Neal Bawa details using AI agents to automate real estate underwriting and reduce headcount by 40%
Summary
Neal Bawa, CEO of Grocapitus, discusses integrating AI into his $400 million real estate portfolio, claiming to have replaced 40% of his workforce while tripling productivity. He describes building over 100 AI applications, including an automated underwriting system that screens deals, performs financial analysis, and generates presentation decks. Bawa emphasizes the use of sub-agents and cross-AI auditing to mitigate hallucinations and ensure accuracy. He contrasts the low token costs of these systems against traditional human salaries, highlighting significant efficiency gains in acquisitions and asset management.
Key takeaways
- 1 is me creating an entire employee. I instead of paying $120,000 a year, I'm paying roughly $200 a a year, not a month, a year, uh, for the tokens and the work that I've done. Plus, of course, my work, right? So there's uh obviously work that I am doing up front, which was probably twenty. 20 hours of work, to get the app, the workflow, the code all completely built.
Unidentified speaker (speaker_2) · 44:06
- 2 AI is not good at checking its own work due to assumption bias; using a different AI model (e.g., ChatGPT auditing Claude) is necessary for effective validation.
Unidentified speaker (speaker_2) · 44:46
Episode sections
- 0:05 – 5:07
Introduction and Guest Background
Host Alex Perny introduces guest Neal Bawa, CEO of Grocapitus, highlighting the evolution of AI in alternative investments. Bawa outlines his background as a technologist turned real estate investor, noting his transition from selling a tech company in 2013 to managing a $400 million real estate portfolio and acquiring private companies.
- 5:07 – 15:30
AI Obsession and Operational Integration
Bawa describes his deep integration of AI into his daily workflow, stating he spends 80% of his day working with AI tools like Claude. He claims to have built over 100 applications across various departments, reducing his human workforce by 40% while tripling productivity through AI 'employees'.
- 15:30 – 27:13
Building an Alternative Investing App
Bawa demonstrates how to build a custom app using AI to rank alternative investments and their operators. He instructs the AI to research top methodologies, rank them with pros and cons, and identify top vendors, emphasizing the need for high-tier AI subscriptions to handle the token usage required for such comprehensive tasks.
- 27:13 – 32:28
We Are All Programmers
Bawa argues that AI empowers non-technical users to become programmers by turning ideas into functional applications. He contrasts building static apps with his goal of creating agentic AI employees that can make decisions and execute tasks autonomously, aiming for 80-100 hours of weekly productivity from an 8-hour workday.
- 32:28 – 34:36
Automated Underwriting Workflow
Speaker 2 describes replacing a $120,000/year human underwriter with an AI system that screens broker emails against specific buy-box criteria. The AI pulls financial documents, runs initial underwriting, and generates presentation decks, handling the majority of the analytical workload.
- 34:36 – 39:30
AI Expert Training via Sub-Agents
The speaker explains a method to train an AI (Claude) to become an expert by instructing it to use sub-agents to research the web, read books, and build a deep, indexed knowledge base on a specific topic like multifamily syndication underwriting.
- 39:30 – 40:44
Document Retrieval and Basic Underwriting
Once a deal matches the buy box, the AI accesses the broker's website, signs up, and downloads the T12, PL statement, and rent roll. It then performs a basic underwrite in 60 seconds to determine if the deal meets profit parameters.
- 40:44 – 42:10
Full Underwriting and Presentation Generation
For qualifying deals, the AI conducts a full underwrite by emailing brokers for missing data, then generates a 32-slide PowerPoint deck with charts and rankings. It schedules a 'Shark Tank' meeting to present the deal to the investor.
- 42:10 – 44:06
Post-Meeting Audit Loop
After a Zoom meeting, the AI grabs the transcript, answers investor questions, and updates a dashboard. It iterates on the offer and presentation based on feedback, eventually generating a purchase offer in 60 seconds.
- 44:06 – 44:46
Cost and Time Comparison
The speaker states the AI system costs approximately $200/year in tokens compared to a $120,000/year human salary. He estimates it took 20 hours of upfront work to build the workflow, which is written in Python.
- 44:46 – 48:03
Cross-AI Auditing Strategy
To prevent hallucinations, the speaker uses a second AI (ChatGPT/Codex) to audit the work of the first AI (Claude). He notes that AI is poor at checking its own assumptions and requires a different model to perform a comprehensive audit.
- 48:03 – 57:48
Multi-Methodology Auditing
The speaker instructs the auditing AI to attack the underwriting from three dissimilar methodologies simultaneously. This forces the AI to create separate scenarios and verify the deal's robustness across different analytical frameworks.
Assets discussed
| Asset | Kind | Stance | Context |
|---|---|---|---|
| BitcoinBTC | crypto | neutral | Obviously, there's real estate, there's Bitcoin, there's uh metals, you know, gold, silver, etc., and others like it, right? |
| Gold | commodity | neutral | Obviously, there's real estate, there's Bitcoin, there's uh metals, you know, gold, silver, etc., and others like it, right? |
| Silver | commodity | neutral | Obviously, there's real estate, there's Bitcoin, there's uh metals, you know, gold, silver, etc., and others like it, right? |
“Stance” is how the speakers talked about the asset, not a recommendation.
Claims worth verifying
Bawa has a $400 million current real estate portfolio and has bought/built about $1 billion of real estate since 2013.
Unidentified speaker (speaker_2) ·4:28
How to verify: Check SEC filings for any registered investment advisers associated with Neal Bawa or Grocapitus, or public real estate transaction records if the assets are publicly traded or listed on MLS databases. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Primary sourceBawa's company has had 1,200 accredited investors, with 40-50 investing through Advanta IRA.
Unidentified speaker (speaker_2) ·4:28
How to verify: Review Advanta IRA's public case studies or press releases from 2026-08-18 to 2026-11-18 mentioning Neal Bawa or Grocapitus as a guest or partner. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Bawa spends $8,000 per year on AI tokens, which he equates to 25 employees.
Unidentified speaker (speaker_2) ·11:59
How to verify: This is a personal financial claim; verify via Bawa's public blog posts, LinkedIn updates, or interviews between 2026-08-18 and 2026-11-18 where he discusses his AI operational costs. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Bawa reduced his human workforce by 40% while tripling productivity.
Unidentified speaker (speaker_2) ·11:28
How to verify: Check Grocapitus's employee count on LinkedIn or Glassdoor over the last 12 months (2026-2026) and compare with public announcements of productivity metrics or deal volume. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
OpenClaw is an open-source agentic AI tool that came out in Q4 of last year.
Unidentified speaker (speaker_2) ·29:47
How to verify: Search GitHub or AI research databases for 'OpenClaw' and check its release date and description for agentic capabilities. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Bawa built over 100 applications in the last two years, with 25 for personal use.
Unidentified speaker (speaker_2) ·9:18
How to verify: Verify via Bawa's public portfolio, GitHub repositories, or interviews where he lists specific applications built between 2026 and 2026. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Bawa's AI spend is $650 a month.
Unidentified speaker (speaker_2) ·11:59
How to verify: Cross-reference with the $8,000 annual figure stated in the same segment; calculate $8,000 / 12 months. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Bawa works 80% with Claude and 20% with employees.
Unidentified speaker (speaker_2) ·5:26
How to verify: This is a subjective estimate of time allocation; verify via Bawa's public statements on his daily routine in interviews or blog posts from 2026-08-18 to 2026-11-18. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Bawa's AI employees are named after AI models (e.g., Claude).
Unidentified speaker (speaker_2) ·11:28
How to verify: Check screenshots or descriptions of Bawa's internal tools or public demos where AI agents are introduced with names. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Bawa claims AI is worth $5,000 per week to users who pay $20/month.
Unidentified speaker (speaker_2) ·8:38
How to verify: This is a subjective valuation claim; verify if Bawa has published this specific calculation in articles or videos between 2026-08-18 and 2026-11-18. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Neal Bawa's company no longer has human underwriters; the process is fully automated.
Unidentified speaker (speaker_2) ·33:41
How to verify: Check Grocapitus career pages or employee counts on LinkedIn; verify via company filings if public, or contact company directly. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Primary sourceThe AI underwriter costs approximately $200 per year in tokens, compared to a $120,000 annual salary for a human underwriter.
Unidentified speaker (UNKNOWN) ·44:06
How to verify: Calculate token usage costs based on API pricing for Claude and ChatGPT; compare to industry average salaries for real estate underwriters. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
The AI rejected an underwriting 14 times before accepting it on the 15th try.
Unidentified speaker (speaker_2) ·45:59
How to verify: This is an anecdotal claim; verify by asking Neal Bawa or checking if he has published logs/cases of this specific event. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
The basic underwrite takes 60 seconds, while the full underwrite takes hours.
Unidentified speaker (UNKNOWN) ·40:24
How to verify: Benchmark against industry standards for manual underwriting times; verify via Neal Bawa's published case studies or demos. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
The AI generates a 32-slide PowerPoint deck for each deal.
Unidentified speaker (speaker_2) ·41:24
How to verify: Review demo videos on Multifamily University or request a sample deck from Neal Bawa. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
It took 20 hours to build the AI workflow.
Unidentified speaker (speaker_2) ·44:26
How to verify: Anecdotal; verify by asking Neal Bawa for the breakdown of hours spent on coding vs. prompting. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
The AI uses sub-agents to read every article, podcast, and book on the topic to build a knowledge base.
Unidentified speaker (UNKNOWN) ·34:52
How to verify: Technical verification by reviewing the code or prompt structure shared by Neal Bawa; check if the AI cites sources from its training data. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Multifamily University offers free AI demo videos.
Unidentified speaker (speaker_3) ·53:33
How to verify: Visit multifamilyuniversity.com and check for free video content. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
The AI signs up on behalf of the user on broker websites to download documents.
Unidentified speaker (speaker_2) ·39:50
How to verify: Technical verification; check if the AI uses browser automation tools (e.g., Selenium, Playwright) to handle CAPTCHAs and sign-ups. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
The AI sends emails to brokers asking for missing data without copying the investor.
Unidentified speaker (speaker_2) ·41:04
How to verify: Verify by observing the AI's email sending behavior in a demo or by reviewing the code logic. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Do your own research
- Verify Grocapitus's actual headcount changes and productivity metrics against public filings or press releases.
- Audit the claimed $8,000 annual token spend by reviewing actual API usage logs from providers like Anthropic or OpenAI.
- Test the 'OpenClaw' open-source tool for security vulnerabilities and data privacy implications before enterprise deployment.
- Compare the 20-hour build cost and $200/year operational cost against industry benchmarks for custom software development and maintenance.
- Investigate the accuracy of the AI underwriting system by backtesting its deal rejection/acceptance rates against historical real estate transactions.
- Check for conflicts of interest if Bawa recommends specific AI tools or subscriptions that may have sponsorship ties.
- Review the legal liability framework for AI-generated financial underwriting and presentation decks in real estate transactions.
- Assess the reliability of cross-AI auditing by testing if different models consistently catch hallucinations in complex financial scenarios.
Full transcriptexpand
0:05Unidentified speaker (speaker_0): Warning. Exposure to the Alternative Investing Advantage Podcast may result in increased financial confidence, independent thinking, and a desire to invest beyond stocks and mutual funds. Subscribe if you're serious about taking control of your financial future.
0:20Unidentified speaker (speaker_1): Alrighty, everyone. Thank you so much for joining with us today on the Alternative Investing Advantage Podcast. Today we're very pleased to welcome all on Neil Bala with Grow Capitus. And we've uh we've we've known Neil for quite a while here at Advanta IRA. And uh, you know, over the years we've we've done some content with him. And one of the great things about the people that we get to know in our line of work is kind of how things can evolve, right? And especially in today's climate of investments and technology, the advent and the continued continue evolution of AI is something very exciting and maybe some degree of trepidation to watch, depending on kind of what side of the coin you're on. But to completely ignore it is to do yourself a disservice, right? It's kind of one of those things that's here, the genie's out of the bottle, and we can either adapt and understand and grow with it, or you know, for lack of a better way to put it, be left behind. So, you know, the conversations centering around AI are as varied as the day is long. And we're not really here to pass judgment on it today. We're going to be looking at how some of
1:20these tools can be used when it comes to investment decisions specifically related to alternative investments, right? So, you know, if you've been a longtime listener of this show, that can mean everything from, you know, real estate and all of its beautiful forms and fashions to cryptocurrency to uh dead instruments to precious metals. Again, the the spectrum of things that fall under the uh alternative asset umbrella are again as varied as the day is long as well. So utilizing these tools in this arena is something that hasn't been done a ton up to this point. Again, certainly in all aspects of finance, there is the use and the application and the integration of AI, but we've seen it a lot in the more traditional security space. But looking at it through the lens of decision. making in alternative assets is something I'm really excited to speak with you today about, Neil. If you wouldn't mind, give us a little bit of background about yourself. I know we've uh had you on a few other uh avenues here, but uh go ahead and give yourself an introduction.
2:14Unidentified speaker (speaker_2): Sure. Um I am a recovering technologist, so computer science degree. I've had a successful tech career. I live in Silicon Valley. Um and um um, you know, ran a tech and education business for 14 years. uh hundreds of employees, so not a startup. Um, and uh sold it in 2013 because I live in Taxifornia and I had the big fat tech salary. I was paying, you know, 40 to 50 percent of my income to the man, and I got into real estate for its incredible tax advantages. Did that for about 10 years, just myself and my family buying smaller single family or fourplex type buildings, and in 2013, when I sold my company at a very large taxable event, I decided to go full hog and buy large multifamily, which turned out to be really easy to do. I needed investors because I ran out of my own cash. But it starting 2009, I had basically opened a meetup group inside of my company. We had large conference rooms, so and I didn't want to fight barrier traffic to go to another meetup. So I invited uh other people to attend my meetup. which was about the data science of real estate. Basically, how do you use data and science and scientific methods to to invest in real estate? And I that I was very passionate about that topic. We started in 2009, and by 2013, I had 5,000 members because I'm in Silicon Valley. There's a whole bunch of nerds here,
3:33right? You throw a rock here, you're gonna hit a nerd, it's gonna bounce off and hit another nerd. And so I mean there's there's a lot of us. And so it was it was a huge group of people, and we weren't allowed to pitch anything or things like that for four years. We were just basically trying to rank cities and figure out what matters the most when it comes to real estate profits. You know, is it job growth, is it income growth, is it home prices? What really makes the biggest difference to um to real estate profits? Um, and that was a passion we would meet once a month, and um and that you know. in 2013, when I bought a very large multifamily, realized that I'd run out of my own money, I sent an email to that list. And immediately a huge number of those people decided to invest with me because they'd known me for about four or five years. And it started a new career for me in 2013. And since then, we have bought or built about, you know, um a billion dollars of real estate. The current portfolio is about 400 million. And uh we've had 1,200 accredited investors, of which about I think 50 are from Advantage, roughs roughly 40 or 50, I think, have invested through Advantage, not necessarily at this moment, but over the last decade. Um and um so a lot of IRA investments, not a lot of cash investments into real estate.
4:47Eventually we started buying companies as well because we realized that. real estate, the cash flow was decreasing as real estate got more and more expensive. So we started buying uh what we what we call boring companies, water infrastructure companies, CPA companies, things like that. And then about two years ago, I fell into this full-on obsession with AI. So I I'll just stay say state this to start this podcast, Alex. I have an unhealthy level of obsession with AI. I start my day at six o'clock in the morning by checking 30 plus Claude strings that are running overnight. And there's probably eight of them active at that point of time. I give it instructions. And then throughout the day, I have changed my day entirely so that I work about 80% with Claude and then 20% with my employees and my investors. Now obviously that 80% with Claude definitely involves my investors, my employees, and my properties. But I'm essentially working through one uh software. And you know, Claude is one of the four AI softwares that I use. So that's that's what sort of brings me to this point. I'm I'm a I'm a tech guy that's in real estate, obsessed with AI.
6:00Unidentified speaker (speaker_1): Fantastic. Well,
6:02Unidentified speaker (UNKNOWN): you know, to
6:02Unidentified speaker (speaker_1): kind of look at that, uh, you know, I think that's probably a good place to start as any, is kind of the the implementation and kind of looking at how you're using it. Now, when it comes to again, utilizing AI, especially in the context of what you do, I mean, I'm assuming it's probably fair to kind of bucket it into some silos of the actual application decisions that you're looking to have it do, right? I mean, it's easy enough to, you know, most people say, Hey, Claude, what do I eat for dinner? tonight? Okay, well, what do you want to eat? But when you're kind of applying that as a layer on top of an operational uh organization like this, what are some of the key components that you're using it for in the context of of your real estate investment or your real estate um uh operations, if you will? Are you looking at it for like decision making on day-to-day operations? Are you looking at it from like future to like uh market identification? or you know current risk assessment? Like, what are the things that you're primarily focused on utilizing these tools to help you accomplish with that part?
6:59Unidentified speaker (UNKNOWN): So
7:00Unidentified speaker (speaker_2): I'll I'll start by setting this uh as a as a picture because the scope of this is so enormous that you're gonna struggle with it, your your listeners are going to struggle with it. AI is the greatest change to human history since the invention of fire. Take the smartphone. the personal computer, and the internet, gather them together, multiply by a thousand, you still haven't hit AI. The reason for it is all three of those the smartphone, the personal computer, and the internet, were tools to improve our productivity. They made us more productive, and in the case of the internet, they also made us more knowledgeable, right? AI is pure intelligence. intelligence. And it's not just raw intelligence because we humans get the ability to shape that intelligence into usable intelligence. AI is raw intelligence, and nothing like an intelligence explosion has occurred since the invention of fire. So understand that we are at a point in history that is unlike anything that we've ever seen. No technological innovation before. this has created intelligence. AI is almost unlimited intelligence. I'm going to give you an example of this so that you understand my point of view, and I hope that you think that this is the right point of view. A lot of people buy uh $20 chat GPT accounts or $20 cloud accounts.
8:18They're very common. I can explain to Into you s this. These accounts reset every week, right? And you're thinking, hey, so really if they reset every week and they cost me twenty dollars a month, it's five dollars a week. That's what really Claude's worth. I'm I'm here to tell you that you need to take that five dollars a week number and you need to multiply it by a thousand. Because the intelligence that Claude or ChatGPT or Perplexity or Gemini are giving you per week should at least be worth $5,000 to you. In my case, many more zeros, right? That's what it's worth. And each week, if your limit expires and you don't use it to 100%, you've basically wasted $5,000. I want everyone to start thinking of that. You're being given an incredible amount of intelligence every week. Stop looking at the $5 bill. Start understanding it's worth $5,000. for you, for your health, for your business, for your life, your entertainment, for all of these things, right? I have built over 100 applications in the last two years. 25 of them are just for my life. Really, really cool stuff that I'm doing that has nothing to do with my business, but it really improves my life, right? So I still haven't answered I Alex's question because I'm still really setting up just how big a deal
9:38this is. I use AI in every single aspect of the business. There are 17 applications in my company just for the marketing department. There are 13 applications for the acquisitions department. There are seven for the asset management department. There are five for the compliance department. There are uh four or five for the operations department. right? So you can understand that there are apps, real apps, and these are like apps like the kind you have on your phone, though we mostly use them in the browser. Um, you know, because you know, when you're working, you're you probably have a PC or Mac in front of you. So you tend to use them more there, though they work fine in a in a on a phone. So we've basically taken every aspect. Oh, I forgot one important one research. We have seven applications just in the research. department. Because before you buy anything or manage anything, you first have to research it. We have nine applications to manage our investors, right? And these are all apps that we built over the last year because and and we asked ourselves one question. And this is actually the same question that Alex just asked. Okay. And and I use this as an example. So Some I'm I I really didn't tell Alex to ask this question. he just asked it by himself. Most people are interested in telling Claude or Chat GPT, hey
10:57Claude, what should I have for dinner? I have no interest in that. I'm interested in telling Claude and Chat GPT, make my dinner. Okay? Now we all know currently Claude and ChatGPT cannot make dinner. I get that. But in a business, they can make my dinner. In a business, they can carry out thousands, and currently our Claude is carrying out thousands or tens of thousands of tasks. So my workforce has decreased by 40% in the last year, while the total number of tasks and productivity has tripled. Why? Because I have over 25 new employees. They all have names, but they're all really clawed. Am I spending $20 a month? Not even close. My current token spend or my AI spend is $8,000. So I'm spending $8,000 a year, so about $650 a month, right? For a very large amount of
11:59AI tokens. And those $8,000 have given me about 25 employees. Another way of saying it is I'm paying about $200,000, $250 for an employee that works 24 hours a day, seven days a week, at an average speed that is 100 to 1,000 times faster than my human employees. It doesn't replace my human employees entirely because that's impossible. Maybe, maybe in 10 years it will. But it adds to them, it makes them more productive. So when I have a $150,000 employee today, I'm able to get $500,000 of work out of them simply by wrapping them in a layer of AI, assistance. And those assistants are going up the ladder in my company just like people do. You know, people start at the bottom and then become, they become a supervisor. and a manager and a director and a and a VP. Those clod assistants are also getting promotions. It is quite likely that within the 12 next 12 months, I will have a Claude employee that manages humans simply because they have a higher level of intelligence and understanding of my uh you know, company. And I realize that a bunch of you are cringing right now, thinking, what the heck is this idiot saying? I mean, how can you know AI manage humans? The answer is it's intelligence.
13:18People with a higher level of intelligence and a higher level of experience typically tend to manage others that have a lower level of experience or intelligence or both. And so that's the world that we live in today. I'm not running one of these AI companies, Company, so I don't have an agenda, right? I'm just a user. I'm just using these things and they are absolutely extraordinary. And I I don't think that I'm capable of understanding the full impact of AI. But I certainly think that 99% of the audience listening out there doesn't. And hopefully I've given you sort of a little bit of a chink of just how powerful AI is. So back to you, Alex. But I don't know long answer, but I really had have to lay it
13:59Unidentified speaker (UNKNOWN): out.
14:00Unidentified speaker (speaker_1): No, for sure. And and I can appreciate the
14:04Unidentified speaker (UNKNOWN): the
14:04Unidentified speaker (speaker_1): you know the utilization aspect of how you're you know implementing some of these things. But specifically kind of what I want to dig into is a little bit more of when it comes to investments, right? You know,
14:19Unidentified speaker (UNKNOWN): kind
14:19Unidentified speaker (speaker_1): of the core focus of the show is talking about investments as it relates to either decision making or performance. or again, just kind of general application of these other types of alternative investments. Again, not things in the stock market, because there is, you know, tends to be a little bit less of a public-facing data set for an LLM to learn from with private investments or alternative investments in general. Um, you know, it's not as aggregated to a central location. So it's this again, picking on real estate, uh, some it's of something the we like to do a lot in this uh industry. When it comes to again, generally speaking, the audience is going to be people that are going to be wanting to place money into real estate, whether as a limited partner or maybe making decisions on single-family homes, debt instruments, again, the scope of kind of being the more retail-facing um investor, if you will. So when it comes to those kind of avatars. of investors, what are some of the ways, at least, that you've seen maybe it's something that you've done personally, or how you're seeing other people implement AI tools into that endeavor of making informed decisions in this space of real estate?
15:30Unidentified speaker (speaker_2): Okay, how about this, Alex? Over the next 25 twenty-five minutes, you and I will go through the entire process that you mentioned. and we will build an app. We will build it live. You it won't, you know, we won't be able to show it to people. We're gonna build an app. And anyone that has the transcript of this podcast, you know, and and do you guys put this on YouTube, right? Or uh Apple or whatever it is, right? So
15:57Unidentified speaker (UNKNOWN): all you guys have
15:57Unidentified speaker (speaker_2): to do is to figure out how to get access to the transcript of this podcast. And you will be able to replicate the app that we are talking about building today, right? Call it a super app. We will call this alternative investing edge. That's the name of the app that we're now building, right? So the first thing is to build an alternative investing app, we need to define what are the alternative investments that we're interested in. Obviously, there's real estate, there's Bitcoin, there's uh metals, you know, gold, silver, etc., and others like it, right? So, first, in your mind, you need to give, and I'm gonna use the word CLARD for AI. Obviously, you do not need to use Cloud, you can use ChatGPT to build this application as well, right? So the first thing we're gonna do is basically we're gonna say to our AI, hey, the app that I'm building is mine. It's for me. I'm not selling it to anybody else, I'm not selling it to Alex. I'm building it for me. So I need to tell you, I need to tell you what I'm interested in. Now, that's one way of doing it. I would urge you not to do it that way. I would urge you to tell AI, you know what? I I want to learn from you. I want research from you. So why don't you tell me in today's environment, in today's markets, what are the top five alternative investing methodologies? I'm pretty sure
17:17uh metals are going to be in there and real estate is going to be in there. But let's just say top five. Now, why not every investment methodology? Because you're going to get overwhelmed. The problem with AI is it gives you way, way too much data. So pick five, right? So if you guys are okay with that methodology, instead of giving it a list, I think it'll be better, right? The next thing is okay, so now it's picked five for us. Great, it gives us a list of five. Then you're going to say, okay, Claude, I want you to rank these. I want to know the pros and cons of all of these. And what I want you to do, Claude, is I want you to use sub-agents. Now you don't need to write this stuff down. Remember, you have the transcript, right? So I am now instructing Claude in the application that I'm building for you, right? Claude, I want you to use subagents, right? And I want you to assign a subagent to each of these verticals. The goal in the app is to make the app extremely comprehensive, but also very, very easy to understand. And Claude, when you build this app, you will use lots of colors, you'll use a dashboarding methodology, you will make it highly extraordinarily visual. Please don't give me a bunch of stuff to read on the screen. I want visuals, charts, graphs, trend lines. I want this to look like McKinsey built this for me. I want it to look like Anderson Consulting built it for me. And I want you to give me a tab at the top for each of those alternative investments. Okay? Now,
18:36I want you to take your time, Claude, to find the data. Don't just go out and find basic stuff. I want deep level research. I realize that my $20 account is going to run out of tokens. I'm okay with that, right? Do your best. And Claude, by the way, I know that there's some free uh places out there that you can use tokens in. So use them as much as you can so you're not burning through my $20 limit. So when you're done, Claude, my application would have tabs for these top five that you come up with. And you've given me the pros and cons of each of these, right? You've You given give me the research. You're telling me why, you know, let's say real estate is ranked. three out of five, and Bitcoin is ranked two, and gold is ranked one. Why do you think that is? And it has to be all English language. Think that I'm five years old, right? Give but and make it highly visual for me. Now that you've ranked it, now I want you to go find the the vendors, like the people that are offering this. So let's I'm I'm gonna going to use real estate as an example because a lot of Adbanta Advanta people are into real estate. right? They understand real estate. So I want you to basically find me the top 25 companies that are offering options to me to invest in this place. And I don't just want you to find the companies, I want you to find places where these companies aggregate. There are places
19:56on the web that, you know, uh aggregation happens where people are aggregating all the Bitcoin vendors. people are aggregating all the gold and silver vendors, people are aggregating all the real estate investors. And Claude, your job is not just going to be to find these vendors. I actually want you to take the next step. I want you to go in there and I want you to log on and sign up to these portals. Now I realize, Claude, that a bunch of these, you're gonna need my help to sign up. It might ask for my cell phone, it might need you to send send me a code. So do your best. Make sure that you do 99% and you make me do the 1% and build this app for me. Right? And and yes, I realize that substance are going to be manual, and you'll come back and say, I can't log into this website. You have to create a password. Make it easy for me. Tell me, give me instructions, and I'll do my 1%, Percent and you can do your 99%. But when you've built the app inside of those tabs, there need to be sub tabs. The first tab is the research tab, right? That tells me why I'm interested in Bitcoin or I'm why I should be interested in real estate and what the pros and cons are. The second tab are the operators. Now you've built me a tab where you're actually going out to these people and you're getting me 25 Five operators. Then that operator tab it needs to be highly visual. Don't just make me a list of operators. That's not useful, Claude.
21:16I want you to basically analyze it. Think that you are a consultant, the world's best consultant for real estate. Right? Gather knowledge from the web and then go in. Once you've gathered the knowledge, go in and say, How do I visually tell Alex what he needs to see about these companies and understand about the different surveys. services that they're offering and the pros and cons and why I've ranked this company above the other ones, right? Neil Bauer's company might be in there and probably won't be at the top. That's okay. Bottom line is now you're ranking all of these companies for me, Claude, right? And showing me a visual on what these companies look like, where you found them, things like that, the good, the bad, the ugly, right? In each of these places. And then as a future option, don't build this yet, Claude. I want you to even give me the ability to click a button and invest into these things, right? So there's got to be some future code that you're going to write, right?
22:17Unidentified speaker (UNKNOWN): And
22:17Unidentified speaker (speaker_2): that's version one of the app part. When we are done, I need to have at least for five of these things, 125 ways in which to invest, all ranked by quality. And then I need to have the right research. Okay, I'm done with spec for version one. Now, as you can guys can imagine, apps that I build are version five, version 10, right? It would probably take me three or four hours just to describe all of them. But in the last six or seven minutes, what I've done is I've given you everything that you need, not just the basics, everything that you need to build this app. There's only one piece of really bad news for you. This application, which I gave you, cannot be built using a $20 Claud account or a $20 Chat GPT account. So if you really want to build the app and customize it for yourself, you're probably gonna have to bump up to the $100 level for one month, build the app, and then bump it back down to the $20 level. Because once the app is built, you don't need more than the 20 bucks. But while it's building, it is going to be very frustrating because it's going to keep running out of the $20 tokens. Remember I spend $8,000 on tokens and I have all kinds of ways in which I I get free tokens. All kinds of ways, right? It still costs me $8,000. But of course I'm building over a hundred apps a year. So I know this was a long process, but all you need
23:37is the transcript here. and a working account. It cannot be a free account. Now, the one thing I didn't tell you is where is Claude going to host this? So I'm going to basically say, hey, Claude, since this is a beginner user, you can host it on Vercel, you know, or equivalents of Vercel, you know, help this user sign up for a free account. Please don't, you know, make it a paid account. There's lots of these free accounts like Vercell Vercel on on the web. or equivalent so that they can go in and start the app. If they start using the app, sure at some point they might have to pay a $10, $20 a month fee. But initially while you're building the app and they're playing with it, make it a free service that's available on the web cloud. There you go. Now your hosting is taken care of. The app is somewhere on the Vercell website or equivalent website. It's an AI vendor that hosts these apps. Uh and they have all kinds of tools that help you out. But you're not going to use those tools because your Claude's going to build the app, right? It's going to do all the work for you. You're just, you know, you need a place to click on your phone or click
24:37on your iPad so that you can actually enjoy this gorgeous app that Claude has just built. The end result of this app, it is going to look absolutely stunning. And you're going gonna to be all bragging at every coffee table event or or dinner about what you built. right? But Claude is going to build it for you. All you needed was this description. You realize now how powerful this is, right? I have just built an app for you, customized to you, that is going to look stunning and going to do the research. The key thing is this was just the research component, right? And this was research for alternative metals. I obviously my research looks different from yours, but that's all that's needed. Now, the truth is you'll probably talk with Claude 50 or 100 times to improve, tweak, add a radio button, checkbox, drop down. Yeah, none of those are required. You're gonna do them because you like to see stuff and you know look at stuff a different way. That's just you tweaking and improving. Even when we buy a brand new car in 2026, we play with you know
25:37the the Apple interface or the Gemini or Google interface on our car and we move the met, you know, buttons. We do little customization. That's what it means. But the description that I gave you is enough to build a gorgeous app for
25:48Unidentified speaker (UNKNOWN): you. Does
25:50Unidentified speaker (speaker_3): that
25:50Unidentified speaker (speaker_2): answer
25:50Unidentified speaker (UNKNOWN): your
25:50Unidentified speaker (speaker_2): question, Alex?
25:51Unidentified speaker (UNKNOWN): No,
25:52Unidentified speaker (speaker_1): it definitely does. And it's it's interesting to kind of look at it from that perspective. A lot of times when And I've talked to people about AI, uh, which in fairness hasn't been a ton on the show as comparative. to some of the conversations that I've had. So again, setting at least the standard for the fact that the conversations haven't been uh numerous, right? But more often than not, um, when talking to people, it's more based on prompting AI to complete a task, more or less kind of bound in that task, right? So prompting it for kind of a decision-making decision making process, you know, one, two, three, four. Now, what you're kind of alluding to, or I guess like mapping out really, is app development, right? So more of kind of a continual use case for something that it can build, right? Which is again the first time I've had this conversation. So to expand more on that, what is what was kind of the the the impetus for kind of going that direction? Because again, AI, the one of the beautiful things about it is. that there's no kind of right way to go about doing it there's more effective ways there's more efficient ways to use it for your end goal but it's very amorphous in its kind of interaction that you have with it right you can have it build you an app you
27:04can have it do graphic design you can have it do a lot of things but what was kind of the the driving factor towards
27:11Unidentified speaker (UNKNOWN): you
27:11Unidentified speaker (speaker_1): know app development for lack of a better way to describe it is it just for again, that specific interface, or is there something that doing going the route of having it develop some type of like
27:23Unidentified speaker (UNKNOWN): like
27:23Unidentified speaker (speaker_1): user interface app that is you find more effective than other methods of using AI?
27:30Unidentified speaker (UNKNOWN): I
27:30Unidentified speaker (speaker_2): wasn't trying to build an app. I the example that I gave you will help you open your eyes to just how powerful. you are today. An app is a stepping stone to where you're trying to get. Okay? Sure. An app is app basically can this transcript and the steps that you will take after that, it'll take you a few hours, will absolutely leave you 100% convinced of this statement. We are all programmers. We are all programmers. The world that we live in today. a hundred percent of the world's technologically equipped, you know, population, let's call it people with smartphones and with some amount of money. We're all programmers. We just don't know it yet. We haven't figured that out yet. And why are we interested in programming after all? I mean, most people don't like programming. I happen to come from a programming uh back end uh you know background, but I didn't like programming when I was doing it either. It was boring. I love programming now because today my programming is about me getting my thoughts, my
28:33ideas, my concept out into reality, building something like Iron Man was building a suit in the first movie, and he's talking with Jarvis, right? And visually you can see the suit being built. I feel that way. I feel like Iron Man, right? So no, Alex. I am not trying to get people to build an app. And I wasn't trying to build an app myself. I just feel that building an app is the way for you to empower yourself to basically take that first step towards what AI is doing. What I was trying to do, which I've achieved,
29:07Unidentified speaker (UNKNOWN): is
29:08Unidentified speaker (speaker_2): I was trying to build employees. My goal is to build low-level, mid-level, and high-level AI employees. And my goal to do that is to massively increase the productivity and reach of my company so that we can increase our profits, increase our the power of what we are doing. Because this is what everybody else is going to be trying to do. I'm just a couple years ahead of everybody else, right? Agentic AI has only existed for nine months, just so you know. The kind of AI that can become an employee or a senior manager came out in Q4 of last year through an open source tool, not from Anthropic, not from Chat uh OpenAI, but from an open source tool called OpenClaw. The moment OpenClaw came out, and I looked at what it did, it was the first of its kind, it was agentic, it did things, it wasn't designed to be a chatbot. It did things. I realized where we would get in a year. Now, in a year, we've gone very, very, very, very far. And the next year we're going to get really, really, really, really far. But the the bottom line is what I was trying to do was to create people or virtual people that do actual work. An app is the best way to start that journey because an app is the best way to communicate with Claude. in a way that's
30:27visual for you. Now, here's what I didn't like about that app that I told you about. Again, you can only do this in steps. This app still, once it's built, you're gonna be blown away by it. You're gonna be like, oh my god, I built this app. But in the end, this app still requires you to read. It still requires you to make decisions. What we are doing in our company is we're having
30:50Unidentified speaker (UNKNOWN): it
30:50Unidentified speaker (speaker_2): becomes be full cycle. We're having it take those decisions as well, so that it's like an employee. An employee doesn't just do the work, the employee makes decisions, and then based on those decisions, it does more work, and then it makes more decisions, and then it does more work. That's what we're having AI do. And I'll I'll give you some examples of that if you want to, because uh especially real estate examples, example, so you sort of get an idea of what we're talking about. But no, I wasn't after. apps. Apps are a way, a means to an end. The means to an end is I want to work, I want to be Timothy Ferris. Tim Tim Ferris is famous. He's the guy that wrote the four-hour work week. He says work four hours a week, right? But get 40 hours of uh productivity. My goal is more ambitious because Tim Ferris, when he wrote this book, didn't have AI, right? My goal is far more ambitious. because today what I want to do is I want to work eight hours a day, and I want 80 hours or 100 hours of productivity. And I want that productivity because throughout the day I want to be talking with humans and virtual AI that are all trained to do tasks. So now I have a very, very large workforce without paying for a very large workforce. Does that make sense?
32:05Unidentified speaker (speaker_1): No, absolutely.
32:06Unidentified speaker (UNKNOWN): So
32:06Unidentified speaker (speaker_1): I would really like to kind of dig in because I have some questions regarding kind of the execution um the decision making part of it. So let's kind of dig into that, what you were alluding to of like how uh you are moving towards that you know final execution and decision-making stage with utilizing these agents uh through AI.
32:26Unidentified speaker (speaker_2): Okay. I'm buying building large apartment complexes. You could be buying single-family homes. Everything I'm about to say applies to you equally as it does to me. There's no difference, right? It applies, all of it applies to you, right? I'm going to give you my examples, but you could just take those examples and apply to buying a single-family single family home and rehabbing it, or you know, let letting it out for rent. So
32:53Unidentified speaker (UNKNOWN): I
32:53Unidentified speaker (speaker_2): already told you how the research piece works. So hopefully I've sort of filled in that gap. So let's assume that I did my research and I created these funnels, funnels being properties that I want to buy or land that I want to build
33:06Unidentified speaker (UNKNOWN): on. Now
33:07Unidentified speaker (speaker_2): imagine the next steps. I want to replace an employee that I pay $120,000 a year and an underwrite. underwriter. An underwriter is also called an analyst, but in the real estate world, we call them underwriter because what they really do is they take an opportunity like a single family home or a $100 million building and they analyze it to see if A, we should buy it, and B, how much should we pay for it? Two questions, but there are hundreds of pieces of input needed to answer the two questions. Do I buy How much do I buy it for? We've automated that process. My company no longer has human underwriters. So here's what the process looks like. We first spent months talking to all of our brokers to make sure that all of them, when they send us deals, they're going to deals at my mycompany company name.com, growcapitus.com, right? So deals at, so it's going to one email address. And occasionally, if a deal accidentally ends up in my mailbox, I just forward it to deals at
34:10growcapitus.com. Now, I have a Claude which is really, really well trained. Now, how did I train this clod? Because by by definition, people have found this, and I'm sure you found it, Alex, that when you ask AI, sometimes it gives you a really good answer, and sometimes it gives you a really frivolous. answer. Like you clearly, I mean, if you implement that, it's going to be terrible, right? You know that. So the first thing that you should do is when you want to do something specific, you first want to tell your claude or whatever you use, ChatGPT, to train itself. And here's how the training goes: it's as simple as this, and you don't need anything more than what I'm about to say. Claude,
34:52Unidentified speaker (UNKNOWN): I
34:52Unidentified speaker (speaker_2): need you to become the world's best underwriter for multifamily syndication. To do this, I need you to spend a substantial amount of time using sub-agents to go out to the web, find every podcast, find every article, and even read books. Claude, you're aware that there are many libraries on the web that have free books. Sign up for all of them and read. books and convince me that at this point you have built an incredible knowledge base
35:21Unidentified speaker (UNKNOWN): that
35:21Unidentified speaker (speaker_2): is current as of 2026 about all of the best practices of multifamily syndication underwriting. You must truly be an expert.
35:32Unidentified speaker (UNKNOWN): You
35:32Unidentified speaker (speaker_2): must understand all aspects of the job and be able to prove it with me to me. Do not do this at a surface level, do this at a very deep level. Save your knowledge in a database and make sure that that database is indexed and is always accessible to you within a tenth of a second, Claude. This is it. This is the only recipe you ever need to turn any AI agent into a an expert at whatever time. topic you wish. Right? To me, it was real estate underwriting. To you, it could be underwater basket weaving. And it will basically do the same thing. It will read every article on the web and every podcast and every YouTube thing that it can find on underwater basket weaving. Right? So this this is it. This one minute description. is what most people are missing. Because without this, you just get generic clod. And generic clod
36:33is a little bit good at 9 million things. Specific clod is really, really good at underwater basket beeping if that's what you told it to do. So now it will take hours, and again, you're going to run out of the $20 account. So sorry about that. Um, because it's gonna spay it's gonna send a whole bunch of sub agents out into the web and read everything voraciously that it can find and basically put it together for itself and say, now I really know underwater basket beating. Right? So now that you've done that step, we did it for the underwriter. We did many more things beyond this, by the way, but those were specific to our company and our beliefs Leaves and how we underwrite, et cetera, et cetera. You don't need to know those.
37:11Unidentified speaker (UNKNOWN): And
37:12Unidentified speaker (speaker_2): then we said to it, now that you're ready. you're the world's best multifamily underwriter. We want you to look at everything that comes into this email address, deals at. Now it took a little bit of work to get Claude access to deals at. And it's going to be a little frustrating. And guess how we did it, right? We asked Claude, how do we get access? Well, I can't do this, I can't do that. Yes, yes, but how do we get access? Well, there's this and that and that. Yes, yes, but how do we do this? So we had to do a few things here and there. And then eventually it basically figured it out with our help. Right? So it now has access to this email address. Throughout the day, incoming emails come in from brokers from all over the United States that are selling buildings. It opens every email, it looks at it, and we've instructed it to open the email. And basically we have these five or six filters that you know you would have as well, right? When you're buying a home, you're not going to buy a home of any age in any market of any size. You're going to have your own filters, right? So I want to buy a single family home uh in Columbus,
38:12Ohio. And uh I, you know, up to 2,000 square feet should be newer than the 1980s, uh, blah, blah, blah. And these are, you know, these are the parts of the city that I want to buy them in. Maybe that's your profile. Mine was, you know, I want to buy a multifamily property that's under 25 million, under, you know, 150 units or more, blah, blah, blah, blah, blah. Right? So bottom line is, I will I basically created my initial profile and I gave it to Claude and I said, hey, don't spend a lot of time underwriting this. Just as you take it in, rip it open. And if it doesn't match my profile, just move it, don't delete it, just move it to a processed folder. Why? Because I have an app. And now the processed counter in that app, which is at the top left, goes up by one. So it now instead of saying processed 4,142 uh properties, it now says 4,143 properties. I didn't understand, uh I didn't underwrite those, but I did process them, right? So I increase that so my my investors can see I'm working on
39:13Unidentified speaker (UNKNOWN): them.
39:14Unidentified speaker (speaker_2): If it did hit my sweet spot and one out of 10, 1 out of 20 hits my sweet spot, then the next step that it does is it actually goes out to the web. Because guess what? All we've got is an offering memorandum, which is basically a marketing document. We're not interested in the marketing because it's all gibberish and you
39:30Unidentified speaker (UNKNOWN): know,
39:30Unidentified speaker (speaker_2): there's a broker that said a bunch of things that are maybe true, maybe not. Who knows? We're interested in the numbers, right? Well, the numbers are not in the email. So Claude has to go out to the broker's website. There's usually a link in the email. He clicks on the link, goes to the website, he signs up on my behalf. Again, you'll have trouble with this. It took us a while. We figured it out. You will too. It signs up. Then it goes in, grabs all of the documents, the T12, the PL statement, the rent roll, and a few other things. It grabs them all. Then it goes to my Google Drive. Then it creates a new folder with the today's date and the name of the property and the city and the state. It goes into the folder. It puts all of these files in. And then it starts what we call a basic underwrite. A basic underwrite is designed so that Claude Clause can finish it within five to seven minutes.
40:24Unidentified speaker (UNKNOWN): The
40:24Unidentified speaker (speaker_2): underwriting part only takes it 60 seconds. The rest of it is for it to read all these documents to take all of their data so that can put that into this basic underwrite. If the basic underwrite works and it's within our parameters of profit, it then immediately turns around and starts a full-size underwrite, which is a lot more work and takes hours for it to do because it has to send emails to the brokers. It has questions about this or that, you know, the roofs, uh what about the this and what about that? And what about the salaries? You haven't included this, you haven't included that. This sentence doesn't make sense, whatever it is. So it's sending, you know, a bunch of emails out. Is it asking me about these emails? No, it's sending them out. Does the broker does it copy me on these? No. It's his job. I don't I don't have the humans copy me on every single question they're asking every single broker. So why should I have my AI do that? So it sends those questions, then it monitors the inbox. Eventually the answers come in, it takes those answers, it finishes its underwrites. What's next? It spins up a PowerPoint deck. And this is a templated deck. It has 32 slides. It takes all of the information and puts it in there. Because I am really lazy. I don't want to read an Excel spreadsheet. I want to see a beautiful deck. So it creates charts and graphs and trend lines and it gives a ranking to this to this particular property compared to the last 100 or 1,000 that it's looked
41:44at and builds me a beautiful chart. Then it invites me to a meeting. Uh, it's called Shark Tank, and it's on Fridays and it's 60 minutes long. And so the part that I'm building now, right now, is it presenting to me. I'm almost there 99% of the way there, but uh current, you know, as of four weeks ago, I had to actually go in, open the PowerPoint, and go through it.
42:09Unidentified speaker (UNKNOWN): And
42:10Unidentified speaker (speaker_2): I would ask questions. I'm in Zoom, right? So I would ask questions. Uh Claude, what about this? What about that? What about this? Now it doesn't answer my questions because it's not able to answer my questions yet. But when I finish the meeting, it's waiting. And the moment it sees the meeting being finished, the Zoom meeting, it grags grabs the grabs grabs the transcript, listens to everything, takes my questions, goes through the document, comes up with the answers, and updates a dashboard with all of the answers that questions that I've asked. And then it it sends a message to me in Slack saying, Hey, thank you for all those amazing questions. Here are my answers. Link. I click link, it gives me all of the answers. I review the answers and some sometimes I have more questions. Um, so I might ask more questions. At this point, I have to make a decision, right? I'm being presented with this data, it's my job to do this. Many times I ask more questions, that requires it to email people. So it takes a while, right? AI is only doing a small amount of work for a small amount of time, but humans You know, are in the loop, so it takes a while.
43:11Unidentified speaker (UNKNOWN): So
43:11Unidentified speaker (speaker_2): now it's basically coming back to me, giving me the answers. and eventually I might get to the point where I'm like, let me make an offer. Previously, our underwriters had to go create a template, make an offer, put stuff in, check. Claude does all of that and it does it in 60 seconds. So I click a button, it makes the offer, it sends the offer. Now I'm waiting for the other party to come back. I could continue along this, but you know, this is not a five-hour five hour podcast. But you understand this is the reality today. It does everything that I've said. Only the last week, in the last week, did it actually physically present to me because we've now got a voice agent from a from OpenAI that actually has the ability to present, right? So it talks to me and says, Hey, I want to present this property to you. It's in Chattanooga, it's 154 units, just like a human underwriter would. I've gotten to that point, but it's pretty damn raw. I think it'll take me a month to fix it. So
44:06Unidentified speaker (UNKNOWN): this
44:06Unidentified speaker (speaker_2): is me creating an entire employee. I instead of paying $120,000 a year, I'm paying roughly $200 a a year, not a month, a year, uh, for the tokens and the work that I've done. Plus, of course, my work, right? So there's uh obviously work that I am doing up front, which was probably twenty. 20 hours of work, to get the app, the workflow, the code all completely built. I've never seen the code. Uh it's written in a language called Python. You will never see the code either. You you your job is to basically have good conversations with Claude, and that's what I do. I have good conversations. I tell it things, right? Um, I'll give you one more tip before I stop this segment. I have a second AI, which is never this first AI. If the first AI is Claude, the second AI is uh OpenAI or Chat GPT that goes and checks the first AI's work. Just like humans, um, AI is not good at checking its own work. It makes because it uses assumptions and then basically it sort of gets sold on its own assumptions. So whenever checking, I never use the same AI, I use a different AI. So I have, as you can imagine, I have you know hundred $100 dollar accounts for open OpenAI AI. as well. And so, but what you can do is you can tell Claude, Claude, when you're completely done, I want you to hand over the entire package to ChatGPT, have it
45:26run a comprehensive audit, have it come back and give you feedback. If you agree with its feedback that you've made mistakes, fix it, send it again. If again it finds mistakes, fix it, send it again. until it comes back and says, I don't see any problems. And that's the package that I see on PowerPoint. The one that one AI has written, another AI has nitpicked over it. And AIs are really horrible at nitpicking. You would hate them if you were the one that were being nitpicked on. But luckily, both AIs are patient and they go back and forth. I've actually seen one instance in which Codecs Codex rejected an underwrite 14 times before it accepted it on the 15th try. Right? Now these 14 tries only took about two hours or three hours, and I wasn't sitting there looking at it. I just happened to come back to the chat and realize, oh, rejected, rejected, rejected. I counted it. I was curious. 14 times the second AI rejected the work of the first before it eventually said, I think you've got it. Right? That's an employee. That's where the world is going. That's where you're going.
46:32Unidentified speaker (speaker_1): Sure. And I think the kind of that's a that's a good kind of segment to kind of bring this in for a landing because the the biggest hurdle that at least I've run into, I guess, personally, and I certainly do not use the uh tools of AI to the degree that you do by any any appreciable measure, but is the the con the the concern of validation, right? Making sure that the assumptions that the particular agent that you're using is giving you accurate and verifiable information. Because again, it is making decisions for itself, but it you know can kind of uh you know what's the word I'm looking for, Or kind of um you know think that it's it it assumes that it's correct, right? And the the validation of the information that it's giving you, the output it's that it's giving you, being accurate to the task of which you've set it to. So when it comes to that kind of validation and that check to make sure one, it's not hallucinating, two, that it's giving you accurate and actionable items for whatever that you're doing. Have
47:32you found that the cross check, or have you ever run into problems with it telling telling a, let's say, client? Claude to OpenAI or you know, whichever other one that you're using to check it, are there certain things or maybe just kind of best practices that people can utilize? Again, are there certain things that work better for cross checking? Are there certain ways to do it besides just a cross prompt telling you to do something that can yield the most accurate results and and outputs that you're looking for?
48:03Unidentified speaker (speaker_3): Yeah. Um it's it's straightforward. Um if Codex, Codex, by the way, is the Claude equivalent for OpenAI for those that don't know. Codex is the coding application name, just like Claude Code is the you know, anthropic application name. So Codex, when I say codex, codecs I mean ChatGPT. Chat GPT. Um when you're setting that one up, the the auditor, the checker, basically say the following to it. Your job is to audit all of the underwriting, all of the analysis that Claude Code is handing over to you. You need to become an expert at underwriting auditing. Not just at underwriting, but at auditing. You need to send subagents sub agents out to the Word and build a knowledge base so that you know every single question. Read every book, every podcast, every um uh YouTube article out there about the art of underwriting and how you can successfully audit it. Then when you're actually doing the audit, attack the audit from at least three different perspectives. The perspectives cannot be similar to each other, they have to be this dissimilar. similar. So imagine three different methodologies. So, in one way, you're not auditing once, you're auditing three different times using three different methodologies.
49:22Because you're doing and that's it. That's that's the end of that prompt, by the way. Um, and that what what that does is you end up having three auditors because even though it's codex auditing, it's now forced to write three separate scenarios and then use those scenarios to attack. the underwriting and see if it holds up. That makes sense?
49:43Unidentified speaker (speaker_1): Oh, absolutely.
49:44Unidentified speaker (UNKNOWN): No,
49:45Unidentified speaker (speaker_1): yeah, it's always and that's kind of what I've, you know, at least personally struggled with is kind of the the validation part of it. But it definitely makes sense when and maybe it's kind of from my own perspective of not being you know specific enough or for kind of following that prompt trail down to a minute enough degree to where I have kind of confidence in the outputs. So again, I think this is a lot of uh, you know, really great tools for people to look at. And,
50:10Unidentified speaker (UNKNOWN): you know,
50:10Unidentified speaker (speaker_1): the concept of utilizing uh again, I understand the con what you were getting at with, you know, the app is not necessarily the goal here, but it is a good uh illustration of just how powerful this stuff can be. And developing an app kind of gets you to that goal. But the utilization. of subagents in the context of how you're prompting these things to get multiple avenues of examination for your prompts, I think are some really powerful uh components that you've, you know, looked at. Now, if people are looking to kind of, you know, learn more about this besides asking AI to teach you about how to use AI, um, if they want a little bit more of kind of a human touch or just, you know, to kind of learn more again about this on you know, without having to delve into AI specifically, are there any tools that you recommend that people um can utilize besides just asking AI to explain itself? Because some people uh you know might prefer just uh the more human touch of the explanation before they kind of dive headfirst into the interesting and exciting and albeit sometimes scary world of uh AI utilization?
51:13Unidentified speaker (speaker_3): Yeah, there are. Um So first I'll say this that 99% of all problems that people have related to AI is because of a lack of specificity. The more specific you are with AI, the better your end result will be, right? I come from a world of computer, you know, writing software. You realize that you can have a one million line of code software that breaks because I put a comma instead of a semicolon in one place. So we're, you know, I'm wired in my head through a four-year degree program to be specific. And so that works for me. But you can, anyone can wire themselves to be specific, and you certainly don't need a degree to do that. So be more specific and you'll get a better result, right? And be more paranoid and you'll get a better result. I'm very paranoid. So I'm always adding on layers and layers and layers and layers of audits and checks. right? And still stuff breaks, just so you know. I mean, doesn't always work. Um, to answer your question, the best way to learn AI is
52:13to watch other people use it, right? Because that really I mean, human beings, we learn by watching. If someone's playing baseball and we're watching them play baseball, we don't have a problem swinging that bat because we've now watched somebody play baseball. right? It it makes it much easier. So there are two different ways of doing it. One is practitioners. I'm a practitioner of AI, right? So what I do is every couple of months, I teach a course which has no PowerPoint. So there's no PowerPoint at all. It is a demo. And what I do is in 40 minutes, I demo 20 apps. And they're all different apps, you know, every time it's a bunch of different apps. different focuses. So one time I might say, here's 15 marketing apps within my company that I'm using to uh find investors for our projects, or to find uh properties, or to find brokers, right? Whatever it might be that I'm looking to find and optimize. Here's a bunch of marketing apps, right? So as it happens, on the day that we're recording, and I'm not gonna say the date, on the day that we're recording this, I have 850 people. signed up to watch me do 20 demos. So I'm doing 20 demos in 40. Um, the good news is by the time, you know, but you know, by the time obviously you watch this podcast, that would have already been done. But we record them and we provide them for free. There's no
53:33limit, it's it's on our website. The website is multifamily university, right? So just type multifamily. university into Google, click that link, and then go in there and then look for those AI. You know, it's I I don't remember what it's called. Uh today's presentation actually is called AI and Real Estate, 20 demos in 40 minutes. Go find that. It's gonna be a big banner. Click on it and watch it. Now in 40 minutes, you would have seen 20 real estate applications. And I think that will create aha moments about which of those 20 mattered. I can tell you most of them won't matter to you, but like two or three, you'll be like, ah, I need to build that right now, right now. Right? As soon as I finish this video, I'm going to go build it. Right. That's what I'm hoping for. The other way of doing it is find practitioners on the web. I watch YouTube videos two hours a day. And you know what what's interesting? I'm no longer watching YouTube to learn. I did that before. I'm watching YouTube to find things that I can add to my code repository. So when I see a feature, guess what I do? I stop watching the video, I click the share button, I copy it into Claude and I say, give me the top 10 ways to build this feature into my repository. Repository is a
54:47technical word which means a bunch of code. Right? So I basically say, I want 10 ways of using this in my repo. Then I read its answer, and often, I mean, it says stuff like, you already do this, Neil, or you do this better, this is not better, right? So I say stuff like, Okay, I'm not convinced, benchmark whatever this guy is talking against ours. Let's do a bunch of simulations and you tell me if we got an improvement. And then 30 minutes later, it says things like, Ah, you know, actually this was a lot better. than what we were doing, and here's why. And so I'm like, okay, roll it out across the repository. And it basically makes 80 of my applications better. You notice I actually didn't learn what the person did. I just watched enough of the video to figure out if there's something in there that can improve my AI implementation. So often I'll watch 30 AI videos in 15 minutes because I'm just looking at what is it that they're talking about. You need to go find those people on YouTube. And that's the easiest thing you'll ever do. Because all of us, when we watch a YouTube video, YouTube gives us more of those videos, right? So go watch five. Go search for five. And then you'll have an endless number of them in your feed.
55:52Unidentified speaker (speaker_1): Awesome. Well I think that's a really uh kind of great place to package this. We could talk for for hours, but uh my producer likes to keep these uh bound to about an hour for our uh for our uh kind of content creation so again, really appreciate uh you being on with us today. It has been, you know, a a very eye opening uh part for me too, 'cause again, with with my utilization of AI personally, it's been more kind of specific question prompt answer and that's basically it. The the theory of kind of creating an app. And I'm I'd be honest with you, I'm gonna go home and give it a shot tonight. I got uh I got some free time after I hit the gym tonight to uh go and uh you know beat down the doors of Jericho, if you will, with uh a few AI prompts. So definitely gonna see try my hand at making an app and uh tomorrow's a new day. But uh if you wouldn't mind, repeat your uh website. So if people are l looking to get uh kind of more ingrained in this and learn some more, what's the what's that website again?
56:43Unidentified speaker (speaker_2): Multifamily University or Multifamily U. you. That's the letter U.com. And then the course you're looking for is a AI in real estate. And uh yeah, let's go beat down the doors of Jericho tonight.
56:56Unidentified speaker (UNKNOWN): All
56:56Unidentified speaker (speaker_1): right, fantastic. Well Neil, really appreciate your time because uh, you know, the most valuable resource we all have is our time because as much as AI can give it back to us, we can't create any more of it. We all have the same 24 hours in a day and I appreciate you taking an hour out of yours to uh us, sit here and talk with us today. This has been another edition of the Alternative Investing Advantage podcast. Again, I always encourage people if you do have questions or you want to learn more about this stuff, engage with our guests. That's why we bring them on. We want to, you know, open your eyes to new ways of learning things, of alternative ways of looking at things. And certainly AI is going to be a very big component of that for all of us through the rest of our lives. So, Neil, again, really appreciate you being on with us today. Thanks everyone for joining, and we'll see you on the next one.