled work. It's helping create tools.

Craig (22:32)
Huh.

Jack BeVier (22:52)
And if that's a threat to any humans, it's probably a threat to the VAs, right? Because the work that we have as small businesses outsourced to the Philippines and Colombia and Egypt are the easiest work to just have the manager teach ⁓ an AI how to do. So I was like, that's a fun use case. Or that sparked a devious idea. So one of the Filipina VAs

Craig (23:16)
Hmph. Hmph.

Jack BeVier (23:19)
that we work with is a recruiter. And she's been a recruiter for 10 years. So she comes in, she's an external recruiter. She uses LinkedIn recruiter. We use JazzHR to organize, basically like a CRM for recruiting. So she's got the LinkedIn recruiter account. We use Culture Index from a personality profiling perspective. And we've got a database of freaking like a million resumes, like an insane number of resumes over the past like

15 years. uh, and so she gets in, you know, a month and a half ago, she gets in, she starts using all the different systems and she's done a couple of placements for us. Like, and she's just paid hourly. She's staff and, um, works with our HR team and brings people in and, and she's a recruiter. Right. So I sat her down and I was just like, Hey, download Claude desktop. Here's the, you I invited you to the account. I was like, you know how to do your job better than anybody else. You were the best recruiter.

Right. Like you've been doing this for 10 years. Here's how you use it. And she saw me. She I demoed that like Baltimore Public Records database thing that I put together and she was just like, my God, you're so amazing at AI. This is amazing. I was just like, no, no, no, no. Like I'm just a housing nerd that like so I knew what to tell it. And ⁓ I said, and you are a recruiter. You know exactly what you need so that you don't have to context switch between the four different systems that you have to log into.

David Moses (24:27)
you

Jack BeVier (24:38)
And so, hey, just build exactly what you want. And here's, you know, here's how you connect things and here's how APIs work. And I sat her down for 30 minutes. It couldn't have been more than 30 minutes, but I would be like downloaded Claude desktop and I invited her and she clicked the link and I said, Hey, here's where you start typing and told her about sessions and folders and that stuff. Just the basics, the real, real basics, right? Gave her a learner's permit. And then I just gave, and then I just got rid of the limit on her tokens. And I said, go and

She gives me a daily update of where her app's at and it's pretty damn, like so, you know, I think we're like four days into it and it's like getting damn good, like damn good. So if I want to freak out, you want to freak out the Americans, you're not, like the Americans who don't learn AI are not going to lose their job. The Americans who don't want to learn AI are going to lose their jobs to the Filipinas who learn AI. And that's the threat, right? Is like, if you don't want to pick up these tools,

There are people all across the world who are very interested in working and might be more motivated than you. And I don't know whether they're motivated by fear or greed. I don't really care. But Serena with Claude code is the best recruiter that Dominion has ever met. So I'm really excited about that.

Kris Garin (25:49)
Well, I think there's gonna be a lot, a lot of people who are two or three times more effective than people who aren't using AI. There's gonna be a smaller number of people who are like 10 times more effective in organizations. And then there's gonna be a handful that are like 100 times.

David Moses (25:50)
That's awesome.

Jack BeVier (26:07)
Mm-hmm.

Kris Garin (26:07)
Right.

And so those gaps are going to compound, particularly in the next three, five years, you know, before the new table stakes become clear. And then whatever that is, it'll be what it'll be. Right. Like I've got an opinion on interest rates too. It doesn't matter. Really like doesn't matter. But like, you know, that, you know, in between there's going to be these gaps.

You know, and that's, and, and, and, and I think what happens in the space between those various gaps is going to be super, super interesting in some areas, potentially seismic and other areas, just meaningful shifts. But I think to be, you know, I think, I think you want to, for sure be one of the two to three times. Like just to protect your ass. It'd be really nice to be one of the 10 times. And I don't think anybody gets to know where they were going to be one of the a hundred times until it's in the rear view mirror.

us to settle in and it's all over. like, don't think about that. You know, but that's, I think.

David Moses (27:00)
That's a,

I agree with that. ⁓ So I, so we talk a lot about like what's gonna happen and everyone's worried and you know, how do we as small businesses actually move, you know, move the needle forward? What do we do with our people? And I think that I've had so many different thoughts on this and ideas and I think what it comes down to is,

Craig (27:04)
Yeah Dave.

David Moses (27:25)
you know, growth comes from struggle. And I think that as we remove resources, people, tools, expensive things that we're paying for, and put people in a position where they have to struggle to do more with less, combined with the tools to do that, I think that that

human nature is to adapt and I think the people who will adapt eventually, I think you accelerate that process. I think that that's, I think I said it last time or the week before, I think people are scared and I think at some point that fear is either going to motivate somebody to do something or it is going to be some,

something that happens, they lose their job, they get put in a situation where they have to struggle, and only then will they start to move forward. And I agree that all of those things are okay. I think all of those people are gonna be okay. It's just a matter of putting them in a position to get there faster, and I think it's struggle. I think we put people in a position where they have to...

where they have to, know, they don't have this person to rely on anymore because that person's gone or that person has been moved to another part of the company. Or you remove a tool that was expensive and clunky and didn't integrate well with anything. And they were relying on that. And all of a sudden they have to struggle. And now it's like, oh, but by the way, we've got this tool here. We've got these tools here. And if you use them, you can recover from

the loss of this thing and I think putting people in a position where they have to adapt I just I I think you're going to Get people to that moment faster and and I and maybe you'll be able to save some people you know in the organization who Really needed to struggle before they got up to do anything and sometimes those people can be really really good at it

Jack BeVier (29:21)
Yeah, I hear that.

Kris Garin (29:22)
I think that's a perfect articulation of it. And then I think what it captures is like, this is a people moment. Like these are not, like you didn't name a single technical challenge in making that point, right? That's what limits most people.

David Moses (29:32)
No.

Jack BeVier (29:35)
Yeah, absolutely.

Craig (29:35)
So I met with.

I met with a young guy just out of college. We actually interviewed him here at Dominion and he went to work for a fairly significant self storage development company and basically spends his entire day just searching through scouring costar looking for land opportunities. And obviously they have their box of check marks that he's got to figure out. And so

We met, had a couple of drinks no more than a month and a half ago. And this kid had never touched AI. And, but he had this, he had that thing, Kris, he had that curiosity and he had that creative curiosity. And he was like, well, tell me what you're doing. And I just, I had an app that I built for the badge scanner, Jack, like went to a, went to a conference and they charged 450 or $600 for you to scan a badge with the, at a conference.

And that's per person. So I was like, what the hell with that? I'll just build an app that scans the badge and saves every saves all the leads. And it worked perfectly. So I was showing him this and a couple other things I built. And he would he was like, I got to go home. I'm going home right now. And he calls me the next day. He's like, I'm up until two in the morning. Fast forward a month and a half later, this kid is literally working next to the CEO now. CEO can't believe what he's built. And he came by the office yesterday, Jack, and showed me what he's built. And it is mind blowing with this kid built.

it is mind blowing what he's built in a month and a half. So ⁓ it's that creative curiosity. It's that just unleashing people on great tools and kid has never, never, never wrote a line of code in his life. And yeah, that's exactly what.

Jack BeVier (31:08)
Kris, for everyone's benefit, what's the tech stack that you're working on, both from a traditional software perspective as well as what are you guys building stuff in?

Kris Garin (31:20)
So AppFolio is our property management, know, MGL. So a lot of things kind of run off of that. We schedule repair and maintenance off of something called Zuper for like field services management. And, you know, we've got DialPath and...

for phones and we've got an asset management kind of investor portal called Juniper Square, which a lot of groups use. Some other stuff, that's kind of the core company-wide stack. What we decided in early 2024 was that we were gonna abandon everything that was an open API.

And so we really embarked on the project and we spent, you know, a fortune, you know, to do stuff that now I think you could do for a fraction of that. And probably a lot of that is going to get just chucked, you know, because like what you can do now is better, but that's fine. That's like.

that you know, that that's, you you buy the ticket, you take the ride. But so we, at the time, we kind of came in on the, on the tail end of no code.

So we really went all in on Xano, which is a low code, no code database. And basically we pulled everything, basically said like, we don't care if you integrate with that folio because everything integrates with us and we can make the connections there. so...

you know, that was, that was really the journey of, you know, 2024 to kind of get, get everything into that, into that mode. And we were able to do things really fast at that time. That was like pre-clogged code, you know, I mean, that was, know, and so, um, it, it, you don't really hear people talking about low code, no code anymore, which is really interesting, right? Um,

David Moses (33:04)
Just

everything is.

Jack BeVier (33:08)
Yeah.

Kris Garin (33:08)
It's just

right because it's, know, but, but, but I, and, and, and, and, and, and, and there's a comment I want to circle back and make about what's, what's not important and what's still very important. So, but we, so we got all that in and then we were able to.

to do some really cool things. So we decided we wanted to use ⁓ voice AI to answer phones and schedule maintenance calls and all that stuff. And we kind of demoed the various tools that were available on the market. And it didn't really feel that any of them really kind of understood what we needed to do. And so we stood up something in basically six weeks. That was, what's that?

David Moses (33:43)
With 11 Labs? Did you use

11 Labs for it?

Kris Garin (33:46)
We use 11Labs for the voice and then it comes in. So phone call comes in through dial pad, gets routed to the 11Labs agent. There's various kind of prompt injections happening so that it knows who it's talking to, what unit it's talking to. Ultimately it goes through the entire process of triaging a...

maintenance request, figures out the amount of time required to put on the schedule, figures out where that property is, who on the maintenance scale, so you can schedule the whole thing. would go to the appropriate maintenance tech with the appropriate skill set on the right side of town, pick two or three windows, and this is all conversational, right? And it's pretty good. We over-indexed on what was emergencies, because we were, and so it was, that cost us some overtime, but we figured it was

cheap

insurance, right? And then we've sort of tuned it. And so, yeah, we didn't get it where it is today in six weeks, but we went from zero to one in six weeks because everything was on that platform. You know what I mean? And so, you know, what we learned, so, you know, so that's, know, and now like all of our calls come in, right? And so maintenance calls get fully scheduled.

Jack BeVier (34:50)
And that's the end of it. And now, let's make this cool skit.

Kris Garin (34:55)
If it's a billing question, whatever, it'll get triaged and then it'll turn into a ticket. That goes to a property manager who's like, this is the issue, this is the resident, this is their history, this is whatever. So they're not going to spend 25 minutes on the phone listening to somebody's story. They're going to come calling back with a solution. And so that's been pretty good.

David Moses (35:14)
That's awesome.

I have couple questions about that. ⁓

Kris Garin (35:21)
Yeah.

Jack BeVier (35:22)
And

David, for Kris's benefit, give him give 30 seconds on your tech stack so Kris understands your context.

David Moses (35:29)
So we also use Dialpad. We basically use AppFolio. We also have a community management company that does like HOA and condo association management. That's in a software called Sync, C-I-N-C. we basically mirror all of our property management databases to Mongo, which is a, you know, it's a no SQL database. And we...

Essentially, we actually just stood up an after hours maintenance agent as well. And we're kind of struggling our way through, how does it get more context? How does it have that context faster? And kind of playing around with all of that. But that's essentially my tech stack. I do love Dialpad because webhooks for every step of the call, you can really do a lot with it.

But I guess my question, twofold. Number one, do you kind of create a certain number of tokens worth of context programmatically every... So that context is ready if a call comes in? is all the data just in the databases? And it's actually making a call.

Kris Garin (36:35)
Oh, it's all in the database. So it's

David Moses (36:39)
and trying to pull that and then pulling that data back.

Kris Garin (36:39)
making a call. So there's a tool that's making a tool call to, know, so it matches the phone number. So the first thing is, you know, hi, so and so that it appears you're calling from, you know, 123 Main Street, is that right? You know, and then it'll pull up.

David Moses (36:48)
Mm-hmm.

Kris Garin (36:59)
You know, it'll pull that up. so, you know, there'll be, you know, by the time it's done, you know, there'll be, it's, it'll create a work order and we don't use that folio for work order management, but they have to match. Zuper. Um, uh, because we found for scheduling and assigning and all that stuff, it's just, it's been, it, it, it, it, we, we kind of, we couldn't figure out how to do a good job, you know, managing a distributed workforce in the field using our folio to manage their time.

David Moses (37:09)
What do you use? OK.

I

agree with that. We use property meld, which I'm not.

Jack BeVier (37:28)
During the course of the.

Kris Garin (37:31)
Yeah,

PropertyMeld, we came very close to using PropertyMeld. ⁓ Yeah, no, we liked it. think the Zooper, they were super open. So we were able to get right on the phone with their product people.

David Moses (37:35)
I'm not recommending it or not recommending it.

Kris Garin (37:47)
they were excited that we were trying to do what we were trying to do. We just, had conversations with them. And so it was like, you know, so we were able to say, Hey, what could you do this? And they're like, yeah, yeah, we could do that. You know what I mean? And stuff. it was, it was, it was, ⁓ it was helpful in that way. use company cam. So now like, if I do cut, like we send somebody to do diligence on company cam, you know, like we're basically just using that as a front end to collect photographs.

You know, and then everything gets sucked into our database and the AI is doing it like cloud vision and all that stuff. we'll, you know, we'll, we'll analyze like the, you know, the, the, the, the plate, you know, on the water heater to inventory what the item is all that stuff. And like, so, but there's like a combination, like you have to do, you have to like do a little bit of description on the photo and, and, and have it append.

David Moses (38:31)
Mm-hmm. So.

Kris Garin (38:35)
You know, and then, and then it's like, if Claude is looking at the photo with that little bit of context, it's totally different.

David Moses (38:40)
That's right. Yep, is absolutely the key there is give

it. We don't actually, we haven't coached any of our, you say company cam, that where somebody is dispatched, they don't work for you, but somebody's dispatched to go and take pictures?

Kris Garin (38:56)
Well, it's for anybody, but we use it for kind of due diligence and acquisitions and stuff. So you can, you can assign anybody contractors, whatever, but it's very good about geo tagging. And then it's got all these AI features that we totally ignore.

David Moses (39:01)
Gotcha. Okay.

Yeah, it's funny because we went to dial pad because it had all these AI features and we didn't use a single one them.

Kris Garin (39:11)
Yeah.

Yeah. Yeah. Because

what the, the, what the, what the company cam AI thing, which is it's not bad, but what it does is you can walk through, you can narrate, take photos, talk to the thing while you're going, and then it'll produce like a pretty good report in PDF. It's just like a dead document. Right.

David Moses (39:34)
No, that's great. We wind

up ingesting kind of like everything that the tenant called in about the chat history, because in PropertyMilled they chat with the tech or the manager in PropertyMilled. Any phone calls, we take the...

Kris Garin (39:47)
Yep. Yeah.

David Moses (39:52)
the transcripts those or the summaries of the transcripts of those phone calls. All that becomes context in every single AI run through a photo. So it really has, you know, and then we analyze all the photos in a particular work order. We kind of give them a description and, you know, call it any red flags, whatever it is, on a photo by photo basis. But that all kind of just gets, that becomes the context for the whole work order review, which kind of, you know, because you could have a picture of this and it's missing a, you know,

Kris Garin (40:02)
Yeah. That's great.

David Moses (40:20)
some data that is in another picture. So it really needs the whole, you know, kind of the whole suite of pictures to do it. But Company Cam I haven't played with, so I really don't...

Kris Garin (40:22)
Yes. Yep. Yep.

So what we started, we've been using it for a long time, but like what I wanted to do was, so we're walking up a large multifamily property and trying to come up with a capex budget kind of see if we're gonna make an offer, whatever. And so.

⁓ we basically came up with, with the, it's a structured kind of insight going to each unit, you know, take these pictures. We've got a checklist in there, do these, right. And, and then all of that gets sucked into our database. And then, and then, you know, so that's, that's kind of one, one type of context. And then we've got, you know, our cost tables, another type of context. Yeah. We've got like our AIA contract format, another type of context.

David Moses (41:10)
Mm-hmm.

Kris Garin (41:10)
And it does like a pretty darn good first pass. Yeah. And it's like, just what we think. It's just an expression of our view, right? Like nobody has to do anything. It's just, it's just, it's just, you know, cause the algebra is in one, like there's, there's almost no AI in there. There's almost zero AI in there, right? It's all it's, basically just, it's like, it's like approximating square footage.

David Moses (41:25)
There you go. Yeah.

Kris Garin (41:33)
and like, you know, in like high, medium, low, like, is this a, you know, you know, are these mechanicals at the end of their life kind of midlife or new, right? Like very basic high level stuff, you know, that it's like, but then it's like, I see molt.

David Moses (41:42)
Mm-hmm.

Kris Garin (41:46)
Like Claude sees mold. Claude's like, want to take a closer look at this unit. And then also it's like, you can feed it. We always ask for like a 12-month history of work orders. And it'll say, there's a concentration of leak calls in this end of the building. Figure out what's going on. You know what I mean? And so that's where we focus. It's pretty great. ⁓

David Moses (42:06)
Yeah, that is, it's really, really

cool.

Kris Garin (42:09)
but it's all, and so this is the thing, like I think, so we, thing about low code, no code that like, were able to build stuff really quick, but it was like me building an Excel model. And I wasn't doing this by the way, we brought in somebody who did a bunch of stuff that I don't know how to do. It was very impressive, went very fast, but it was sort of like, you know, it ended up being kind of like a lot of, because you're just, you're just making connections, right?

And so, know, the hill that I, you know, am now prepared to die on, you know, for all this stuff is, you know, the marginal value of code is like rapidly going down to zero, right? Like code doesn't matter. Architecture matters a lot.

Craig (42:51)
Yep.

Kris Garin (42:51)
Schema matters

a lot. And what's interesting is like, you know, that the people who know how to code, you know, they're not important because of their coding ability. I mean, you need them for quality control and like all that stuff, but like,

If you didn't learn how to code, you're starting from zero on like these core questions of architecture, schema, things like this that really, really, I think define.

you know, the utility of something, the difference between something that can be very efficient or just be like this big kind of blob that gets very bogged down in itself or whatever it is. so that's been like my, like when I built these things that topple over, like they've been the most valuable. So we built a lot of sub-words. What's funny is like low code, no code, like it didn't enforce architecture. And so like we, so what we're now kind of actually going because

I'm sort of on the other side of this and it's not the only way to look at it, you know, but, you know, this is sort of where I, where, where, where I am on it is like where we've actually gone into a SQL world for database. Like we're doing everything in Postgres, you know, and, and,

but it's also super efficient because like nobody's writing SQL code, right? Everybody's just like, like, like, cloud code is great at SQL. So like that's, you know, so it's like, it's, you know, I spend like most of my time when I'm trying to build something, thinking about like, you know, the database architecture and like schema and all that stuff. I'm like, how do you, how do you get the stuff that is not written down?

you know, into a formal kind of structured way, because then you can delegate things, you know, at a completely different level.

David Moses (44:30)
Yeah, I think that is absolutely true today. Schema architecture, so ridiculously important. But I don't know if a year from now that's still gonna be true. Because I think that one thing that the models are getting better at exponentially is taking context.

in whatever format, in whatever haphazard way that it is that has been memorialized, i.e. a phone call, an email, a text thread. These are inherently unstructured, difficult to understand for right now, difficult to understand and make sense of and action on.

⁓ things. But I think that going the direction it's heading is that we'll want to just do what we do, right? Have conversations in natural language the way Claude code takes our ideas and creates some kind of structure and architecture and code behind it. I think that's kind of where it's going to head in the future. And I think that, you know, vector databases become

I think more important because they understand, you know, it's not just a cell and the field has to be of this type, right? It is, you know, someone's just gonna say what they're gonna say and yeah, she's gonna figure it out. And it'll be a question of how much context it has to have to figure it out, right? And that's my...

Craig (45:54)
I still think though,

to Kris's point though Dave, if it's not, I still think that a data analyst and a data architect are probably...

David Moses (46:01)
⁓ there's no doubt. just don't know if that's gonna hold true.

Craig (46:01)
So like, you know, and I'm not really, I don't know if that's

going away like.

Kris Garin (46:05)
How long

is going be there? I'm in violent agreement, David, with the overall comment. What I think we're going to find, I suspect we're going to find. And like, days on which I'm wrong about my big claims generally outnumber the days on which I'm not. take it with the appropriate grain of salt. yeah.

David Moses (46:23)
Me too.

Craig (46:23)
Yeah

Prediction from Kris Garren. Here we go.

Kris Garin (46:29)
You you look at like these, you know, these headlines like, oh, like the AI bubble doesn't work because you're these companies spend, you know, hundreds of millions of dollars and they get no value for the AI projects. And like, that's organizational, right? That's organized. like, you're going to, and you're going to, you know, you're going to be able to stroke a check to have Palantir park a bunch of forward deployed engineers or whatever that is in, you know, at Marriott and just kind of figure it out. Right.

But I think in the middle market, in the lower middle market, the level of, you know, and at the level of individuals who are not animated and excited by the idea of staying up till two in the morning, you know, the inner workings of their soul to Claude Coe. Right, which believe it or not, like outside of like this little Hollywood squares here is like most people, right, which I forget.

Yeah, because most people are like, what are you talking about? Are their eyes glazed over? And I think the big question mark is, is that last 3%, 5%, 15 % of undocumented context? Is that a pebble or is that a boulder?

David Moses (47:20)
You

Kris Garin (47:41)
right, is that the stuff that you can't get anywhere because it's like I know it when I see it and I don't really like to talk about it and I couldn't explain it to you even if I wanted to because that's not the kind of person I am. Like what is like where what is the importance of that and I don't think we know, right? That's where it's going to stop.

David Moses (47:57)
No, yeah, think the answer is we don't know.

Kris Garin (48:01)
And it's either going to stop at 85 % or 99 and a half percent or somewhere in between. And maybe it stops so close to a hundred percent that it doesn't matter anymore. My gut is people being what they are. It's going to, it's going to be material, but it's not going to matter everywhere. And it's like, you know, somebody explained like, is AI like actually intelligent? Like, you know, or, you know, like

If it can sort of fake it well enough that nobody can tell the difference, like, does it matter? Right? You know, for the purposes of like getting done, like it does not matter, right? For other purposes, it probably matters very much. like, you know, that's, I think what it comes down to. Because I think everything you're saying, David, is exactly right. And yet there will be things, you know, and it's like, is it for want of a nail?

you know, the whole structure collapsed, or is it, you know, actually we can build pretty, pretty well without that nail. Thank you very much. You know, that nail wasn't as important as everybody thought it was. It could, could really be either my, suspect in some combination, in some situations it's going to be one, some situations it's going to be another. But I think technically everything but that, I agree with you, goes away very quickly. And now we find out just how big a deal that actually is.

David Moses (49:15)
Yeah, my favorite ⁓ quote on predictions is, if I had a crystal ball, I would walk differently.

Kris Garin (49:21)
Yogi Berra said that, you know, I never make predictions, especially about the future.

David Moses (49:26)
you

Craig (49:26)
your comment of staying up and bearing your soul to it until two in the morning is hilarious. I had a an interesting few evenings where I decided that so I this very good friend and she was like a she was the global head of talent for BlackRock in Hong Kong. And then she kind of worked as like the whisper between like Fink and like all of his VPs. So brilliant and at the age of like 32.

So, she, she, I knew exactly what she was studied in, you know, like all of the sort of certifications that she had. And there's one in particular called Enneagram. And if you ever, and, and this woman was one of the most accomplished coaches and sort of that I've ever met. And, ⁓ she told me about getting this Enneagram certification. And she was like, it was the most grueling thing I've ever had to do, Craig. I would never recommend it to anyone and never go through it. So I,

I thought one night, well, let me just, let me just program. Let me just program a skill that it literally is an Enneagram expert. And it, and I told it, and then I said, and I want you to speak to me and sort of like an ontological coach. And there's a whole different coaching paradigm behind that, that I, that I trained it on. And when I tell you on the second night of having a session with this thing, it told me things about myself that I knew, I knew, I knew.

like, but like, no one has ever verbalized that. And I was like, Holy two nights of like talking to this thing for like a half hour. And it's telling me about like, subconscious contracts that I made as a kid with my brother, like, and it nailed it. And I thought to myself, like, what are the use cases, the infinite use cases outside of what we do in our daily lives that are so compelling, that are that that are frankly life changing.

and appear to be happening at a more rapid pace these days. And then like, how could you then tune that same knowledge base to talk like Tony Robbins to you, or talk like David Goggins to you or talk like, you know,

Kris Garin (51:22)
Ready, make,

David Moses (51:23)
or convince your staff to learn about AI.

⁓

Craig (51:28)
I'm telling you, man, it's

a powerful thing that I think will only get way more powerful.

Kris Garin (51:33)
Yeah, yeah, the persuasive I mean, you know, I think, you know, I've had it with and without the benefit of like various kind of like, you know, personality tests and stuff I've done with you like, you know, I just I, you know, I have

I have had it kind of like assess periodically, like my working style, right? And my decision-making style and like all this stuff and like some of these patterns. And it's very insightful. it definitely like, it's like, it's pretty good at like, I'm like, where am I making the same mistakes over and again? You know, like as I build it.

Craig (52:04)
Did you guys know that a couple of days ago, a

couple of days ago, Kris, OpenAI came out with, think, an add on, I don't know much about it, just kind of heard about it in passing, where now it can essentially record your screen all day long. so, you know, if you're a knowledge worker at your computer all day long, you're either training it to make it, you know, make you way better or training yourself out of a job. I think the jury's still out on that, but...

but it is an interesting, I think you'll be seeing way more of that from the anthropics in the opening.

Kris Garin (52:33)
Yeah, that's super

interesting.

David Moses (52:35)
That's the point, right? The

point is to train systems so that you don't have to do what it is that you do. The big question mark is what comes next. Because it's going to get there. think everybody, the sooner everybody just realizes it's going to get there. Whether it, you know, at some point you're going to jump on the train, right? It might not be at this station.

Craig (52:47)
Mm-hmm.

David Moses (52:57)
it might be at the next station or the station after that but at some point you're gonna jump on the train because the train's going.

Kris Garin (53:02)
Right?

The thing right now is I think everybody on this call is sort of, you know, obsessively trying to build their own station. Right? And that's exciting.

And like, I'm, you know, I can't help, like, I can't help. This is like what I'm doing, right? I'm just like, it is the most interesting thing, you know, to me. But like, you know, are you going to, is there going to be, you know, is somebody going to, is somebody going to just like pop up a nice station with like a nice cappuccino stand and a sandwich shop, like right next to the station we've spent the last three years, like trying to build. Then maybe it doesn't even matter, right? But I think that's like,

Craig (53:23)
Kris it.

Kris Garin (53:41)
So for the people, being early, it's just another way being wrong from the standpoint of investing or whatever. I mean, that's kind of a truism. think in this case, this is all time well spent. I think it's a legitimate question. Time will tell. Are we?

educating ourselves in ways we find personally interesting and we'll get some marginal utility. Or like how quickly does this stuff all catch up? And I'm kind of taking David's side of the argument about architecture and stuff more than my side. I think the barriers go away on more of like a five-year kind of timeline. And I think there's a pretty big window for people who are really doing this to compound that growth. But maybe it's two. I don't know. Maybe it's 10. We'll find out.

That's the only question I have about all this is not like, it coming? And is it going to be transformational? But like, what is the appropriate use of energy and effort and opportunity, cost and resources to kind of try to front run it? that, yeah, go ahead.

Craig (54:44)
Kris, have

you built anything sort of, you know, you guys are all titans of industry and real estate, but like, what are the, do you have any personal use cases that you've built anything for outside of your business?

Kris Garin (54:55)
Yeah. So I'm going

to, yeah, I'm actually going to ask Jack to, to, to, to, test for me. My, this kind of like obsession project I've been working on, but it basically, it's, it's sort of a, it's kind of like an AI assisted, you know, to-do list and notes app. so, so basically it's in, it's in, you know, email teams, Slack,

you know, and it sort of got, goes through and it picks up, you know, it goes through like 30, 90 days of your email and builds kind of like a basic map of your world, you know, and then it's sort of picking up, you know, 60-ish percent, 80%, you know, at first, you know, of like the stuff that would make it onto a to-do list.

Craig (55:40)
Interesting.

Kris Garin (55:40)
You know, and then, and then you can kind of manage and kind of clear. It's not telling you what to do. It's not telling you how to do it. It's not telling you what's important. It's not scheduling, but just the ideas. Like for me, I have so many things coming at me. like offloading the offloading, the cognitive load of maintaining a to-do list is like been one of my, you know, kind of, yeah.

Craig (55:48)
What is it doing?

Does it?

Does it then like sort of like put things in buckets for you? Like, so you can focus on this aspect or something like that. Like what's the output of what it's doing?

Kris Garin (56:10)
Yeah. So I've

got, I've got, I've gone, um, I've got, I first tried to have it put everything in buckets and it couldn't get the buckets right. And it was overly prescriptive, you know, and all that stuff. so, you know, it's focused much more on kind of like markers of, so everything comes in, it, you know, it gets sort of, you know, it maps the people nodes, like, you know, all this stuff for kind of like semantic similarity and like all this stuff. So there's a lot of stuff kind of going on.

Craig (56:20)
Yes, yes.

Kris Garin (56:38)
you know, behind the scenes, but what it's really just doing is it's saying, you know, this is a task. Do you want to do it now? Take it off the list. Has it already been handled or do you want to hit snooze and be reminded about it later? So, you know, it's a very simple interface. then you can also put your own, you know, can take a picture of a...

Craig (56:53)
That's interesting.

Kris Garin (57:02)
your car registration and save it on your notes, then you can query that at all. So it's got all that context. So basically, it's like the idea is very flat, very, very simple surface, but then it's aggregating.

you know, cause for me it's like, get five minutes and I'm like, what do I need to do? What am I going to do it? So I can like look at that list and every time I've tried to, then it's like, what do people, and it's also good at recognizing the nature of like open loops. So it's like, what do I owe people? What do people owe me? You know, what am I maybe not paying attention to that I want to, and then I can flag things. So I could say like, you know, if we were having a dialogue about this topic, I could create a topic.

Craig (57:25)
That's interesting.

Kris Garin (57:39)
and then things that come in, it would tag relative to that topic. And then when that topic's done, it just goes away. Right? And so, yeah.

Craig (57:44)
man.

That's very cool.

David Moses (57:48)
Did you build it in Claude Coe? Did you build it in like cursor lovable? What did you build it in?

Kris Garin (57:53)
⁓

I generally am doing clog code in, in, in, in, in VS code, like in the ID and then it's all in super base for, for, for the database. And then the app is in Verso.

Craig (57:59)
Yeah.

Is it have you ever done any like does it run on your iPhone as it's strictly desktop?

Kris Garin (58:11)
Yeah,

no, it's a web app. I mean, I've got it saved down as a web app on my homepage. So it's like, it's the, it's the main, it's what I'm using is like my main to do list right now. And like, yeah.

Craig (58:21)
Jack, have you have you

⁓ considered so like I just got access to build an app on top of outlook about a week and a half ago. I'd been working on it without access and I'm still trying like after I got access I was like, man, I'm going to develop emails and you know, just like I'll spend three seconds a day at know, five times a day in my email.

And I still have yet to sort of find the killer app for all of that access. And I'm wondering if you've had it. Jack, have you considered that? Like, what's how do we use? How do we take all of the email that we get and make sense of it in some way that like, you know, like, I just don't know it yet. I haven't I haven't gotten there yet.

Jack BeVier (59:00)
Yeah, so I feel like I'm going to I'm going to I promise I'm answering your question. I feel like, you know, because software has become so cheap because coding has become so cheap. Now everyone can create custom software and that's very exciting. So and everyone's going to create custom software for themselves, their particular use case. But if we ignore the differences in human personality, there isn't like there is no such thing as a productivity app that everyone's going to like.

because there are, I'm making up numbers. I'm completely making up numbers because there's 20 different clusters of personality in the world. And those 20 people are fundamentally biologically wired different. Like I, I believe that there's a biological basis of behavior that we call personality. And so like, you're not going to have a God app, you know, they're not going to have one app to, to rule all others because it's so cheap to create a custom one.

that Kris, I imagine you're building your perfect productivity app, but it's different, but it's gonna be different than my perfect personality or my perfect productivity app. Cause like my weaknesses are not yours and my strengths are not yours, you know?

Kris Garin (1:00:06)
So that's what I'm very

interested to kind of find out here. what I, you know, the initial versions of it, for sure, right? Cause ultimately I think if I can use this for different people in the company, all of sudden then it's like, you start to get some visibility in like priorities across the company and you know, things that are in the open loops and all that stuff. But I think,

I think as soon as you start saying like, you know, what's important and like how do do that? And like, then like 100%, you're right. I think what's the question is like what markers are there?

of what you think is important and what David thinks is important, what Craig thinks are important that show up in the metadata, know, that show up, you know, is semantically that independent of how you're using that can reflect the priorities you're putting on it, you know, but.

David Moses (1:00:55)
Yeah, I might be a little,

Jack BeVier (1:00:56)
Yeah, I thought-

David Moses (1:00:57)
I might be like a few degrees off from Jack's assessment in that I would say, I guess I disagree, but not completely. I think that the ideal productivity app kind of takes into account that there are all of these different types of personalities. And what it can do is identify where you're inefficient and where you're dropping the ball.

Jack BeVier (1:01:19)
Yes.

David Moses (1:01:20)
regardless of your personality type.

Jack BeVier (1:01:20)
Yeah, that's exactly what I think. I think, yeah, like the best productivity app would be one where you can select your personality and be like, there's 20 different, there's actually 20 different versions of it. It's all gathering the same data, but it's presenting things in different priorities and the reminders are of a different nature. And the to-do list is featured more prominently or less prominently as like a driving feature behind it. Cause like,

I get along just fine in email. I'm one of those people who I have zero unread emails. I have zero things in my junk mail and my to-do list is my flags. And when there are more than 15 flags, my stress level goes up and I work on those flags until my flags go down below 15 and then I can sleep at night. I am like, so I'm like, I'm like organization is not like the thing for me. What I could use help with is a reminder to have lunch with so-and-so.

and like to call some, know, to like call this person like that's where I am weak, but it is not, you know, so like where I'm weak is completely different than what you guys are weak. So like it's, you know, it's going to be a different thing that like actually makes my life better. But a better to do list is not interesting to me. Like the flags work just fine for me. Like.

Kris Garin (1:02:30)
This is why I ditched my aura ring. This is why I ditched my aura ring, you know, because it was like, I would get this like, it would, it wasn't because it wasn't interesting information, but then I'd get this alert saying like, you your readiness score is low today. So like, should really take it easy, you know, and I'm like,

Jack BeVier (1:02:33)
You what?

Craig (1:02:34)
You ditched your what?

David Moses (1:02:35)
is his aura ring.

What it's really saying

is keep your subscription.

Kris Garin (1:02:55)
I'm like,

Craig (1:02:56)
Solid

advice.

Kris Garin (1:02:56)
you're right. I should take it easy today, you know, which is like not what I need, like at all. What I needed someone to say like, you didn't sleep very well last night, so you're gonna have to get your shit together. It's not gonna be a great day. Focus, get ready to show up. And that's helpful. That would be helpful to me, you know, but I couldn't set that personality. But you will be able to in like five minutes, right?

Craig (1:02:59)
Ha

David Moses (1:03:01)
You

Jack BeVier (1:03:08)
So order a triple shot. Yep.

Craig (1:03:12)
You just ⁓ need my Tony Robbins guy.

Jack BeVier (1:03:21)
Craig

had this I Craig Craig introduced this idea to me three four months ago and you were and he was showing me what he was building with his open claw with Rocky and he built it in a way that it was like he had different motivators he was talking about the Enneagram and had different motivators and he was just like this is like this like look I've created this place where I come to work and I'm happy here because it reminds me to do push-ups and it talks to me like Tony Robbins and and I'm like my god that's fucking beautiful I would I this is not where I want to live but

Kris Garin (1:03:49)
Yeah.

Jack BeVier (1:03:49)
my God, like, look at this.

But like, there's a version, right? But there's a version for everybody. And so I think that, you know, it's going to be like, what's your role plus what's your personality equals what? That'll be like the, that would be a really, I don't know if that's going to like the end state, but that'd be a really interesting stop on the train. You know, it'd be a really interesting station for the train to get to.

Kris Garin (1:04:10)
So I think the last component that's missing there is fragmented context. And if whatever you're applying to whatever those role, personality, whatever, is not limited to these fragmented domains.

but can have a more holistic view of what's flowing across your screen, what's flowing through your world, all that stuff. That, I think, is where the architecture comes in.

Craig (1:04:37)
think when you when you

when you take a look at, know, like, I have I have my work and I have my personal life and there's sort of KPIs that that are, you know, like, if we're all it's sort of trying to be with our better selves. There's there's certain things that you have just have to do every day. And so part of the app that I built was like a dashboard of personal, like

you know, hey, tell me, tell me what my checking account looks like every day. Tell me what my, what my weight, you know, I'm going to take my weight, my blood sugar, cause I'm diabetic. Like all of these things that just, you know, and, so that's on a dashboard and the dashboard is, is obviously powered by something that knows me well and has a voice and sort of like its own personality. And like, that's, that's what I was showing the Jack. And I thought to have that in my hand, like, you know,

just to kind of keep me updated on exactly where I am and with my personal goals and my life goal and my work goals. Not a bad thing, not a bad little app. So I see where you're going with the productivity app.

David Moses (1:05:33)
As a visionary at my company, that's kind of my role. I think the perfect assistant for me, the perfect AI productivity tool for me would be something that recognizes what's not interesting to me and just does it. Not reminding me, this has to, just do it. Don't tell me about it.

Jack BeVier (1:05:33)
The ⁓

David Moses (1:05:54)
you know, don't remind me 16 different ways. Because all I built when I built my own AI assistant was a constant reminder of all the things I don't have time to do. And it's really just, you know, and kind of throwing in the face how poor of a delegator I am. And it's just, really what it's doing for me. Like I need something that's just gonna say, you know what?

Not this is something that's important to David or this is something that's interesting to David, but this is something David is gonna wanna do and this is something David is not gonna wanna do. And the things that David is not gonna wanna do, just go do it.

Jack BeVier (1:06:25)
Mm-hmm.

Kris Garin (1:06:28)
So I cut off, go ahead, Jack.

Jack BeVier (1:06:28)
Yeah, like Kris, you

will you mentioned you mentioned the context like filling in the context and and yeah, like I guess I just wanted to make this point that like, if you listen to a conversation, like if there's a recorded conversation or you know, and like, and I listened to it, and another person listens to it and another person listens to it, we that we did not all interpret the same things we did all not take away the same things from that right? Like

So so and so heard that Sally is upset. Another person heard that we have like some goals that were might not be on course on course to meet. And I heard that there's an appointment on Friday afternoon. And those are like the take. You know, if you ask me about like, what are the most important thing or what did you learn from this conversation? Number one, number one, two and three are different from three different people. And it's a function of their personality. In my opinion, it's a function of their personality. Like the data comes in, gets processed based off of different literally, literally physically different parts of our brain.

And we call that personality. it is those are self reinforcing mechanisms, right? Your personality gets more extreme as you get older, not less. Right. Like because those grooves just get deeper and deeper and deeper. So like to the extent that an A.I. can listen to that conversation and interpret it through those different lenses and understand the nature of your bifocals is different than the nature than somebody else's and then fill in those gaps. That's like, yes, I agree, David. That's a super powerful tool.

David Moses (1:07:51)
Brain-Computer Interface.

Kris Garin (1:07:52)
Uh, the,

Craig (1:07:52)
I think the

Kris Garin (1:07:53)
uh, the, the, the first iteration of like my list thing, um, actually, you know, was, uh, was set up to, to, to draft and send emails or teams messages. And so, you know, I could sort of say, you know, just, you know, you have draft response to this and it would show up in my drafts folder, you know, in outlook. could take a quick look at it and it would be, you know, and, um,

Like, so like, like that's as close as I got to like the delegating concept. What I found was,

I just, it didn't know me quite well enough yet. Like I didn't try, like didn't know, it didn't have enough context to know what, like it was quick, it was still quicker for me to write it.

And it was, at the end of the day, was a context and meaning issue, not like a mechanical issue. I think that the delegation function was available to the extent that the thing that had to be done was to respond to so-and-so's email that I've been avoiding.

Right. Like that's like, that's technically easy, but like, you know, the nuance is all in the context and the meaning and, ⁓ and it's, and it's, and the hallucination is expensive in that context. Right. You know, and, so like, you know, like low tolerance for risk, you know, and, and so that's, you know, for that risk, other risk I appear to be okay with, ⁓ somehow, but,

Craig (1:09:06)
haha

David Moses (1:09:07)
Yeah.

Jack BeVier (1:09:10)
Mm-hmm.

Kris Garin (1:09:18)
Yeah, I mean, think that's all it says. I think I think the personality thing, you know, can't be overstated Jack because I think it's it's the human thing at the end of the day. I'm like you rapidly hit hit this this point where you're not, you know, now you have things that are kind of complex and it's hard to explain. It's hard to explain why something's important and how something's important and why the situation is a little different than this situation. Right and that.

You I think it's, you you, you, you, you, you try to kind of, you know, AI stuff in that category and you run into the limits of, you know, of, of, of prompting and context management and all that stuff pretty quickly. And it's super interesting, you know, because that's like, that's where. That's where, you know,

you're gonna, the people who are good at managing AI, like are always gonna be necessary, right? And then those lines may change, but right here, right now, you know, that's where I hit it pretty fast, still.

Jack BeVier (1:10:11)
I

mean, this is just a this is just a prediction. But I think that like, you know, like the the dot sold the dot MD, the dots, the skills files. I feel like there's a missing piece of that before this stuff gets really good. And that is that you have to give it a personality like you like and it or you have to make or you have to give it all the personalities. But then understand that it's not it's going to be not you. It's not going to it's, like the the

If you don't give it a personality, it will not mimic a human. so to the extent that you need it to the maximally increase, you know, if you want it to interact with other humans, well, then it needs a personality. Otherwise, it's obvious to humans that it doesn't have a personality. And we're like, it's clearly just a bot. But if you gave it a specific personality, then it's like, this might just be Sally that I'm texting with, you know.

Craig (1:10:52)
Well, that's what they're, that's what they're.

personalities will all live inside of the data centers that are being built. That's where the intent models are coming from. It'll have perfect memory of everything you've said to it and the tone and the intent. that's what the five million square foot that data center is for.

David Moses (1:11:05)
It may be if

Well, that's the, I think, you

touched on it, Greg. mean, the reality, like, what is coming is the data centers that are coming, you know, whether these are terrestrial or in space, are, the idea is, I think the driving force.

Craig (1:11:23)
Yeah.

Kris Garin (1:11:28)
Are they going to be in space so that they can get solar power? Is that what they're...

David Moses (1:11:30)
in

Craig (1:11:30)
Yes, yes.

David Moses (1:11:31)
space because,

you know, who wants a data center in their neighborhood? I think that's...

Kris Garin (1:11:35)
Yeah. That's

an interesting take on space-based solar that I hadn't heard before. Yeah. Yeah.

David Moses (1:11:40)
yeah, Elon

Craig (1:11:40)
it's happening. Very much happening.

David Moses (1:11:42)
Musk

is, I think, driving that and think Bezos is probably not going to be too far behind. But whatever it is, it's context window. Like everything we've said, is, like why do we need context engineers? Like why do we need, why doesn't, why can't I build an AI with a personality that's consistent? Like why can't I...

What is the missing link with all these SolMD files and skills files and all the different... It's context window, right? If the context window were 10 million tokens, if there were no practical limit to what could be ingested and computed, then yeah, think that would be... I think from that perspective, we have a long way to go because even if you had a 10 million...

Craig (1:12:07)
memory files.

David Moses (1:12:26)
token context window, it can't reason through that in a fast enough time to have a conversation over the phone. It might be able to email back and forth, but that's okay. And that may be very, very useful. But in order to have the speed and the efficiency of what a human brain can do, it needs to be able to take all of the experience

cumulative experiences it's ever had distill it down to what is I guess what it can convince the the Being whether it's human or non-human on the other side that it you know that it did Come up with a meaningful and correct Response to what that thing is and I think it's I think it's context. I think it's just a tremendous amount of of context that is going to be

necessary to do that and speed the reason through it.

I don't know. It's all really, really cool.

Jack BeVier (1:13:22)
I that you can,

I think that you can like, I think that you can brute force intelligence with a large context window. Like that's kind of like our proxy for a more complex neural net is just a larger context window, you know, in the LLM as opposed to neural net. like, so like, yeah, larger context window certainly helps, but a smaller context window or a lesser model given

more clear instructions given better architecture and schema has the opportunity to perform just as well. You can use a lesser model with better architecture and get the same results as a stronger model with a larger context window, I would argue. So the best of both worlds would be the best model with a very thoughtful architecture and schema.

Kris Garin (1:14:04)
Well, that's the, I mean, think like, if you're building something, I think you wanna have, like, you might be calling three or four different models and different levels for different tasks along the chain, right? Because for exactly that reason, which is by the way, what we do, because like, when a lion comes.

you know, running around the tree, you're not like, hmm, like shaggy mane, like long claws, like big teeth, you know, like moving really quickly, like, let me think about this. It's like, like, you know, it's like, then you're up the tree, right, you know, before you even think about it, right. But then, you know, other times you're thinking about what is it, you know, what does it mean to be a human being and like,

David Moses (1:14:33)
you

you

Kris Garin (1:14:48)
You really want to slow down and get the noise out of there. that's exactly how we function. We're prediction machines. This is how unconscious bias works. This is why you get all kinds of...

Our brains work very differently than the LLM kind of prediction models, but we still like, we fill in the minimum amount of information required for us to jump to a conclusion, right? That's like why Thanksgiving is so hard, right? So like, you know, we like, so like, well, not amazing enough to still be doing it, thank God. But the,

Craig (1:15:18)
Kris, you are the king of metaphors, sir. You must have been an amazing journalist.

David Moses (1:15:22)
you

Kris Garin (1:15:26)
the like, it's why you have to fight against that so hard to be like a reasonable open-minded person. That's like what we are wired to do. And then we have to sort of resist that, you know, to grow. But it's like, it's this, you know, I mean, it's, we're still prediction engines, right? You know, in our, in our way, in a different way than the LLMs are.

David Moses (1:15:33)
Yeah, I think.

Jack BeVier (1:15:33)
Mm-hmm.

Kris Garin (1:15:47)
You know, and then, so yeah, like I've got like, you I've got my like haiku and then I've got my like 4.7 opus, whatever, right. You know, and. You know, I'm not, I'm not bringing the same, you know, loop inside my brain to like every, every situation anymore than I'm bringing the same model to like, you know, different steps of the process. And that's how you manage a token budget or latency or, you know, whatever it is.

Craig (1:16:12)
Anything else, Mr. Bevere, Mr. Moses?

Jack BeVier (1:16:14)
Now, Kris, really appreciate your time, man. Thanks so much. Looking forward to talking with you more about the stuff that you're working on. So thanks for your time today.

David Moses (1:16:17)
Thanks.

Kris Garin (1:16:22)
Yep, thanks guys, this was great. Thank you for having me.

Craig (1:16:25)
Dave, any last thoughts?

David Moses (1:16:26)
Well, are we going to go back to committing to building something in the next week?

Craig (1:16:30)
Well, it was such a lively conversation today. I couldn't show you guys my one shot from yesterday of a, of an app, but, be glad to show it off on the next one when it's at beyond a phase two of the build of nine. So hopefully finish that up today. I'll have something to show. Jack, did you, ⁓ so, Kris, Jack, Jack built an amazing tool and then it toppled over on itself. And on the last podcast he was talking about, like, yeah, I'm just going to rip it down and start over again. And your prediction was, is that

because you had gone through the pain of building the first one that the second version would be much faster, better, stronger. What's the outcome? Where are you with it?

Jack BeVier (1:17:07)
So yeah, I'm using, I'm ripping off this that you mentioned before at the beginning of the podcast, that John McCullen came up with this basically like starter pack where everything's pre-wired. There's a bunch of company context in there. It's just kind of like a, feels to me just like a stronger foundation to build. And the idea is like, let's, we'll continue to refine it and make it better and put more best practices in there and like put more, you know, put more architectural context into it. And so, and the company will manage that file.

Craig (1:17:29)
functionality.

Jack BeVier (1:17:35)
And then the managers can use that as the baseline and customize it for what they're interested in. I'm taking, I'm, I'm going down that path and using that, using that baseline. Like John gave it to me two days ago. So I haven't rebuilt everything yet, but I feel like I'm building on a stronger foundation. So we will see next week.

Craig (1:17:53)
Build it as a modular tool, sir. Not a bolt, not a bolt.

David Moses (1:17:56)
Greg,

Claude code or Codex 5.5?

Craig (1:18:00)
Don't know enough yet. I had a great session with it last night. I'll let you know on the next one. ⁓ I was impressed. It was ⁓ very, it was fast. 5.5 max was very fast. And I thought the, I thought that it's assessment of what I was building in cloud code was pretty spot on.

David Moses (1:18:06)
with code with 5.5.

Craig (1:18:17)
All right, guys. Well, listen, Kris, again, thank you so much for your time. Very generous of you. I can't imagine how busy you must be. So thank you for coming on. Love to have you back on again when you have, you know, when you have something to report, Kris, like when you're working on something like, so yeah, just, yeah. When you, when you really start using the tools. So thank you, Kris and all the best to you and folks. We hope you enjoyed this one.

Kris Garin (1:18:32)
Anytime.

David Moses (1:18:34)
Yeah, I mean, like when you're doing something with AI, jeez.

Craig (1:18:45)
You can catch us online on YouTube or anywhere where you can find podcasts. It's forked.ai and we'll see you on the next one.

Kris Garin (1:18:52)
Thanks.

Craig (1:18:53)
Thank you, Kris.