June 2026 AI Agents in Small Business: An Update

Welcome, welcome, welcome back to another episode of Lazy Leverage, the internet's number one small business operations podcast, downloaded over 4 billion times last week. I am joined by my heterosexual life partner, recently bearded, LeadSimple-wearing, property managing friend, Peter S. Lohmann. How are you today, Peter?

You got the middle initial right.

Good work.

I'm doing great, John. How are you, John K. Matzner? I don't actually know your middle name.

I'm gonna keep guessing. I'm not gonna tell you the answer. Uh, you have been moving around a lot, being able to take advantage of the fact that you have hit, uh, you know, you can travel and things don't die. How have your travels been to date?

Travels have been great. Was down in Outer Banks for a week with the family, back in Ohio right now. Um, we got some other trips. I'm gonna be going to Vermont later this summer for the first time, so But you are not doing a crane—

you are not doing a crane break.

I'm not doing a 4-week crane break.

I think I saw a newsletter about that, right?

Yeah. So I'm not taking off the whole month this year. I did it for 4 years, kind of like prove the point. Not feeling the need to be gone for 4 weeks this year. So I'm kind of breaking it up. I'm still taking about half the month off and then another, you know, several weeks throughout the summer. So—

Love to hear it. So today we are talking about a couple different things in the larger family of AI, but they are all coherent. You want me to kind of lead us into this and just jump right into it?

Yeah.

Okay. So as many people know, there is a lot of nonsense out there about AI. Excellent work. A lot of nonsense out there about AI. About 2 weeks ago, we got, I got, I organized about 15 business owners, all of whom did over $3 million a year in revenue, into a conference room for basically like 2 and a half days because, and I brought our president of AI, Zucky, who's like technical. We had a developer in the room, stuff like that. And it was a really, really fun experience where everybody got to show and talk about the challenges they're having and the problems they're having. And so, um, some of the things we're doing after building— we're almost crossed like 100 agents at this point, and we're accelerating the pace at which we can launch them. Um, and so wanted to talk about some of the kind of counterintuitive things we learned about actually putting this stuff to work in businesses. Not just from our own lessons, but what we heard in the room as well. And also, we'll also tease a new benefit for Crane members, but only if you listen to the end of the episode. Do you like that? I feel like they do that in real podcasts. So stay tuned for a brand new Crane member benefit, but only if you stay till the end.

Yeah, that's how the pros do it.

Okay, cool. So first caveat, which was everybody in this room was a a business owner. So, and it was specifically home and professional services. So we intentionally said no SaaS, no e-com, no venture-backed. We specifically did not want venture-backed people. Like, they're banned if they take venture. It's just a different mindset, right?

So this was a, this was a real-world, in-the-flesh meetup in San Diego, right?

Correct.

Correct.

A couple of days we did like, yeah, we did like you know, a dinner one night and then all day one day, dinner, half day the next day, and everything.

Laptops open.

Laptops open, questions, discussion, debate, whatever. And because I was like, as always, I just build what I want. And I was like, I want to hear where everybody's at. Mhm. You know, now we were probably a little bit selected to like, I would say, early adopters. So like our mutual friend Rich Jordan, who's become a really good friend now, um, was in the room.

You think like a dozen guys getting together in San Diego to talk about AI We're early adopters. Really?

What gave you that right? Do you think, wait, I can't tell if you think that's true or not. Yeah, of course. Are you breaking my balls here? Of course.

Okay. Like anyone who's gonna fly across the country to talk about AI.

What are you talking about? Are you breaking my, what are you breaking my balls here for?

No, I just like, you're so far on the bleeding edge that you don't realize that like most people aren't even talking about AI.

Fair enough. So the, The, like Rich is a good example, or Derek, if you know Derek, really smart guy who owns a shutter manufacturing installation wholesale company in Arizona, super, super smart and totally like, dude, he said in the thing and I was like, guys, did you just hear what he said when he was introducing himself? 'Cause we're all introducing ourselves and what we're doing. He's like, yeah, you know, I've got a GM these days, so I'd say I'm probably spending about 25 to 35 hours a week on, uh, coding tools for our company. Wow. This guy's like, and I was like, guys, you just, did everybody hear that? As everyone's like, anybody else in the city? We were all like, because of the, you know, the leverage you can feel. And we all feel kind of super early. Um, just in terms of, and, you know, he had some really good questions about what vehicle that manifests, but we can put about that separately. But what I wanted to talk about here and the why people should listen to this is I want to talk about a few. We'll see how many we get through. There's, I'm gonna do a bunch of write-ups about some of the things I learned, but there's a couple really cool and easy ones, um, that I wanna get your input on and kind of unpack with you.

Is that cool? Okay, let's go.

Good intro. Anything else up front? We'll do more of these, by the way. Um, we're gonna do more of 'em. I wanna do 'em probably at least every quarter, if not more.

Yeah. I mean, I was sad to miss it. Um, I'll definitely be at the next one. Just scheduling conflicts. Uh, cool. Yeah, it sounded awesome.

Yeah, it was great. You would've learned a lot and your, your brain would've been spinning. Um, and what I— the other thing that worked for us well in this context, and we'll see how it does in other things, which was Zucky and I were kind of the co-hosts. So Zucky's president of, uh, AI at Sagan, and he's a developer. He— I think he was a civil engineer by like undergraduate training, career entrepreneur, SaaS companies, whatever. He was kind of like the technical lead within the room, and I was kind of like the how-do-I-make-more-money lead, or like You know what I mean? Like kind of the, the, like, how does this— we're talking about GitHub. How does GitHub have to do with my HVAC company? Like, I'm good at that level. And Zucky's good at saying like, here's what you need to know about GitHub.

Yep.

That's great. So it was actually a very good, like technical and non-technical kind of like we alternated kind of sessions and modules and stuff. So it was really good. Okay. The first was, I'll do the, some easy ones first, but we've talked about this endlessly, which is global talent and AI agents are the same pursuit. It's just leverage, and it's all driven by the same motion, systems and pushing things downward and clarity of expectations. And so almost every single thing you say, you could swap out the word AI agent for global talent, and you can't tell the difference.

Yep.

Right. So that's kind of— It's just hunting for operating leverage.

Correct.

It's just hunting for operating leverage. And we will define operating leverage in this context as revenues increase linearly and costs do not increase linearly. Operating leverage. Like, that's the book definition of operating leverage, which is my revenues increase, and I do not have a proportional increase in expenses. How do I break those two lines apart? And global talent is absolutely a way to do that. And technology and code and agents and agentic workflows and automations are another way to do that. But I think a lot about breaking that line.

The challenge with both are, you know, global talent as operating leverage and to, to a certain extent, AI agents as operating leverage is, um, oh, on a long enough timeline, all of the advantage is eroded, uh, and absorbed by your competitors and you're back to square one. So like, and what I mean by that is if you were the first, if you were in the first 5% of property management companies to start working with global talent, Every dollar that you saved on payroll went right to your bottom line because you could hold prices constant and look, or, or lower them a little bit and look great, you know, versus your competitors. But now that 90% of property management companies are now working with Global Talent, a lot of that advantage has been eroded and it's all priced in. And so your margins, you're no longer enjoying the margin advantage that you did. Which means you've lost your operating leverage. Um, now AI agents are like the next phase of this. And if you, again, if you're on the bleeding edge of AI agents, which you are, which you are by definition at this moment in time in, in early 2026, um, you have the chance to run this playbook again. Uh, and you're going to be well set up to do so if you've been working with global talent, which is what John was just saying. Um, but it's unfortunately like eventually and, you and I get your opinion on how long you think this is gonna take. Once AI agents have proliferated throughout the industry, your competitors are all using them, you're, again, you're kind of back to square one.

There's a lot there for sure with what you just said. I think it, you know, that assumes effectively, you know, saying all margins get eroded eventually is another way to say that, right? Depending on the industry, depending on the category, depending on lock-in, all those kinds of—

Well, not all margins. Like, I'm talking specifically about like an operating leverage. Like, there's moats that are defensible, but, you know, working with global talent is not a moat.

No. Well, and that's— so, you know, I just pulled it up. Munger, Charlie Munger, has a quote that says— and it's not exactly this, but it's the same idea— he says— and I'll go on a rant, a short rant about this— he says, technology can either help you or kill you. The difference is whether the customer gets all the savings or some of the savings go to the shareholders.

Exactly what I was trying to articulate. Yeah.

It was like— and so it's like Global Talent can help you. Are you just— like these guys write, it triggers me. These people write on Twitter saying like, I'm a one-person company. And I'm like, bro, you're not solving for competition because guess what I'm going to do? I'm going to start a three-person company that answers the phone with a person. And then I'm going to show off that I built a billion-dollar three-person company. And then somebody's going to say, you know what I'm going to do? I'm going to build a 7-person billion-dollar company, and then I'm going to have a better value prop. It completely— it's not the reason— the firm size is also affected by competitive dynamics, and downward price pressure, and moats, and lock-in, and it is not solely determined by how many people I need to answer the email. It's based on oftentimes competitive dynamics, and I think that's completely ignored when these people talk about, well, like maybe in some industries that have these structural impediments to new competitors entering, but in most businesses, those don't exist. And so if you start showing off your Facebook agency, that's one person, I'm like, "Bro, I'm gonna have 3 people and have better account management." Yeah. Hey, I do exactly what they do, but, right? And so I think when you think about your value disciplines and your value proposition and the density of competition, you have to go back to your, On strategy. I forgot what it's— but you have to look at a market analysis.

Porter.

You have to look at— yeah, Michael Porter. Competitive strategy. Thank you. You have to look at ease of new entrants. You have to look at, you know, competitive differentiation. If you're one of two property managers in town and the other guy's like an old guy in his 60s, very different than if you're in downtown Chicago in a price-sensitive market with B2B buyers doing—

Yeah.

Right? Those are just different things.

Yeah.

So I think, yeah, there's a lot there, but I wouldn't use that. I would also— last thing I would say is billions, if not trillions, have been made in the difference between early adoption and mass adoption.

That's fair. Yeah.

Like billions, if not trillions.

Yeah. Just because the advantage won't last for 50 years doesn't mean you shouldn't exploit it for the next 12.

As Warren Buffett says, "In the long term, we're all dead." So it's like, well, in the long term, everybody will— It's like, "Yeah, I'll call you from my jet." Yeah. So yeah, I wouldn't do too—

Made with dollars that weren't defensible against a moat. You're like, "Yeah, I guess they weren't, but they're flying the plane pretty good." Correct.

"Hey, you want a lift? I'll pick you up in the plane." "Yeah, you're right. It was an undifferentiated product that I timed right." My money's still good here.

Yeah.

So, okay. So that was one, which was it's the same motion. It's systemization. It's leverage. It's all that fun stuff. Okay. Second one, pretty straightforward, but I almost want to start this with every single one of these kinds of things, which is theory of constraints and speed-based operating system is the filter for is AI actually going to make me more money? And that is increasing the contribution margin or unit of your constrained resource or on-time task completion of non-constrained resources. So we're not gonna sit here and think about CSRs answering emails quickly unless that's gonna manage my on-time task completion to a more appropriate range or increase the contribution margin of my constraint. I'm not gonna do a whole thing on that, but like we started with that because what I didn't want people to come out with was a bunch of tactics. Mm-hmm. That was just like, oh wow, I didn't know that Claude could do images. And then you spend all these calories and you spend all this time and you're like, I'm not making any more money.

Mm-hmm.

Yep. So I'll pause on that. But it was like, what do I do? It's like, we got a fancy new ERP. It's like, we're less profitable.

Right.

Because we didn't attack our constraint.

Yeah.

And—

Attack.

We didn't, we didn't, uh, contribute, respect, increase the contributions, respect our constraint. Yeah, go ahead.

Sorry. Sorry, Justin. Uh, the way I'm thinking about that now, per our last episode and others, is, you know, it's like a quadrant. You've got those two things you had just articulated, but for the growth engine and for the ops engine.

Yeah.

So there's like four, you have four opportunities, constraint for growth, on-time task completion for growth, constraint for ops, on-time task completion for ops.

Fantastic.

So like put it on a whiteboard and whatever you're thinking about working on, if it doesn't go into one of those 4 quadrants, it goes in the trash can.

It's fantastic. And right now, when you're— you have an open Saturday and Jim has the kids, rather than messing around trying to replace LeadSimple, right, go try to figure out how to enrich stale Zillow listings with MLS data to be able to send better direct mail to units that haven't leased up in 30 days that are in your geography. What are you doing? You're attacking your growth constraint, which is qualified conversations with third-party managed owners.

Yes. Or hire—

Yes.

Or hire somebody to do all that. Yeah.

And direct them at that problem.

Yeah.

Not we need to have headshots on the website and we don't have to hire a corporate photographer. We used AI. Right?

Cool.

So it's like, I say that again only upfront because like everything is downstream of choosing the right problem.

Yeah, and I think the only thing I would add there is if you're just getting started with AI, or even if you're not, don't let this prevent you from playing. Like, if it's fun for you to mess around with the headshots on the website and you're just getting started with Claude and you want to explore the image generation capabilities, go for it. Because I would, I wouldn't, I like, would I want everyone in my company and everyone listening to have is a working knowledge of the, the boundaries and the frontier of what's possible with AI. And you can't do that unless you actually mess with it and play with it and, oh, that doesn't really work yet. Or like, oh, now there's a better tool for this or whatever. So you, you need to find that mixture of play with an understanding that it's not gonna make you more money, but it is, it's like an education, uh, type of activity. And okay. I'm seriously, we need to hit our numbers for next quarter and I'm not gonna be playing around with image generation right now. I'm going to take the things I've been playing with and learning for the last 90 days and attack one of those four quadrants.

I love that. I think you're exact— there's separating learning AI from making more money with AI. Playing is a huge part of learning AI. Just playing and doing weird stuff and whatever, but just finding random things in your business and saying, I'm gonna try to use AI for this is like the worst of both worlds. It's neither play nor constraint. Right? So that's—

Yeah.

That would be an example of where it's like, no, either go play and try to do some fun stuff and whatever, or be focused and disciplined in what fires you choose to let burn versus which ones you focus on using this innovation. By the way, this is a framework for all innovation in your company, right? This is not just AI specific. This is, should I get a new CRM? Well, does it increase the contribution margin per constraint unit or on-time task completion in one of the two engines, right? Like it is the same exact question or answer or whatever.

Okay.

Questions on that one before I keep going?

This episode brought to you by McDonald's.

Quick, quick ad read.

Okay. Let's all figure, we should figure out ads for the show at some point, by the way. Like totally. Yeah, absolutely. Dropped the ball on that. So congrats, listeners, you're getting an ad-free Lazy Leverage for like 200 episodes.

Yeah. Okay. Um, 3, I'm gonna do 8 total. Okay. We're gonna do 8. So, cuz we're getting through these pretty quick. Okay. 3 was, I think within your organization, you need to have 3 or maybe 4 different layers of AI access. And this was based on a mistake I made. And you need to have different tiers of user. Now, a lot of what you read on the internet is these tech-native companies, and that is a different, that is a different beast. I'm talking about home and professional service companies. Okay. Level 1 is people who chat with AI and they can spot opportunities, right? Meaning that new executive assistant of yours. Should be able to chat with it to make her writing better, to make better decisions, to do research, whatever it is. She should also be able to say, I found this problem within the business that is a great potential AI use case, but I'm not gonna go to Claude Code and spend 7 days trying to launch this in Railway.

No.

Level 1 is like, I'm a chat user, I'm a spotter, and then the third one is actually, And this is a big one for us. I'm a user of point solutions that were developed by other people that use AI. So as an example, you could have a dev or you or whoever develop something that lives in Slack that calls Claude API, but it is a history of a particular address that you guys own, and nobody even knows that it uses AI in the background. It's just something in Slack that everybody in your team uses. And these level 1 people are users of AI point solutions solutions, but for the way they use it, it could be the magic Christmas elves.

They don't even know.

So that's what I think level 1 is, is chat spotting and user of point AI solutions.

Thoughts? How many levels did you say there were?

3 or 4.

Okay. Got it. Level 1. Clear.

You like level 1? Mm-hmm.

Right? My only question would be like, for these, for the chat, like go back to the chat for a second. What is the current best practice here? Like, do I need to buy like a Claude core?

That's what I was just gonna say. Whatever. That's what I was just gonna say.

Okay.

What I was gonna say right here, I literally have use whatever's cheap and already there. So if you have a Google profile, use Gemini. It comes free with Google. Gemini is for this type of user. You do not need to give everybody team access to enterprise cloud. No, whatever's free.

Okay. I feel like we're on the cheapo Google business plan at RL, cuz we have like 30 or 40 users.

Okay.

So I don't even know if we actually have that, but I will check.

Yeah, I mean, it should not be, you know, like there's ways around that, but like a free Chinese model is appropriate.

Like you could use Chinese model.

Really? Yeah, yeah, they're fine because you're just using it to like, you know, spell check this.

I mean, okay, but what if they're like uploading like rent rolls and asking questions and stuff like that? Like, is there a data—

That is not level 1. That is not level 1. I would not call that level 1. I said chat.

Yeah. Like, like upload a PDF and ask questions about it. That's, that's chat.

Okay. Uh, it starts to get complicated because what type of document is it? Are you doing this? I, let me get through the 3 and then we'll determine different use cases. Okay. Level 2 is what I call the coworker level. And generally speaking, they need to earn their way through 1 to 2, meaning they have to show that they're like AI pilled. And they graduate to level 2. Level 2, the comp would be like Cloud Cowork, right?

Mm-hmm.

Little bit more robust. You can save files, you can save your brand imaging, right? Um, they're not getting read/write access to systems of record, but they might get read access to the, the systems that affect their department, like read-only access to the rent roll via Rentvine. Okay, they don't have this global read. Why? Yeah, why are you smiling? Because I dropped Rentvine? Yeah, yeah, every time I drop a property management vendor, you always laugh.

Because you don't— the thing that's funny is you don't always pull the same one. Like, you could have easily said AppFolio or LeadSimple there, but you pulled out Rentvine. Mix it up.

Like, yeah, right, mix it up. Love it. Or Airtable, right? Certain bases in Airtable, they have read access to, right? Not global read/write to bases. Okay. Key thing here is you pull them up. You don't force them to this level. Like I've seen, you can't just say, well, you're the head of marketing, you should be at level 2. I've not seen— you need to have people who raise their hand. You want the kind of people who say like, I'm using it so much, I bought my own plan. Would you guys reimburse me? Totally.

That's like level 2.

Yep. And there are, there are definitely people in your organization doing this. And those are people that you, pay for the cowork plan, you give them extra usage, you train them a little bit more, you send them to Sagan training. Like, that's level 2, broadly speaking. And those are people who are like the early adopters within your company, regardless of seniority. Like, we have very junior people doing this, we have very senior people doing this. This is not judged by seniority. It's like how AI builder. And then the third one is, um, basically what I would call the builders. And those people are basically deep and unrestricted access. These are the people that are building the point solutions. These are people that are experimenting. They have read/write access to every single system of record. Cloud code, baby. Cloud code, Max Plan, frickin' Hammer Time USA. Those are your level 3 people. In a lot of different companies, that's just the owner, depending on your size. That's just the owner, right? For me, it's Zach. Right. Zach, who's like been with me for 10 years, Bincy, uh, um, Zucky, our head of AI. Yeah, obviously him. There's, there's like 4 of us, 5 of us, but everybody else is like, why do you need that API key? Right. Because I do not want people with read-write access to my Airtable who are like, have, you know, bypass permissions messing with like a key data source. And then the last one, I would say the level 4, whatever, is what I would say is, like, essentially external or expert providers. Meaning in some instances, like, if you don't have, like, full-time dev in your company, full-time developers in your company, you need to raise the hand and say, I need, like, I need the next level, like, even my level 3 people, because I'm a property management company, I don't have a team of backend guys, like, we're gonna put this in front of real customers, I need to throw some dollars at professionalizing this agent. I'm not going to just ship something vibe-coded to all 1,000 of our fucking owner investors.

Yep.

So I would call that like a kind of level 4. Okay, I'll pause on that, digest.

I think it's very coherent. I hope the model providers eventually make this a little more built into their products. Like, you could select these different levels within your Claude professional subscription or whatever, like your corporate subscription. Um, but yeah, I mean, this all makes sense to me.

And— But like, let's say what it's not. Let's say what it's not. It's not giving everybody Claude code, right? It's not having random people saying, we use a lot of AI at this company, so please build some things to make your workflow better. It's not doing company-wide training on cloud code with a bunch of freaking maintenance techs. There are people doing that right now.

You told me that you used to, you were doing like hardcore executive training or for everybody on like coworking and everything, and you scaled that way back. And this is like your learnings from that.

This is my, this is, which is like, there's gradation and, and you can't, you can't go to your head of maintenance saying, okay, we're an AI-fueled PM company now. And they're like, uh, I use it to rewrite my emails. Now you might have some global talent from Guatemala who's in that group who's just like, I put my own dashboard. I hope it's okay. I plug— and you're like, yes, you, you're level 2. Like, yeah, yeah. You keep going. Right. You go hire a level 3 person or the owner does it. Right. So like, that's where, that's where this is best practices distilled. Like there's companies that are like, we mandate everybody uses a certain amount of tokens every month. And we're like, What are you doing? Because you're gonna have people building the same things. You're gonna have, like, you're gonna have leaking access. Your security is gonna be a shit show. You're gonna be having, you know, different versions and version control you need to figure out. Like, come on, guys, you can't, you can't half-ass your way into this. Like, you have to be disciplined. And this is a result of kind of mistakes that we, everybody in that room had seen or experienced. I really— there's a lot of wisdom distilled in there.

I really wonder if we got some older folks on this show who were around when intranets and the internet sort of became a thing in business, because I have to imagine that like when business intranets, business networks, and when the internet became available broadly to corporate America, I have to imagine there was a similar dynamic where it was like some, some owner of the company, like, was at his buddy's business and saw that everyone at the company had a computer that was networked and was like, yes, everyone now at my roofing company will have a computer with a network connection and everyone will sit at a computer. and use it to do everything, right? And back then the software was garbage and there was like, everything was horrible, right? But, but it was so clear at the time that this was the future and it was very powerful that once you saw it, you couldn't unsee it and you just sort of felt like everybody should be using it, right? So I just wonder if we could talk to some people who were around during that time and like let them reflect some of these learnings. Maybe we could save ourselves some trouble. Yeah. And they would be like, oh yeah, like we had the same exact thing with, you know, desktop computers. Here's what we learned. Here's what we tried that sucked. You know, here's where it actually, like, we did this and it put us way ahead of our competitors. You know what I mean? Like, this isn't the first technological revolution that businesses have dealt with.

And that's when we talk about theory of constraints, the whole Eli Goldratt thing was about rolling out ERP, not the ERP process of ERPs, the SRP. It was a different. Term for it, right? But absolutely, this is all, you know, history doesn't repeat itself, but it rhymes.

Yeah.

Right? That's the old line. And so I think a lot of this rhymes. But when you look at things like companies figuring out that they spent $15,000 on tokens and they didn't realize it, they're going to start saying like, what ROI did we get? And they're going to come back to Ethereum. Like, so a lot of this stuff is like, hey, I'm playing a few moves ahead here. I see where this is going. Maybe this isn't the right answer, but these are the right, like—

This kind of reminds me of like—

—questions.

The old joke about like, people make this joke about like boats and pools. It's like, you don't want to own a boat. You want to have a friend with a boat. It's like, you don't want to be on the bleeding edge of AI, but you want your good friend to be on the bleeding edge of AI.

Correct. You want to be a Sagan or Crane member.

Correct.

And they'll tell you what you have to do.

Let them figure it out and make these decisions.

One of the things that, One of the things that we came out with is that we're gonna basically turn a lot of our internal member-facing content into AI stuff. And it's gonna be like, here's a 3-minute distillation. Should I give an Open Claw to every member of my team? Short answer, no. Keep listening if you wanna hear why we think it's too early for home services companies to give Open Claw to everybody. It's like, thank you. Thank you. Now I don't have to worry about it. You guys will let me know.

That's great.

It's like an internal education piece. Okay, quick, quick one that's a little adjunct to what we just said, which is a huge difference between experimentation versus production. When you're building AI, saying I got something to work on my local instance of Claude code and I was up all weekend and it's this like morass of freaking plug-together shit and API keys and whatever.

I don't know what you're talking about. I've never done that.

Never heard of it. And it's like this, it's, it's like this duct-taped creature. And then you go to your COO who has real jobs and you're like, hey, we can analyze rent rolls based on Airtable data. And they're like, what in the eff is this aberration? It is wobbly and brittle and the access isn't right and it's— this is not ready for production. This is an experimentation level build.

Mm-hmm.

And we all as an organization, we're talking a lot about taking things that start in experimentation, which is where they should start, and keeping those separate from the people who are working in Airtable right now every single day. And if Peter comes in and starts going, hey, I built you an AI tool so that you don't have to respond to emails anymore, and it's an experimentation tool, and then it breaks in 3 days, and Peter becomes an IT guy because you kicked everybody out of Airtable to save $99 a month. Like, there's a huge difference between experimentation and releasing this to a 725-door property management company into a core process.

So real. So what is the solution, actually? How do we bridge that gap?

I mean, that's, you know, come to the next Sagan X event. But what I would say is, what I would say is there's just, there's a lot there. Like there's a lot. I mean, what I would say is that, you know, one of the ways that we're at least trying to position ourselves as an agent offering is you talk to Zucky and you are head of AI or folks like him or other providers, whatever, like good ones, not like some guy who used to be a Zapier consultant 3 months ago. He just talks about it differently. He talks about version control. He talks about maintenance. He talks about Git. He talks about— and I'm like, me put calculator HTML site, send email. And he's like, wait, what? But what if they need to roll back the changes because the HTTPS certification gets decertified? I'm like, okay, so can you do all that? He's like, yeah, man. Like, I was like, okay. So I think some of it is knowing when you need that next level of professionalization, whether that's in-house and you want to bring in a backend dev or a whatever, or you want to try to find, you know, an option to get some help on the— like, that's like saying, when do you know you should self-implement EOS versus use a coach? Like, what would be your answer to EOS, Peter? When you're ready to get to the next level? Like—

Yeah. My answer to that would be when you burn out on trying to implement it yourself and get bored.

Yeah.

And stop caring.

Or you hear, or you hear somebody talk and you realize there's another level and you can't access it on your own.

Yeah.

You know what I mean?

Yeah. I mean, you literally bought the domain shipyourweekendproject.com.

Yeah. Because again, I solved my own problem.

Yeah.

Because I'm wonderful at experimentation.

And this, this is for the AI agents, right? Yeah.

Yeah. We're actually doing more than agents, but yeah, it's agents. But the— but this is all because I solved my own problem, which was like, I got it working locally and it's a little janky, but I want to release this to 50 people, you know? So, okay. Questions on that one?

Nope.

Okay. We talked about that. We talked about that. Experiment versus production. Okay. My favorite part of the whole thing was Zucky talked for almost 2 hours about source of truth versus systems of record versus endpoints or like apps, and then the way that they all feed each other. What I would say is it's some very basic stuff to the developer community, but all of us now are basically amateur developers. And so you have to have an appreciation for a database architecture, because that's the reason that Claude code thing you tried to build isn't working, because you have conflicting sources of data and you don't have a way to refresh and update your source of truth. You have a bunch of sourcing conflicting systems.

So you, you know, I have an appreciation for a single source of truth, John. You know I do, because when I move the source of truth, I leave a Word doc that says This has been deprecated. Please go over here. So yes. And, you know, going back years in conversations with Jordan at LeadSimple, talking about system of record versus system of engagement, where LeadSimple is a workflow tool and that they were designing, designing it specifically to be the system of engagement. And it plugged into tools like AppFolio and Buildium, which were the system of record. Yeah. And there was, you know, a really sort of clear distinction between the functions of those two pieces of software, how they talked and interfaced with one another, and the importance of not letting your system of record spread into other tools that can now fall out of sync. And it— yeah, it's—

It's all of that. What I would say is if you're getting stuck at like the endpoint layer, Right. Go like, like you need to look deeper. And that was what was profound to all of us is because a lot of us had just been playing at the like, hey, I just point my Claude at my Airtable database and say, find me this. And it sometimes is inconsistent. It's like, yeah, because you have multiple sources of truth and you have, you also plugged it into your CRM. So it sees the same contact and it doesn't know which metadata is tagged to which comp, right? And then Zucky did a whole thing on, You know what ELT is?

Mm-mm.

Incredible, right? Extract, load, and transform. Getting data into your source of truth, pulling raw data out of your apps, agents, systems of record, loading it into the warehouse, and transforming it to match your schema, right? That's an ELT system. And then a reverse ELT is pushing data back out from the source of truth into your systems of record, the apps, the agents, the endpoints. And like, to a developer, they're like, you idiot, that's like day 3 of like, you know, software engineering. And for all of us, we're like, oh my God, my ELT system, because the schema— and it was so— because we've been playing like outside that room and getting frustrated. And then it's like, no, you just haven't systematized your ELT layer. We're like, oh, so ServiceTitan is really where people work, but we need to be feeding like You know, a Postgres database, which is actually gonna be something that we can carry with us even if ServiceTitan cranks their pricing.

That's the dream, man. That's the dream.

So we'd spent like 2— so you see why this event was like awesome, right? Like this is the crap we were talking about for non-technical kind of like tinkerers.

The way this shows up in our little world of property management and, and maybe yours too, is this concept of owning your data. Yeah, this is a conversation going back 10 years in the property management space of, does— do I own my data? Does AppFolio own my data? Do my property owners own their data? What data do the residents own? If I terminate my account with AppFolio, can I get my data out? Can I load it into another software? What can they do with my data? Can they sell it? Can they, you know, run reports on it? Like, all of this—

Anonymize it? Yeah.

Yeah, huge point of contention.

That's downstream of all this stuff. Yeah, yeah.

So it's really— They're charging for API access, but that's actually my data. So why do I have to pay to access my data? Blah, blah, blah. And you know, the, the, of course the fantasy is what you just kind of said as an offhand comment about ServerSight where it's like, oh no, our data is actually in a data lake or Postgres database or whatever you wanna call it. That's where the system of record is. And all these other tools we use interface for that.

No, no, system of truth. No, no, system of truth. Sure. That's where the system of truth is. System of record where people live and breathe is ServiceTitan. But there's this ETL layer that's pulling things in and out of— or sorry, ELT, not ETL. ELT level. That's, hey, my team likes to use ServiceTitan. Over here in my marketing team, they use HubSpot. Over here in my maintenance team, they use Jobber. Right. But it's all synced into your Postgres database.

I think we just reinvented ERPs. I mean, that's essentially what, like, big companies, that's what an ERP is and does.

Well, and one of the things I talked about, which you and I have talked about on the pod before, which is a lot of this stuff is simply just democratization of stuff that has always been available to billion-dollar companies.

Yeah.

Meaning having a data lake, it's like, oh, so you guys what, do 100, 200 million? You're like, 6 million, or whatever it is. And so like a lot of this stuff, you know, developing internal software is nothing new. What's new is a $30 million HVAC owner saying— That's correct. I wanna own my data and make it transferable and not have to rely on a SaaS guy who's gonna raise pricing every, every month on me or whatever.

Right. Yeah. And there's something interesting about this, like at a small, in a small business, if you write your own software, historically, that's been a negative to your enterprise value. So like if I'm looking at another management company to buy and they tell me they wrote their own software. I don't even— I, I'm just like, actually, my wife's calling. I gotta go. I'm sorry. I don't even continue the conversation because I don't want to try to absorb some weird homegrown, filled with bugs, weird trust accounting situation into my operation. They tell me they're on Buildium or Appfolio. Great. Why? No, I know how to handle that. Yeah. Yeah. Now I wonder what's going to happen moving forward. Is it— is, is the stigma going to come off of homegrown solutions running the backend? Great question. 1 to $10 million businesses, you know.

What I will say that I think AI definitely does, I'll give you like 2 specific examples. I think Rich, the night of thing of the event after this talk, he's like, hey, I just cloned my whole ServiceTitan into Postgres. He just, he just did it that night.

Amazing.

Because he was just like, yeah, I want to get started. Like, it's not perfect yet, but I just want to like get like, so I think some of the historic hooks are, are it like the, the, the transfer cost has reduced significantly, which does one thing. The second thing is, and you know, it's where I'm sure Jordan, if he was still running, uh, LeadSimple or whatever, would be having some serious conversations is, you know, we built a tool. I, I won't be specific, but we can build stuff in the browser that behaves like a human to get data out of closed systems.

Mm-hmm. Yep.

It's not that hard, guys. It's not that hard. As long as you're not like a spammer and abusive about it, which we're not, we're not running 100 API calls a second. We're systematically pulling things out of these systems that historically thought they could triple prices because we're not going anywhere. And it's like, yeah, come build an agent that just syncs any changes in your X database and pushes it into your Postgres. We're like, yeah, we can build that.

There's some real drama about that happening in our industry right now. There's a guy named Taylor Howe who built a Chrome extension called PM in a Box, no relation to RPM in a Box.

Yeah.

That does exactly what you just said, logs into any, your PM software of choice and acts like a user to get all of the data in or out as if it was a user using it. And he's been very public about it and he's gearing up for a fight with AppFolio and others. And he's almost daring them.

They have to come after him.

He's almost daring them to do that. And he's being very public about it. So, everyone's kind of like wondering what's going to happen because if they try and like— they can't really lock him out because it looks just like a normal user. So, then what could they do?

They can't find him.

They can sue him.

They can't find him. The thing is, is that because it's installed on Peter's machine.

Right.

They can't find— they don't know which of their people are clicking buttons by hand and which which ones are using a browser. If you do it right, they can't tell.

Yeah, like randomize the delay between interactions and stuff like that.

We can defeat CAPTCHAs. We can do all that. We are doing that, right?

So it can get more interesting, I think.

Correct. Correct. Okay. Questions on that one? System of Truth, system of record endpoints, but pretty useful, right?

Is there a— did you do a write-up or anything on this?

Not yet. I will. I've been doing my GTD series in my newsletter. So I'm gonna do it after that. Okay, cool. Okay, next one. As compared to what? Perfectionism is the enemy of AI adoption. Lots of people are saying, oh, it hallucinates. And I go, how's Jackie doing? The person who currently does this 10 times slower. What kind of error rate is Jackie making?

Totally.

Right? So, oh, you know, it's sometimes it's like, let's look at 20 of Jackie's emails.

Yeah. Human office worker error rate is 3%, by the way.

Okay. As compared to what? So, you know, we, I've, I've been public about this now. We use in a first step of our screening, we use a voice AI agent, which is exceptionally good. It's better. And I say, awesome. My other option to get the same throughput was to hire 150 people that I pay about $5 an hour to read interview questions that they do not understand. The option is not my head of recruitment, Sophia, doing this, because I need to do 15,000 interviews a month. Option 1: quadruple Sagan prices. Option 2: have my time-to-fill database go from 1 week to 6 weeks. Option 3: hire 150 people to do 8 calls a day, day in, day out, 5 days a week, manage, pay, compensate, fire, whatever. It's like, or I deal with 3% of candidates who say, I don't wanna interview my screening interview with AI.

Okay.

So you have to, you know, it's some old Thomas Sowell, as compared to what? Not what are the issues with AI resume grading? What, you think you're humans? Come on, guys. Come on. As compared to what? Mm-hmm. So that was a big theme, you know, for the kind of skeptics. Okay, keep going. You want to comment on that one?

Yeah, I feel like I have something to say there. I mean, I— just to generalize this out a little bit, this is the same issue. This is a tale as old as time where it's like, I mean, let's go back to something like online property management software. When that first became available, people were like, well, you know, it doesn't, it doesn't have the previous lease right next to the new one. Like, here's my file folder, and here's the old lease, and here's the new one. And I can just compare them back to front and look at any differences. It doesn't do that, does it? You're like, well, no, but I can also sign a lease from Maui. So I can't do that with file folders. So there's always gonna be kind of like the skeptics and the—

Trade-offs. Yeah.

People who are afraid of change and, and stuff like that. It's just the way you frame it up. And you love to say that everything in life is trade-offs and you, you have to go into it. I think part of what's going on is people are so, you know, there's a massive AI backlash going on right now, which I try to talk to people as much as I can because people like us don't run into it much, but it is Like, let me just tell anyone who's listening, most people don't like AI. I don't know if you know that. Most people hate AI. They hate the data centers. They hate the stupid results in Google that are wrong half the time. They hate being on the phone with an AI agent. They hate getting emails that are clearly AI. They hate the AI chatbots. They hate hearing about AI is gonna take their job. They hate hearing about how billionaires are getting rich with AI. They fucking hate it.

Yeah.

So just know that, like, your frontline team does not like AI. They're default negative on it. So just monthly reminder for everyone listening.

Well, and, and it reminds me a little bit of something from a few episodes back that we talked about that I really like, which is, uh, you wanna have every single piece of food that goes out of your restaurant perfect. You need to run a chef's table with 6 seats. You have to be prepared if you want to keep, if you want that and you can't stand to see any mistake made. Awesome. Run a chef's table, 6-seater. You want to open 52 locations. You need to make some trade-offs around error rate and issue, right?

Or even fill one restaurant.

Yeah. Correct. Yeah. So everything's trade-offs. Correct. Everything's trade-offs. So it's like, if you don't ever want to use AI, that's awesome. Have 100 doors. Have yourself like perfect. Or you go to a market where there's no competition. There's a million ways to survive and thrive without AI. But if you come to me with certain types of problems, I'm going to say, here's the trade-off. And I think it's worth it. For us, it was we can interview roughly 100x more candidates. Yeah, okay, I don't mind if I get an email every now and then saying I don't want it. Okay, fine.

Yeah. So when you're dealing with a team member or a partner who is skeptical on your AI solution, that's going to speed some process up by 100x, just recognize they're coming to it, having been oversold. On AI by marketing nonsense and just generally having a negative AI sentiment. Just knowing those things going into the conversation, it's gonna help that go over a lot better.

Last two and then we'll bring it home. Ready?

Mm-hmm.

Okay. I gave one specific example, but I will give others, which is more and more I am starting to view more things as a technology problem. So we recently let go of our very wonderful, nice— our head of content. And what I realized was the only way to meet my expectations with the amount and the quality of content that I expect is to treat this first and foremost as a technology problem. It's an engineering problem to start, which is, do I have enough raw inputs? Do I have enough transformation templates? Do I have enough examples of what's working on LinkedIn? Do I— it is not an exercise in copywriting to start. It is not an exercise in customer service to start. It starts and gets 90% there as a technology problem. And I now put Zach, who's one of the most AI-pilled people that I know, he's our acting head of content, and we are producing more and better content. And then I'm gonna bring in the artist at the end to work off the back of the technology—

Sure.

—tooled process. But I'm not gonna hand an artist saying, hey, use a shitload of AI cuz I wanna be creating 50 different pieces with tracking links and A/B testing and promoting different pieces. She's like, what? I'm not techno. I was like, if you're not technology first in the way we're building our company, you're gonna really struggle. And I'm viewing almost every problem we have as a technology one these days to start.

Mm-hmm.

To start.

Your content example is very, very Hermosi-like.

Of, uh, make more content. Is it Tomo Z?

Yeah, he's all about just making more content and like the algorithm will figure out what's good. Just—

And testing things.

If you're not making 100 pieces a day—

Yeah. And, and I realize that when I'm leaning on my non-technical head of content, she's gonna go from 4 to 8 pieces if I really crack the whip. And if I go to Zach, Zach's gonna be like, we have 400, we just need to rank them, which ones we want to post in what order. And I'm like, That's the right problem. And we need to track which ones are working so that we can put ad dollars behind it automatically.

Yes, that's the right thing.

Correct.

Correct.

Not— you know what I mean? So I'm viewing more and more— what percentage of the problem is solved technologically varies, but I'm starting everything by saying, okay, I want to get back to our customers faster. The first place I'm going is not a customer service person. It's saying, How do I inject technology into this and then build humans on top of it to raise the bar, raise the bar, raise the bar?

Yeah, it's reminding me a little bit of like 10x is easier than 2x type thinking where it's like—

For sure.

You know, you're scaling, you guys are scaling, you're scaling quickly as a company and you're, you have specific domains in which you want to scale very rapidly. And yeah, this isn't, you're looking for an order of magnitude change, not 2x. And the people—

and importantly, though, the people to give me that 10x are not going to be people who give me the 2x.

Mm-hmm.

Right? If you're like— you can't go to a normal accountant and say, "I want to have a live dashboard and have the books on every property closed every day." Mm-hmm. What you need to do is go to a hungry, smart, tech-infused accountant and pair them with a dev and say, "We need to pipe straight into our bank account data using an agent." And like, that's how you're gonna get there, not them responding quicker to email and hiring 7 people, which is where they're gonna wanna go.

Yeah.

So it's like, it's like, and it, it ties into the last point that I think brings everything together, which is as of right now, and I have an article on my desktop that I wanna write about, which is almost everything you're seeing right now is faster horses. Almost everything you're seeing is faster horses. And we're not fundamentally looking at core workflows and thinking, if I made this agent first, what would my expectations be? And the reason why I'm focusing on faster horses is you have to get faster horses to learn enough to build highways and suburbs and cars. You have to get really into it. And right now, human beings are still delivering a lot of the value proposition of a third-party scattered site resident. And you, you have to go through faster horses, but if you were starting from zero with PM and you started with 2 developers, a part-time maintenance guy, and somebody answering the phone from the ground up, would you ever come within 100 miles of Airtable? Or would you just make it all text message native and have it all go into an ingestion engine? And you're like, like, and what does the reformed value proposition look like? So I basically just said, guys, we're all just still in the horse business. Right. This right now, because we're trying to jam a bunch of file folders of PDFs into an agent, and the agent doesn't like that. Yeah. So that was kind of like— it was like a— it was real motivating for me though, because I think about what would you look like if you built an agent-first Scattered Site residential property from, from the freaking studs right now? What would it look like? It doesn't look like what your Chrono looks like.

It would be so fun. I mean, I would do good, good content. Yeah, good content. Like, I would just— I would just declare, like, first thing that comes to mind is I would declare we're getting to 500 doors with no employees. Zero. And the insane things you could do on the journey there, like, the first things come to mind is with all the money that you're saving on labor, like, certain— like, everything's trade-offs, right? So certain— the quality of certain things is going to go down. The quality of owner interactions is going to go down. The quality of tenant interactions when they want to talk to someone is going to go down. You know what solves a lot of problems? Money. So I would go to every owner and every tenant and say, just so you know—

First year's free.

Yeah. Just so you know, this company operates with no employees. Therefore, you're getting a 20% rebate on your management fee every month. And you, you resident are getting a 10% rebate on your rent every month paid by me, the property manager, not the owner. Uh, you good with talking to an AI agent when you've got a clogged toilet now?

Yeah, by the way, by the way, by the way, texting it because that's instant. It's fast. It has access to your entire file, every transaction, right? Like, or you can go the other way. This is all blue ocean stuff, by the way. Redoing value propositions based on overdelivering, meeting expectations, under— this is all blue ocean stuff if you like this stuff or competitive strategy stuff. Or you could say, Look, here's what we're doing. We're going to have no back-office employees, but I'm going to have 3 incredibly well-paid owner-investor concierges, we call them. And their only job is to just drive around Columbus and help our owner-investors. And that could be rekeying a lock, that could be putting a gift basket for the new renter, that could be working with the sheriff on an eviction. And it's a concierge, and I'm going to pay them well, so it's a high-status, great job. They're super— so happy and whatever, but I'm not going to staff a 24/7 call center.

Yeah. I'll give you another crazy idea. With all the new agents and technology, you can now track which properties are consuming the time of your— time and resources of your company. And you could just bill for that.

Yeah.

You could do a Costco model for property management. You could charge $10 a month per unit. Plus costs.

You know, I almost did that with Sagan, right? I almost charged recruiters by the hour at cost. You pay the membership and then, yeah, use my recruiters. They're $15 an hour. They're well trained. They'll search all year for you.

Yeah, exactly.

I don't give a shit. They'll just allocate to your project. Like, yeah. Yeah. And so, so those are all examples of where I can't— I'm so happy with the stage of my career that I'm at where I feel like I'm first in line for the businesses that are gonna be built that are based on cars, not faster horses. And right now, you know, I'm thinking more and more about, uh, this was, I think, my favorite thing that I said in the whole event. And it came up cuz of something Zucky said, which was, everybody says you need to build, or not, I shouldn't say everybody. Sorry, that was a presumptuous statement. A lot of people right now are saying you need to build for agents and humans. And I actually think that's wrong because, you know, you're building for— I was like, no, no, you only build for the agents because do you know how easy it is when something is agent-readable to make a front end for a human?

Sure.

Do you know how hard it is when you just track sales numbers on a chalkboard to get that into a format that the LLM likes? You build for agents and then I can spin up a— oh, you want to do it in Spanish? Via your phone, that's a 30-second build. But the other, oh, it's a bunch of file cabinets and a chalkboard. We want to do an LLM. It's like, kill me. And so, if you build everything AI native and agent native, I think the type of companies and the value you can create for your customers are just— because you don't have all this legacy tech shit. You know, you don't have to change management. You don't have the—

Great.

So yeah, I think it's a super, super exciting time. So those were my— so I'll just summarize. All right. Global talent and AI agents are the same pursuit. Theory of constraints is the filter as to where AI agents actually make money. 3 or 4 levels of access. Not everybody needs cloud code. Source of truth, system of record, getting your business AI ready. Faster horses versus cars. Experimentation versus production. Know the difference. Most things are a technology problem. And then as compared to what, it's all trade-offs. So those were some— I have more, but I thought these would be good to talk about.

Great stuff, man. Yeah. Loved it. Good. Loved it.

Final, final. Oh, well you should end with the announcement because we teased it in the beginning. Let's end with the announcement.

Yeah. So, uh, the big announcement is that Crane members are going to get access to AI Agent builds by Sagan. And John and I are working out the commercial details. We're on the tail end of getting that finalized. It's gonna be an insane deal for Crane members as always. So, um, we know that there's a ton of value that can be unlocked with pretty straightforward agent bills, but not everyone has time or inclination to do that. Sagan has built a whole process to solve exactly that. They've done 100 agent builds for Sagan members, Sagan Prime members, and we're bringing that to Crane members. And so really excited to announce this soon. Uh, we'll have, you know, pricing and all that, uh, Yet to be finalized, but we're close.

It will be eye-wateringly affordable is what I would say.

I can tease that. So, yeah. So yeah, that's session. So cool. Yeah. Um, good.

As always, uh, grant application window's not open.

Um, can't join. Sorry. If you have any, don't join.

Sorry. But any questions, please do not DM Peter on Twitter because I think I'm probably the only one who checks that. Uh, actually I don't actually check it, but, uh, send him a message on LinkedIn.

Peter Lohmann. Yeah, I do LinkedIn.

Yeah. All right. Uh, thanks as always. Please leave us a rating on Spotify. It's great for discoverability. So if you've never done that and you're a loyal listener, please go down and hit 5 stars. It makes a big difference. Uh, we're also on YouTube now. So, uh, follow us on over there. Thanks everybody.