How AI Helps Andres Martinez Manage 100+ Co-Living Rooms With Almost No Staff
Show summary
In this episode, Andres Martinez shares his innovative approach to real estate, including building over a hundred co-living spaces using creative finance and AI-driven management systems. Discover how he’s transforming property management and investment strategies with cutting-edge technology and systematization. In this episode, we explore the cutting-edge of AI in real estate and property management, discussing innovative tools, legal considerations, and future possibilities. Our guest shares insights on automation, biometric security, and the potential for AI to revolutionize the industry.
Key takeaways
- Andres' co-living buy box: 2,500+ sq ft, 3+ bathrooms, corner lot, no HOA, built 1980 or later, because he needs to underwrite deals at roughly 50-55% of value to safely refinance private-money partners out in 5-7 years.
- A seller-financed 1,800 sq ft, 6-bed/2-bath house bought for $235,000 with only $10,000 down (plus ~$10,000 for rooms/furniture) produces the same net profit as an 8-bed, 3-bath, 2,500 sq ft house because the PITI is so much lower.
- He requires a 6-month minimum lease and a deposit covering one month, so if a tenant leaves he has a buffer of economic occupancy while he re-rents the room.
- Running the AI on premium models (Claude Opus, GPT-5.5) initially cost $3,000-$4,000/month in tokens; he solved this by building his own software that runs locally on his computer under a flat Claude subscription instead of per-token API billing.
- He built his own web browser and property management software (Casero) to get around Facebook Marketplace and other platforms' terms-of-service limits on automation, and warns this risks account bans/shadow-banning.
- His AI assistant (nicknamed Kim, for 'Coliving Intelligence Model') is described as a harness, not an agent, that operates with system-wide access and goals rather than fixed prompts, which let it independently call 911 and email domestic violence services during one incident -- a guardrail he later had to correct.
- He recommends using Claude Code instead of the standard chat interface because it can build missing tools itself, centralizing documentation in Notion, and warns against training systems on synthetic data since it does not reflect real-world bottlenecks.
- Roughly 65% of tenant messages arrive between Friday 8pm and Sunday 4pm; automating responses during these off-hours (lockouts, supply requests, maintenance issues) increased tenant satisfaction because issues get resolved immediately rather than waiting days.
In this episode
- 00:00Andres on learning the architecture of AI
- 01:26Current deal flow in DFW: an overpriced, tough market
- 02:31Why Andres' buy box and private-money structure make acquisitions harder
- 03:35What makes a good co-living property: layout, size, seller finance example
- 05:13Managing roommate friction and the co-living business model
- 06:30Zoning, unrelated-occupant codes, and the Room Act pushback
- 08:07Building AI systems instead of hiring property managers
- 10:41Token costs, Claude Opus vs cheaper models, and runaway AI incidents
- 13:05The 911 domestic violence incident and fixing the AI's guardrails
- 16:25Building proprietary software (Casero) and a custom browser to bypass platform limits
- 21:19Advice for investors: Claude Code, synthetic data pitfalls, beta testing with 14 users
- 26:34MCPs explained and why common property management software lacks them
- 31:47Could AI run a $50M commercial apartment complex?
- 36:01Tenant satisfaction, decision fatigue, and the McDonald's kiosk analogy
- 38:57Final advice: build one AI operating system, not multiple agents
About Andres Martinez
Andres Martinez is a DFW-based real estate investor who has built more than 100 co-living rental rooms across Dallas-Fort Worth since starting in February 2024, using private money partnerships and seller financing. He quit his job as a waiter in November 2024 to focus full time on the business, taught himself AI architecture through resources like Anthropic's Claude Code certification practice tests, and built his own proprietary AI property-management system (an assistant he calls Kim) along with custom software called Casero. He runs his school community Coliving Unlocked and has been featured on BiggerPockets and at the Co-Living National Conference.
Listen to the audio
“I feel like I know what I'm doing. I just don't have the license to prove it.”
Andres Martinez
“So it's chaos. If you don't know what you're doing, it is absolute chaos. Every possible bad scenario that you can imagine, it will happen.”
Andres Martinez
“Stick to one, stick to your first girlfriend, man. You find a winner, marry her.”
Andres Martinez
Questions answered in this episode
Can you run a co-living house with 6-8 unrelated tenants in a single-family home without it becoming a commercial property?
Andres says it stays classified as single-family, but cities have code limiting homes to no more than three unrelated people living together. He argues these codes are outdated and that cities like Austin are pushing the Room Act to abolish such restrictions, which otherwise force people who can afford a room but not a full apartment into homelessness.
What size and layout of house works best for co-living conversions?
Andres' buy box is 2,500+ square feet, three-plus bathrooms, corner lot, no HOA, built 1980 or later. His smallest house is actually 1,800 square feet with a U-shaped layout he converted into a six-bedroom, two-bathroom with the kitchen in the middle; he bought it on seller finance for $235,000 with $10,000 down and spent about $10,000 on rooms and furniture.
How did Andres handle the token/compute cost of running AI for property management?
Early on he was burning $3,000-$4,000 a month in tokens using premium models like Opus and GPT-5.5 because cheaper models lost context. He solved this by building his own software that runs as a harness on his own computer under his Claude subscription rather than paying per-token through the API.
What went wrong when Andres gave his AI tool too much autonomy?
A person texted at 2 a.m. claiming domestic violence and asking for a room; the AI, following a 'human safety' goal, emailed domestic violence services and called 911 on its own without his approval. Andres had to go back and clarify the guardrail was meant for tenant safety inside the home, not external situations like that.
Is AI property management easier for single-family co-living or for large commercial apartment complexes?
Andres says commercial buildings are actually easier to automate because they already have on-site maintenance and cleaning staff, whereas his co-living portfolio requires juggling multiple handymen, canceling and rebooking them in real time, and handling constant human-to-human friction between roommates.
What AI tools and setup does Andres recommend to someone starting out?
He recommends using Claude Code rather than the regular Claude chat interface, because Claude Code can build its own tools when it hits a limitation instead of just saying it can't do something. He also suggests using Notion (free) to centralize your processes and documents so the AI can pull from one knowledge base, and warns against training on synthetic data since it doesn't reflect real operational problems.
Resources and links
Full transcript
Andres Martinez (00:00)
I think it came out on its own, man. it’s a little bit scary. So I spent a lot of time learning the architecture of artificial intelligence. Like for these guys, you can go to Anthropic website, they have a great course on how to become a Claude Code certified architect. I cannot take the licensing because they only allow you to be to take the license if you’re an employee of Google or if one employee of one of those big tech companies. However, I took the practice test multiple times and I got a hundred. So I feel like I know what I’m doing. I just don’t have the license to prove it,
Dylan Silver (00:59)
Hey folks, welcome back to the show. Today we’re joined by Andres Martinez, DFW investor, who’s built more than a hundred co-living rooms across DFW in just a few years using creative finance, private money, and systems to turn what could be a management-heavy business
Into something he says takes about twenty hours a month to operate. He’s been featured in BiggerPockets and also has a school community. Andres, thanks for taking the time here today.
Andres Martinez (01:26)
Thank you for having me, Dylan. Thank you so much for the invitation. Glad to be here.
Dylan Silver (01:28)
What types of deals are coming across your desk these days?
Andres Martinez (01:32)
Overpriced.
Tough market right now in Dallas-Fort Worth. more in Dallas than in Fort Worth, but everywhere. It’s getting a lot. for me right now that I would rely on a wholesaler to bring me a deal. There is far and few betweens that actually make sense deal-wise, and they those go like to the VIP buyers, like you would never see in the list. So most of the deals don’t really make sense right now unless you’re doing your own acquisitions, which at the moment I am not doing anymore. So it’s been a slow year as far as buying for me.
Dylan Silver (01:57)
Now, when we talk about underwriting deals and you talked about how it’s challenging to find deals, I think this is a sentiment echoed by folks throughout the real estate space. if you’re looking at single-family, if you’re looking at multifamily or larger multifamily and intercommercial, there’s a lot of folks sitting on the sidelines. But you’ve built the sizable portfolio. Do you find that right now the good deals are pocket listings, so to speak, that are being held by brokers or
Do you have to go out and do a lot of the acquisitions work and you do the old-fashioned knocking doors, driving for dollars, calling foreclosure lists and so forth.
Andres Martinez (02:31)
Yes, if I wanted to get a good deal right now, I would have to do the own acquisitions. But here’s the problem and also with my niche, I do co-living, which is room rentals. So I just can’t buy any BRRRR or any deal that makes sense as a long-term rental. It has to be 2,500 square feet or higher. It has already three bathrooms, it has to be a corner lot, no HOA, 1980s or higher. So my buy box is extremely niche. And on top of that, I buy all my deals using private money partnerships. Right? So we lock in for like five, seven years.
So I need to underwrite for that year, making sure that at that time we can refinance that they can get my partners out safely without having to took them out of pocket myself. So if you add all of that, I need to be buying deals at like fifty five, fifty percent. Which a newer investor is gonna beat me on price because they’re gonna happily pay sixty five, seventy percent. So it gets a lot harder unless I’m doing my own acquisitions right now.
Dylan Silver (03:20)
Now when we talk about co-living spaces and you mentioned twenty-five hundred square feet. Give us an idea of the types of properties that make a good use for this. imagining need to have a lot of rooms, right? That’s gotta be a basic requirement, unless you plan to add rooms, right?
Andres Martinez (03:35)
Yeah, so we had rooms to all of our houses. My smallest house right now it’s eighteen hundred square feet, and that is a six-bedroom, two-bathroom, but the layout was perfect, man. It was like a U-shaped, four bedrooms, and then the living room we turned it into two more. The kitchen is in the middle, surrounded by two bedrooms. So it’s like three and three. They share. the price at that time was I think two thirty-five on seller finance, only ten thousand down. So it was a no-brainer. We only spent like twenty ten thousand for the for the walls, the rooms and the furniture, and then it was running right away.
So that house makes me the same net profit as an eight bedroom, three bathroom, twenty-five hundred square feet because the PITI is so much lower. So at the end of the day, what matters is how much money you take in your pocket. If I tell you right now, hey, my portfolio makes me like thirty thousand dollars a month net, does that mean anything if I’m only taking five because I’m a small percentage operator? Of course I take more than that, right? But we need to underwrite for all of that.
Dylan Silver (04:24)
Co-living, I wanna dive in and get granular here. there’s a lot of folks who are evaluating which real estate asset class to become an investor or to grow their portfolio. And clearly you’ve been successful and passionate about growing in the co-living space, and I think more and more people are. we talked with the CEO of PadSplit, I’m sure you’re familiar with on this show. And I think one of the things that’s challenging but also interesting about co-living, you have
More room, so more a cash flow opportunity, but also more moving parts potentially. some folks might say, well that’s gonna be challenging to manage all these people under one roof. You’ve clearly done a great job at limiting the management challenges. walk me through that, what management looks like in the co-living space.
Andres Martinez (05:13)
Yes, it’s chaos. If you don’t know what you’re doing, it is absolute chaos. Every possible bad scenario that you can imagine, it will happen. So people come to me with that first question like, Man, you’re gonna have a lot of roommate friction. Yes. You’re gonna have three people, yes. And that is a sacrifice I’m willing to make on top of believing this business model and the cash flow because the other options what are mid-term rentals, short-term rentals.
[UNCLEAR IN SOURCE: “Most of that is already banning an entire dialogue world.”] And it’s an entire business model that is more hospitality-based. Co-living is also hospitality, but that’s like it’s a roulette. You’re in the constant, bringing in people, bringing up people. And for example, you mentioned that our company, PadSplit, they’re also a type of hospitality where they do weekly rentals. Imagine having eight people where they can live at any point weekly. Right? Now you have 10 properties and it’s a full-time job plus you need to hire staff. It’s impossible to keep up.
Right. So very early on I was like, okay, if this is gonna work, I need to minimize my operations. So I started in February 2024 and with my first house I told people like six months minimum. Right? Collect deposit covers me for one month. So in case they leave, I have a month of economic occupancy that I’m not gonna lose to find another renter. ‘Cause at that point I didn’t know how fast I was gonna bring the new person in the room. So I didn’t know what my occupancy rate was gonna be in.
So that worked. On that first house I still have five members out of seven bedrooms. They’ve been staying with me since February March twenty four.
Dylan Silver (06:30)
Pivoting here slightly, when we looked at, six and eight people living in one single-family home, it does it become a commercial property at that point?
Andres Martinez (06:37)
No, it’s still single-family.
Dylan Silver (06:38)
Okay. are there any hurdles associated with that, with permitting or what have you, if you have that many people in one single-family home?
Andres Martinez (06:45)
Yes. every city has some sort of regulation in Dallas-Fort Worth as to how many people you are supposed to have in a home. But the code reads like this, right? you’re not supposed to have a family with more than three unrelated people. So for example, right now in my own home, right, I have my aunt, I have my mom, and I have myself and my wife, which by the way do not count as related people. Related people is mom and dad and kids. So I’m already breaking certain code.
And all these rules were written, and very, very far away. A lot of them were to prevent people from having multiple slaves. So it’s like way back on the code. Yeah. And that’s why we see all over the country there’s a big push, especially coming out of Austin right now, the Room Act, trying to abolish all these rules because you’re pretty much telling people, what, you have to be homeless. You cannot rent a room here. I know you can afford it, and this will let you get on your feet. Like a lot of my college students, a lot of my guys who just got divorced, they cannot afford a weekly hotels. They can they to pay child support.
So they have to sleep in their car, they have to go on the streets, or they can rent a room with us long term, in a safe, clean space. So that’s the issue with the zoning.
Dylan Silver (07:45)
Yeah, i co-living seems to be on the rise. And I certainly think there’s a lot of people who are interested, maybe not sure how to go about it. And like you said, they could be cautious or wary of the challenges that go along with it. But you’ve built a large portfolio. At what point do you decide, I’m going to have, property managers responsible versus I’m gonna do this all myself?
Andres Martinez (08:07)
I’m still doing all myself. I never had a property manager. And that’s why we have to talk about the systems and the pain that went through that. as soon as I found out about AI on late 2024, this is when ChatGPT was just coming out. I started learning about not using the ch the chat message, but like what is this thing? Like is this really like are we gonna be into the Terminator future? Is this gonna evolve like that? And once you start understanding the mechanics of how the artificial intelligence works, the genius out of the bottle.
It’s there’s no way back. That thing is gonna take over eventually, some point in humanity, right? So right now, how do I harness that? So I start this is like man, like late twenty twenty-four. As soon as I quit my job, I quit my job in November twenty twenty-four, went to New York with my wife, and now were doing New York, and I was just writing systems. Like how do I systematize this? What is the moving process? Okay, what is the maintenance report? What is the marketing strategy? And then I was like, There’s twenty two steps, this is too much, seventeen, thirteen, twelve. And once you shrink every system and you feed it to the AI, it can do it multiple times. Right? And then
Most of twenty five and early twenty six, the problem was the context and the memory and the memory with AIs. Because they cannot keep up. Right. A moving process is like a ten day step. Right. And then in the process, I’m texting another tenant that is asking me, Hey, is this room still available? I have a friend coming. Hey, no, already was rented. I don’t want to be involved in that. So how do we automate that? So I had to work a lot on the architecture of the artificial intelligence of how to keep it make context. And right now it’s been now forty-five days that I just removed it because before it’s been running for me since like March.
Where you can do everything. But I’m still watching every conversation. Like, okay, what do you say? No, no, no. I didn’t like that. There’s times where somebody approaches for a room and they said, Hey, my husband is beating the hell out of me. I need a room tonight. Right. And this was like two AM in the morning. And I get a call from the AI of the emergency, like, hey, by the way, I contacted domestic services and I contacted 911. And I was like, turn yourself off. That you’re not supposed to do that, right? And I’m getting 911 call to my house or to my rentals.
Talking about domestic violence. ‘Cause I let it go way too soon. Right. So all these little things and adapting ’cause once dude, at the point of where now once you give it hands, it just runs off.
Dylan Silver (10:41)
So with these proprietary tools that you made, did you partner with third parties who handled this? Are you using like a OpenAI API key and reloading that with credits? Walk me through this.
Andres Martinez (10:51)
Yes. So that was the problem early twenty six and why I couldn’t fully deploy it, because I was burning like three, four thousand dollars in tokens. And ’cause for this period over a month.
Dylan Silver (11:01)
Holy cow.
Andres Martinez (11:02)
Because you cannot use the cheap models for this, for this reason. So it always had to be Opus, it always had to be ChatGPT five point five. Otherwise it loses context because the model the way it’s architected, there’s no prompt. There’s no sub agents. It’s just goals. So if the goal is to bring a tenant safely into the home,
That’s why it had the authority to call 911, domestic violence, contact another tenant. There’s multiple things that this thing has that ID and train on it. But because it has a goal and it has all the access to all my system. Like there was a lockout a couple of weeks ago. it asked the tenant for their ID on the camera and a picture, checks message, it texts another tenant. So it checked the locks to see who was the last person to came into the house. Okay, it was room three. And it texts the room three, Hey, by the way, Ryan, can you open the door? Because this guy is forgotten his code.
‘Cause and you read the conversation, they it’s like, okay, this seems to be the right tenant to move in, right? He’s just using a different phone number, but I have his ID and he’s in front of the car, the license plate matches. Should I open the door or should I contact somebody else? Before contacting me. So all of these things that you don’t teach it, but it has the guardrails as to what is safe, and it’s just that’s why you have to keep an eye on it. Like what are you doing and why are you doing it?
Dylan Silver (12:03)
Now, did you pioneer this? Did you follow someone else’s protocol in order to get these interlocking systems like the AI responder that’s looking at the record of who’s in the home? How did you come up with all of this?
Andres Martinez (12:17)
I think it came out on its own, man. it’s a little bit scary. So I spent a lot of time learning the architecture of artificial intelligence. Like for these guys, you can go to Anthropic website, they have a great course on how to become a Claude Code certified architect. I cannot take the licensing because they only allow you to be to take the license if you’re an employee of Google or if one employee of one of those big tech companies. However, I took the practice test multiple times and I got a hundred. So I feel like I know what I’m doing. I just don’t have the license to prove it,
Right? And
When you take out the prompts, when you take out the do this, do that, and you just tell him this is what I have and this is the goal and why this is the goal and show me your steps. And it starts like just running like crazy, like, let me try this, let me try that, let me try that. And once you start seeing that’s like, Okay, I like that. I like that. And it’s just you start telling him like put this here, put this there, and it starts building the systems by itself.
Dylan Silver (13:05)
Amazing. did you have an engineering background before this?
Andres Martinez (13:08)
No, I was a waiter before we sit.
Dylan Silver (13:09)
Unbelievable. When we talked about utilizing AI You gave one example, which I think a lot of people will say, wow, that’s tough, right? Where the AI is calling, emergency services, it’s like, I gotta just figure this out and run with it. At that point, what did the next day look like? Did you take this thing completely offline? What happens after that?
Andres Martinez (13:30)
You read no, I read the logs right away because he had already placed the email to domestic violence from my property management email and he had already contacted 911. Right. But AI doesn’t have voice connection. So he just called multiple times. Like he was trying to reach for help to this person that only sends us one text message on Facebook Marketplace. So after that he calls me, right? And my phone, by the way, I don’t have notifications of anything except my Telegram bot, which is Kim. Kim Coliving Intelligence Model.
Because it’s not an AI; it’s actually a harness. It’s not an agent, it’s not a flow, it’s not an ASCII, it’s a harness. Like you can compare it to Hermes or to OpenClaw. You give it tools and it just runs, right? So it called me and I was like, what is it? Like I don’t see anything on the tenant messages and I text them, what’s going on? Check first marketplace. I open the first marketplace and I see that. And it’s like, okay, why do you do that? And I was leading reading to the guards and one of them is human safety.
As soon as human safety, nine one one, right? We don’t mess with that. The problem is I didn’t tell it human safety for our tenants, not from an elite. So it just it didn’t know what to do. It just read human safety and I start it started fighting up.
Dylan Silver (14:33)
This is mind blowing to me. So you’re going through this and you’re figuring this out on your own. As things are happening and you’re may having to make adjustments like this, that’s a huge example where everyone’s gonna say, wow, the tenant getting locked out is pretty mind blowing as well. What about, we you talk about new clients and interacting with new clients. What about, signing leases and walkthroughs and so forth? imagining you still are involved in that.
Andres Martinez (14:56)
Yes, so right now what it does, it’s it can go on Facebook Marketplace. So for example, if a tenant gives us a move on notice, right, it has a its own workflow, it knows what day of the month to list and at what time, because we have another AI that does a lot of Facebook Marketplace subscribing. So you have to catch the algorithm at what time to post, what picture to post, what is the description, and it changes constantly. Because if you can post a listing and you don’t get anything for a month and you can think the market is soft.
It’s not that, it’s just suck at that part of the marketing. So I spent a lot of fun doing that manually before the AI. Like literally opening Facebook Marketplace. Okay, this listing is day two, it’s still there. Why? Right? Writing it down. Okay, it’s day seven. That listing is not there, but I’m seeing a new listing here. Why? Okay, it has the similar keyword on the description. Once I fade that data to my AI, it starts doing so. Every four hours we get a report. So by the time it’s time to post my listing, within a week we get fifteen hundred clicks, two hundred clicks on a listing. So we get multiple inquiries.
Dylan Silver (15:48)
Now I’ve seen this in the past. it’s been a couple months since I’ve looked at this. I saw that it was somewhat limited in its capacity, it was either OpenAI or clawed to interact with Facebook Marketplace. Have you encountered some of that?
Andres Martinez (16:01)
It was. Yes. So I don’t use Chrome, I don’t use Safari. I had to build my own software to not be limited by their blockers.
Dylan Silver (16:08)
Okay, amazing man. So going from literally waiting tables, hundred room portfolio co-living, and effectively you’re an AI engineer. has other folks approached you and asked you about the engineering that you’re doing to interconnect all of these moving pieces to replicate it for their systems?
Andres Martinez (16:25)
Yes, that for example, this is the first time that I’m actually explaining how powerful it is because before like I did in June I spoke at the Co-Living National Conference about AI in my system and people were like, okay, you guys are still talking to the chat bot. I cannot ex every question you get is how? I was like I can explain you a RAG engineering model and if you’re still only using the chatbot, you’re not gonna get it. Even I don’t I don’t know how it was connected.
Right. I I don’t know how it started firing up. That ability, like start connecting with other systems. The only thing that I can get to is like because it has access to everything through loops. Right. And when people approach me, it’s usually everybody thinks that AI is easy to install. It is easy. I can send you a screenshot of my software and your Claude can replicate it. It will look the same. Maybe for one day it’s gonna act the same and then it’s just gonna break. Because the code has to be constantly maintained. And if you don’t understand what’s going on in the background and you don’t teach your AI
What to build, like you have to be the architect. [UNCLEAR NAME/TERM IN SOURCE: “Cody is amazing at building, but it’s not an architect.”] Right? So you need to know what you’re doing. And it takes for me it took a long time period of training it, watching it, what to the what are you doing? What I’m a freak, I’m a control freak. So I was like over jealous. I was like, what are you doing? Why are you doing it? And it’s always the same thing. Why do you do that? And why you didn’t do this? And he tells you like, well I went through this route, I did this, blah
So right now I have two clients I just onboarded last week where we are gonna build a custom system like this for their operations because they also have two hundred plus rooms, they have three VAs and they are still losing text messages and maintenance leads and all of that is going through the cracks.
Dylan Silver (17:48)
So many thoughts that I have. The first one is you mentioned I think it was two or three thousand on tokens. How’d you end up solving that?
Andres Martinez (17:55)
It runs on my subscription now because it’s inside the computer. So it’s a software inside of the inside of the computer. It’s not running in the Claude.
Dylan Silver (18:01)
Wow. Okay. So did you rebuild like your own AI tool effectively?
Andres Martinez (18:05)
Yeah. Which is it sounds like a lot, but now it’s like once I tell you, it’s easy to copy. what That’s why I don’t have videos about it because I’m I suck at marketing and hopefully this podcast can help me. Somebody’s just gonna see it, pay me for a demo, start running ads, and they’re gonna make money out of my knowledge. And that’s something I’m trying to prevent because I know this is I’m a step ahead of everybody on this. I don’t think there’s anything like that yet.
Where the operating system is the actual employee. Right? You can use AppFolio, you can connect AI, you can build custom tools at random, but if you like I have my own property management software for this. Right? My own property management software, my own web browser, and everything is fed just through the brain of Claude. And it because it runs on my computer, like it I can do pretty much whatever.
Dylan Silver (18:42)
Now did you talked about your own web browser, your own property manager. Did you have to organize and work with your AI tool in order to have those built, or did it build those things for itself?
Andres Martinez (18:53)
No, no, I build it. ‘Cause it w it gets to the point like I cannot do this because of this and because of that. And you let it run and it’s like I can do it. it will tell you I’m not allowed to automate Facebook Marketplace because it’s against the terms of service. Against the terms of service where?
Dylan Silver (19:06)
Yeah,
That’s I think what a lot of people, including myself, ran up against. So you had to make your own browser.
Andres Martinez (19:11)
But see, that’s what I try to tell people. Like, the AI is not gonna go against its guidelines. But if you’re creative enough and if you’re the architect, like why don’t you try this? yeah, I can do that. And then it just goes. But you and you need to have a little bit of that input, like, okay, I cannot do this because of those softwares. Let’s just build our own software. And my software is called Casero, the property management software for co-living.
Dylan Silver (19:29)
Great name. when we talk about being an AI architect, you mentioned not getting the certification because you need to be employed by these companies, but that you took the tests. for folks who are trying to gain, some of the knowledge that you’ve gained, would you recommend them doing the same thing or maybe something slightly different?
Andres Martinez (19:46)
I’m all for doing things yourself because I’ve done everything myself. However, I don’t think everybody, most people are not equipped with the discipline, determination of going through the pain. Except as entrepreneurs, right? We love going through the pain just to get to where we want to go. But another thing is I don’t have a job. I quit my job November twenty twenty-fourth because I was waiting tables and at that point my cash flow was equal. So I was like, why am I doing this? And since day one, I was already I was already reading the systems, right?
Whiteboards everywhere I think it’s like early 2025 I started buying my properties but by then my AI was already a property manager. I still had to click everything yes or no approve and why but it was already starting to build its own data so now it’s not trained and this is another problem people are gonna run into running simulations and synthetic data
That is not the same as the real data. And when I explain that to people, because everybody wants to run into this and I get up, people like, dude, just feed a synthetic data, we start selling this and we’re gonna make [UNCLEAR: “social money’s”] like that’s not gonna work. That’s why you get so much AI slop right? it’s not there yet. But if we keep going through these real cases, and for example, I gave my software Casero away for free to 14 beta testers. I was eating all those tokens. It was like six, eight hundred dollars a month on tokens that they were using, but I got their data.
Where are the problems in the systems? What can we improve? So I start hosting a lot of weekly meetings as well, talking to a group of investors that’s trying to help them, but also trying to know what is the problem, where are the bottlenecks that because I don’t have a lot of bottlenecks right now, right? But they do. So how do we implement that into Kim to make to make it stronger? And so as far as learning it, yes, but because I didn’t have a job, I’ve been doing this for ten, twelve hours every single day until now. But
The models there were a lot slower. You can probably do less now. You can probably let get a lot quicker, right? But it’s still gonna be a pain in the back.
Dylan Silver (22:01)
For folks who are looking at this and like you mentioned, maybe that they don’t have the engineering mind that you clearly have, or they may have, a limited bandwidth time-wise or very intensive job. Where can they start to automate their process? is it something like, hey, get the highest subscription chat GPT model? Like walk us through this.
Andres Martinez (22:21)
Okay, so first suggestion that I give everybody, forget about regular Claude and about Claude cowork, which by the way now it doesn’t even exist, it’s just inside one Claude. And I think the next step is gonna be they’re gonna delete all of that, it’s only gonna be Claude Code. I’ve been saying this since Claude Code came out. Only use Claude Code. Because Claude Code can build things. And if it doesn’t have the tools, it will build it. Claude will just tell you, I cannot do because of this and because of that. my usage, if we go to my Claude account, ninety nine percent is on Claude Code.
Every time it hits a roadblock I’m just build it because it can create any software that is available in the internet, right? And if it doesn’t exist yet and you tell it what you want, it will build it, which is insane. Like it would cost thousands of dollars two, three years ago to build a software. Now you can build everything. So we’re also moving away from the subscription based models into the custom software industry. Because nobody wants to pay three hundred dollars a month. I would rather pay the twenty five thousand up front and have something that you can learn
Upon itself, which is the next step. And what I’m beating most people is in the self learning loop. So start using Claude Code, learn what a repo is on GitHub so you can have access to all your products on the Claude at all times. So I can go to the gym and I’m still working. As long as my computer is send me a screenshot. What do you do? So I’m all day I literally I’m in the sound now at the gym. I’m on my phone, still developing. Right. So do that and
Property management, it’s a lot more difficult to automate, especially for co-living, because we have a lot of human input, right? Like, hey, this guy left the dairy dishes. Okay, why? Right? Who else was in the kitchen? Let me look at the cameras. There are other processes that are simpler. For example, acquisitions are a lot simpler because it’s just math. Right? Give it a list. I need you to email these people at that at that price with that agent. And if they reply, I need you to negotiate to this, and that. So it can do it. It can run that multiple times and you can find
Properties quicker. what else is easier to automate right now? Replying to messages, emails, like all of that is fairly simple nowadays. You just have to go through a little bit of the pain to install it.
Dylan Silver (24:11)
Now when we look at Claude versus ChatGPT, you’re obviously big on Claude. Right now I from the outside looking in, I felt like Claude was ahead for a while with everything that they were doing with Claude Code and being able to integrate it into the terminal and then being able to build projects. Then I felt like with ChatGPT
Code or work, I forget the name of it, almost evened it out and some people feel like maybe pushing ahead, you’re still a Claude guy, why?
Andres Martinez (24:39)
Stick to one stick to your first girlfriend man. You find a winner, marry her, right? all these models, the difference in how good each one is the other, we don’t feel it. Right? We’re talking about Fable versus their other product. I think it’s Astra 6 right now. we don’t need that. I still use Sonnet for most of my stuff. Except Opus is like tenant conflicts, right? But and for people who don’t know, Claude has multiple levels of intelligence. It’s like haiku, sonnet, opus, and fable, right?
Same for ChatGPT, but I don’t think we feel the difference. ChatGPT is going more into the consumer usage, especially now with dots and with I think it’s called sessions, because that’s where we’re heading. You need to create a digital copy of where your knowledge is. So a lot of people have Google Docs, have PDFs of processes that all needs to be in one place where the AI can access it. And that is what starts unlocking like
Hey, I didn’t tell you to do that. And the A comes back like, by the way, I found this PDF from two years ago where you talk about this, so implemented that. I hope that’s okay. Okay. So for that, I like Notion. Notion is free. build your database in Notion. You don’t even need to learn how to use Notion now. I spent so much time this year, dude, learning going to Notion, building the tables, and then they came out with a plugin and you just connect the Claude Code and they it builds the entire thing for you. All the relationships and things. So
Dylan Silver (25:51)
My understanding about a lot of this is Claude built something with the MCP, right? Where it can connect to different tools online that then all the other AI tools are able to effectively piggyback off of. But they pioneered this. And to your point, if you encountered a bottleneck, you could
Effectively say, well design a solution to it. And that’s incredible. It’s like magic because you talked about paying a software engineer thousands of dollars. It was thousands of dollars and would take weeks or months. And then you would have to iterate over it and you’d find bugs that would be hard to catch. And little things like looking for an error and resolving an error could be very, very time consuming. And here it’s done in fifteen minutes.
Andres Martinez (26:34)
Yeah, the MCP is a great example of what Claude pioneered. The MCP allows you to read information out of other software without affecting that software’s integrity. For example, I have an MCP for my coaching course. So my tenants don’t have to go through the 15 hour video that I we have because when it’s three AM and you’re already underwriting a deal, it’s like can I need to know why this floor plan doesn’t work? You just put it in the MCP of Colibrian log and it tells you the answer, right? Because it has all the reasoning, it has
All my podcast recordings, every masterclass that I do, everything is recorded through the transcript. So the AI knows what to do. So you don’t have to wait for next day to meet. That’s an MCP So it allows you to connect to multiple places, but there are softwares that don’t have MCPs. Property management software for us, right? Let’s talk about Innago, RentRedi, something basic for the common land or we don’t have to go to AppFolio, all those big guys. They don’t have MCPs, so you need to create one to be able to connect to those. But those are private companies, so they don’t give you access to it. So how do we compensate for that? Right? That’s when you build a product in the middle.
And it was the same thing with Facebook Marketplace. How do we build a product in the middle? Right? Which by the way, I am in violation of older terms, so that’s why I cannot really say what it is, ’cause if I lose my Facebook account then I’m fucked I don’t know.
Dylan Silver (27:36)
I’m sure you’ve experienced this on some level. I have and many others have. It feels like when you’re promoting basically any type of sales oriented thing routinely enough on Facebook Marketplace that your account just inevitably gets banned or shut down. Like if you’re looking for, renters, I’m an apartment locator, if I’m looking for renters or if I’m a wholesaler, looking for end buyers. If I have enough
Listings on Facebook Marketplace and I don’t have paid ads going out and this is all just free exposure for me, like inevitably it gets shut down.
Andres Martinez (28:05)
Yes. So for example, right now Facebook terms, they don’t allow you to rentals as a business anymore. So you need to run ads. You cannot have a free rental website because they were getting too many fair housing violations and too many lawsuits. Even when you’re talking on the conversation with the renter, you see the pop up in the conversation, like this conversation is being reviewed, right? If there’s any safety violation, we’ll let
So they are reading all of our messages. So the conversation has to be a lot smarter because they don’t tell you that you’re banned, you just get shadow banned. You’re you look for your listing and nobody can find it. And you’re like, why am I getting no clicks? And that’s the everybody jumps into that because they just lower your distribution. So how do you bounce back from that? How do you prevent it from getting there? Because it’s really hard to bounce back now.
Dylan Silver (28:45)
This is interesting because I didn’t think about it like that, the idea of being shadow banned, because someone might be like, How am I being shadow banned? I’m a small account, so on and so forth, but they’re doing this algorithmically. I’ve seen recently, and I’m wondering if you’ve seen this too, a pivot to LinkedIn. I’ve felt this, that it seems like more people are promoting themselves on LinkedIn, even if they are traditionally like social media heavy content.
And so instead of scrolling through reels on Facebook, they may still be there, they may still be on Instagram, but now I’m seeing a lot of those people now curate an audience on LinkedIn. Are you seeing something similar or different?
Andres Martinez (29:23)
It’s funny, I’ve never used LinkedIn, but this morning I woke up and I was like, I need to so the problem is, for example, for me, last week I had a lot of introductory calls with these small companies. They cannot afford a be a software this powerful, right? They wanna pay fifty dollars a month of subscription. That’s never gonna happen. That doesn’t even cover the back end of the subscription, right? They don’t understand how expensive AI is outside of a subscription. So then you come back to like, we need to run under your computer, but your computer needs to be on twenty-four seven. no, that’s too much electricity. All right, dude, you’re not a business owner.
So I was like, how do I contact these other business owners who are definitely in need of this and they can afford it? So I was like, How do I get into LinkedIn? Because all the all the big players are there and everybody’s aiming for a solution.
Dylan Silver (30:01)
Yeah, LinkedIn is incredible. their pricing tiers alone, I want to say the basic tiers like seventy dollars a month, if I’m not mistaken. So that alone puts it over, just to make connections, right? And then it goes up from there. And there’s annual plans, of course. But when we look at and I want to get a little bit granular, you talk about running this 24 hours one machine. It’s a proprietary AI tool.
Immediately thinking of the Apple Mini, where it’s no screen, no keyboard, no mouse, it’s a box. And I remember going into the Apple store and being like, What is this? Right? is that the type of thing we’re talking about here where people could run like Casero Casero, right?
Andres Martinez (30:36)
Yeah, Casero was the early software, but I killed that project. Right now it’s called Kim because it’s only custom because the problem with subscriptions is like you have to make it plain enough for everybody to be able to use it. So it cannot be super custom. Right? For example, on my co-living properties, we provide cleaning, we provide landscaping for everybody. So I had two beta testers. man, we have people who rotate cleaning weekly. How do we remind them of that? And we have another house where we do the landscaping every two weeks. So the tenants are in charge of that and also the gas. To my engineering brain, that’s like
How would you do that? I just pay the fifty dollars for somebody to come do that, right? You’re going through all this pain to have multiple people. And I build all this automations tool, and I was like, this is not efficient. So I just take it off. So I kind of killed that project. But going back to that Mac mini, yes, my early model was called Luna. She lived on my laptop. So you don’t need a Mac mini. You can be any computer. And right now Kim lives in my desktop, like where we are talking. Because I’m not in the computer.
Right. I can control it from my phone and she’s just running her right now she’s just running her things on the background.
Dylan Silver (31:29)
Whether it’s Kim or another one of these AI tools, will I not say right now, but maybe in the next year or even six months, be doing like commercial property management? Could we see, a fifty million dollar apartment complex and D F W be run by a AI tool?
Andres Martinez (31:47)
It is a lot easier to automate a property management commercial property than this. Because you are doing just maintenance where you have a guy in a staff, you’re doing cleaning where you have somebody in staff versus my team is like talking to my handyman number one, handyman number two, handyman number three. Right? Handyman one doesn’t respond in three hours, she goes to handyman two. Now handyman is coming to the property, but handyman one just replied, I’m on my way. She needs to cancel handyman one already because handyman two already said yes. Right. And if handyman
Two doesn’t show up, she needs to go back to Handyman One. It says, Hey, can you still make it today? So all these things that are more custom to our because we have multiple properties, right? A commercial building, they have a staff there. commercial buildings just should have this. There are some tools already for like AppFolio that I that I try to copy. The problem is the same thing because it is a software, it’s like a blanket slate, right? You can only automate so it only takes information, it doesn’t act.
And I think all these people should invest on building their own. You don’t need that. You can build your own custom property management software, right? Every time you run a background check, instead of them getting the money, you get the money. Anytime the tenant makes a payment, that fee doesn’t go to them, that goes to you. All these ups and downs that you can have now, the accounting can be done automatically. Like AI is amazing for that. Right. You don’t need to do QuickBooks. All of these and like five, six months ago, you would hear people talking, man, the back end. You can be high, you can be protected. Try to hack something that was built with Fable right now.
Good luck. Right. Those are the frontier models like it’s getting to that point. Like I don’t understand it. I don’t know how to code. I just know that it works. Right. So it’s definitely something these big complexes should look into and into getting a custom build.
Dylan Silver (33:13)
Again, outside perspective looking in, and for my first time hearing a lot of this, I think there’s some level of disbelief that this is available to them and that they can trust it at all. but maybe while keeping a human in the loop in a large capacity, you could reduce staff and have fewer people at physically, at the leasing office, let’s say for instance.
Andres Martinez (33:33)
This is for example something very interesting that I found out. A lot of the times when I was doing this manually, I don’t want to read another message, do it at six PM on a Tuesday. Because I already laid fifty-four that day. Plus my coaching clients, plus investors. And it was something simple, like can we get another bag of trash? I just I don’t wanna open that message because it could be an emergency or it could be something simple. And I have to get a gamble. I’d rather do it tomorrow. So and that’s happening the same with the staffing.
Either what is a VA or it’s somebody in the apartment complex, the human burnout to deal with other humans, it’s insane. Right? If you have worked any type of customer service job, dealing with other humans is just annoying. Where the AI takes care of all that small friction. So when you need when they for example, right now my tenants, when they need me, I’m a hundred percent available. Because I’m not tired. Hey, what happened? Right? I’m available at 247 now if there’s an emergency. And it will be the same for them. Because the AA is going to take care of the small things.
And that way when they come to a human, you can deal with everything that needs a human sensitivity.
Dylan Silver (34:26)
It’s that’s encapsulates it perfectly. there’s this fatigue with humans to humans and then there becomes this decision fatigue. I don’t want to look at my text till tomorrow because I don’t know if this is something that needs me right now, or if this is something simple. And when we have the ability right now, you’re doing it, to have an AI property manager at the commercial level too, which sounds
Astounding, but then you brought up a good point. It’s actually easier because you have these people in-house. Do you think we’ll see one of these here soon, or do you think this is still because of maybe liability or simply the size of these commercial properties and the legacy influences involved? Do you feel like we’re away from that?
Andres Martinez (35:07)
It’s gonna take somebody very brave because it’s really hard to walk away from systems that work. And as long as they’re making their margin profit, I don’t think they’re gonna switch yet. Because they’re like, Okay, we’re making our target market profit. Why would we take this risk? Because it is a risk. Every time you introduce a new tool, it’s always a risk, right? but what we are seeing the tenant satisfaction, which to me was very important because there’s nobody else that can talk besides me. Now the team is a shield.
Only the important stuff comes to me. Everything that is routinely, like we need more paper towels, we need more toilet paper, the bathroom is there. Hey, when is the cleaner coming? All these scheduling tasks the AI can just do, right? Just the scheduling the vendors on the on the property management complex can be done quickly because hey, I complained about the AC yesterday and they keep calling the front desk. And you I’m sure somebody in the front desk on property management. Dude, that dude on 26 keeps calling me about the fucking AC.
When are they gonna go? Versus they could just be talking with the AI and the AI could be talking to the to the vendor. When are you coming? Have you got that?
Dylan Silver (36:01)
You talk about tenant satisfaction. I think a lot of people maybe surprises you that because they’re thinking, well, people want to talk with a human and now you’ve got AI in the loop, but you’ve had the opposite experience. It’s increased tenant satisfaction. Now, right now, can folks use AI voice or is it still better for text? What’s your perspective from a tenant satisfaction?
Andres Martinez (36:20)
You I don’t have voice on mine because it [UNCLEAR: “barns more talking”] and most people communicate through text message. Uh-huh. Because dude all my most of my messages, like sixty-five percent right now, they come between Friday at eight PM and Sunday at four PM. Cause people are having double jobs Monday to Friday. So an apartment complex, you get all that workload on Monday or Tuesday and people complain and hey, I couldn’t complain about the AC that was leaking, and now there’s a water leak on this unit. All that got handled through the AI during the weekend. This lockout, it wasn’t Saturday at one PM when I was at a soccer game.
Right. all these things I could have a human doing that, right? But it’s still there is burning out and my point was like how much can I automate bis before hiring somebody? Because training a human on this is also a lot of work. And it at the with the money that I make, it would have to be a VA it wouldn’t be it I wouldn’t be able to afford somebody here full time yet. Right. So the tenant satisfaction aspect, it was something I did not expect. I thought there was gonna be a lot of pushback. Pushback. And yeah.
When I talk with these people, it’s like, no man, like I complain about this at eleven PM and next morning the Amazon package was in the on the door. Cause it was at about trash bags, right? I asked for more supplies and she sent them right away. I asked about my lease, I got the answer right away. And when there is a problem, they still escalate with me and you respond faster. Because before I was like, dude, it’s gonna be on Thursday, right? I’m busy the next three days. And it was something simple, like, can you talk with room foreman? Like every time he goes to the bathroom, like he forgets to, like spray the shower down.
It was a five minute conversation, but because I’m scared of that like decision, Burnett, right, that this could be a big problem, that now it’s gonna take me four or five days, I don’t wanna take that call right away. And it’s not their fault that I’m busy.
Dylan Silver (37:46)
Amazing. There’s so many components here. And that decision burnout thing is huge. And I didn’t think about this from the perspective of a property manager. I would maybe think about this if I’m looking at like a customer service line or dealing with, numbers on a spreadsheet and I’m underwriting certain properties and maybe I underwrite worse later in the day because I’m tired.
Andres Martinez (38:06)
One small example. There’s some McDonald’s here two blocks from my house. I hated going there. It’s the worst customer service, went for a walk with my wife like two weeks ago. And it’s like, you want to get coffee? It was like 6 30 in the morning. McDonald’s [UNCLEAR: “still ropping”]. Sure, let’s get some coffee. There’s no cashier. But there’s a person standing there giving the orders, smiling the entire time. The experience before would have been talking to a cashier that is burnt out because 6 30 in the morning, I have to take your order, versus you can do it to the AI kiosk, and this girl just hey.
Here you go. Do you anything else? Do you need sugar? I’ve never been asked if I want sugar on a McDonald’s. Cause they’re just like, here’s your food. I think that’s where we’re coming. Like all this human burnout that is just repetitive work. It’s gonna be taken off and a human is gonna be able to treat another human better because we are not burned out from like human contact.
Dylan Silver (38:47)
Amen to that. we are coming up on time here, Andres. Any new projects or activities that you’re working on? Also anything you’d like to mention directly to our audience?
Andres Martinez (38:57)
Thank you guys for listening. another tip that I want to give you on AI, there’s a lot of talk about right now building the AI team and getting multiple agents. I don’t think that’s the right to right way to go at it. Just buy the bullet and build the AI operating system, whether you want to build yourself or pay somebody else, because that’s what gives the AI power to control everything. Because an agent is gonna lose context versus another agent. You have one agent doing marketing and another agent doing the underwriting.
They’re gonna lose the context on the tone of voice because they have to reply to the to the top. it’s like a different loop versus if everything is circular, they can all talk to each other. I’m only able to do it with the operating system architecture. and then co-living is it’s only growing. Don’t be scared of the city. I’ve had the code enforcement try to shut me three times, I’ve won three times, because they’re wrong. The government shouldn’t be able to tell you if you can afford a property to leave or you’re homeless.
And we are not housing people that cannot pass a background check. We’re not doing group homes. We do background checks on everybody, no violent felons, no evictions. What up? I’m getting divorced, man. I need a place for six months. I need still need to be close to my kid. I need to be close to downtown, but I cannot afford two thousand dollars more on rent. Perfect. Here it is. if you’re buying, make sure you have parking. In Dallas-Fort Worth, everybody needs parking, everybody has cars. So make sure you’re buying corner lots or cars where you can put in the back. And if you want to learn this very well, my school group is called Coliving Unlocked.
You can also find me on Facebook, Andrés Martinez REI. ’cause I give all of this that I’m talking about with AI systems, it’s given to you with the MCP. So you just copy and paste and you can run the business from day one.
Dylan Silver (40:21)
Andres thank you so much for joining us today. Thanks for your time.