Yaakov
===

[00:00:00]

Intro & Guests
---

audioEdwardCohen21197339708: Commercial real estate is going through a massive wave of AI experimentation, but adoption and real-world value remain uneven across the industry. Today on Tangent, I'm joined by the one and only Zach Aarons

audioZacharyAarons31197339708: Thank you. They call me the Aeschylus of agentic workflow. And as we were talking about, I am a one-take wonder. Thank you

audioEdwardCohen21197339708: He's a one-take wonder, and he's also known as the Mythos of Midtown Today we're joined by Yaakov Zar, founder and CEO of Lev. Hi, Yaakov. Where does this podcast find you?

audioYaakovZar11197339708: I am in the Lev HQ on 17th and 6th in, uh, Union Square/Chelsea in New York City. I'm very glad to be here, Edward, thanks for having me. Zach,

audioEdwardCohen21197339708: Where else? The Mecca of real estate, the Mecca of PropTech, Manhattan.

audioYaakovZar11197339708: It's the place to be

audioEdwardCohen21197339708: it is the place to be and be seen. Yakov, everyone in [00:01:00] commercial real estate is talking about AI right now. From your vantage point, where are you seeing real adoption versus mostly experimentation?

AI Adoption: Reality vs. Experimentation
---

audioYaakovZar11197339708: Yeah, that's a, that's a good question. I'll talk a little bit generally and then also about, commercial real estate, of course. we collectively, and maybe some of your listeners or, you know, the, hopefully the people at the company, we live in a little bubble where, like, the only thing happening, at least for me, is AI.

overwhelming amount of activity and innovation and change, all powered by, by this, like, rapid innovation that's happening. And we adopted it like crazy, right? We have automations all over the place. Everyone on the team is using these tools like crazy. And, um, actually yesterday, we were on an onsite for the past couple of days with, with a client, an enterprise client, and they're just getting Claude seats for their team, and they're rolling it out super slowly.

And I was actually just at a conference this afternoon for, for a bit, and th- they too just started implementing Claude amongst the team, and they're rolling it out very slowly amongst people. And of course, they have cost [00:02:00] considerations and other considerations. And when you think about sort of the barrier or the, effort to go from Claude adoption as an individual to org-wide agent operations adoption, it's like light years of difference to make that happen.

So I think we are extremely early. I think that there's a lot of money being spent, and of course, all the news is talking about where, where all the money is going. but I, I think we are, we are early, and hopefully it'll pick up and, and accelerate a bit.

audioEdwardCohen21197339708: To your point, we were the other day at the Real Estate Gala in New York City and talking about, AI adoption on the brokerage side and, resonates what you just said, which is it, it's happening bottom up, even though the ultimate recommended approach is top-down at the organization level.

what benefits do you see from the bottom up, or what issues do you see with that approach?

Bottom-Up vs. Top-Down AI Adoption
---

audioYaakovZar11197339708: one thing I, I think is important is that the tools as they exist today are much more made for individuals. Like, it is a lot easier for an individual to create a Claude account, maybe install it [00:03:00] on their computer and deal with their day-to-day. Because what you're getting with these, is an extremely flexible pathway into using AI and agents, and you sort of start to, to get a taste for what an agent can do, which is again, we- we're in it, so we understand it.

But it's kind of insane that your software is making decisions around where to get data, and it's finding things for you without you being specific about it, and it's able to take these more complex actions. So the people I see who have the most success are the ones who are like, "I installed Claude, and I was up till 2:00 AM every night for a week just, like, figuring out all these crazy things, and, like, I get it.

I get what's happening now." And then inevitably, the thing that happens is they start to use the tool a bit and they're like, "Okay, I could do this, I could do this, I could do this," and they wanna do more complex things. And I hope that that sort of bottoms-up usage and then the desire to go to more complex or more org-wide usage or more sort of like [00:04:00] systemic usage within an organization will get driven by the fact that the individuals who are using these tools can almost like prototype or proof of concept certain activities that are important to their job, and then guide a more enterprise type rollout that's standardizing it a- across the org.

So I think it might work out, it might work out well, which is very different than enterprise software. No one's like buying a Salesforce seat and then being like, "Hey, guys, we need to get Salesforce in the org." With Slack, it happened successfully, which was, sort of the first, take at that

audioEdwardCohen21197339708: first tangent of the day, Zach. Wasn't the Slack founder your collaboration superpower once, the video game guy?

audioZacharyAarons31197339708: Stuart Butterfield, yeah. He, uh, fascinating guy 'cause he's just wanted to build this video game his whole career, and, uh, he continues to stumble into either nine-figure or 11-figure, other companies while failing at, his one sort of North Star, goal. But I, I would love to fail as spectacularly well as he [00:05:00] has, um

audioYaakovZar11197339708: gonna say, where is he walking? I'm stumbling on, in New York City I'm stumbling on heroin needles, not, uh, billion-dollar companies. Must be nice

audioZacharyAarons31197339708: fascinating guy. So, the first company, I believe was, uh, was Flickr. and that sort of grew out of this photo sharing module that they built in order to do something with this massive online multiplayer game, that ultimately sold, to Yahoo, I believe. And then he went back, try to build the game. Game never took off, but they built this internal communication tool for the company to work more efficiently and chat about the game, and that became Slack, and that became the, the main product. and they sunsetted the game. So I'm hoping that he, uh... Well, I hope he comes on this show.

Be great to have you, Stewart, if you're out there. And two, I'm, hoping that, he gets now to build that game, 'cause I would, even if no one else in the world would play it, I would, I would play it. I would play it.

audioEdwardCohen21197339708: Stuart might listen to the show actually. [00:06:00] Zach, talk about what you're seeing on your LP side, especially the real estate folks, and how they're approaching AI experimentation, AI adoption, and, and also if you have any perspectives from, uh, Meta Prop Labs, who's advising AI transformations

MetaProp Labs: Open Source Skills Library
---

audioZacharyAarons31197339708: we actually just launched an open source product today out of MetaPro Labs. It

is an AI agentic skills library for commercial real estate, and it has over 100 distinct, unique, real estate-related tasks that have now been agentified through the skills. We've built about half of them, uh, with our labs team. Half were other open source projects that other folks had done, um, that we found and we've kind of curated it all in one place. so that's an example not just of how far the, the industry has come, but also what I find [00:07:00] fascinating is that of the skills that we didn't build ourselves with our tech team, but the skills that we found, half those skills were built by more traditional web developer types who had a interest in real estate. Half of them were built by not previously non-technical real estate professionals who work at real estate companies that look a lot like our limited partner base. mid-market companies, small companies, large companies, kinda didn't matter. So the idea that a acquisitions professional at a real estate company or a financing professional or a property manager would be staying up all night, like Yakov was talking about, and building their own agentic skill within an AI sandbox, that was [00:08:00] unthinkable even 12 months ago when we were talking to them about it. Now we get on these webinars with real estate companies, and the level of sophistication has just accelerated in a parabolic trajectory. W- the questions we are getting about very specific, highly technical issues these folks are having as they are building their own skills and projects and tools, In whichever foundation model, they're working with. It's truly, it's inspiring, it's exciting. It makes it challenging as an investor to know what types of, AI technology is defensible enough to warrant an investment from one of our funds, and which is very cool looking, but ultimately can get vibe [00:09:00] coded away by a random property manager at a real estate company, and therefore there's not really a need for a purpose-built company around that skill.

but nonetheless, uh, I don't know exactly where everything's going, but, but it is a truly fascinating time to be in this industry and, and I feel very, very lucky that I get to do this. I, I just started texting, for example, all my friends in the real estate industry just started sending them this skills library and, like, people are going crazy.

I mean, people are really digging this stuff and embracing technology like they never have before

audioYaakovZar11197339708: Yeah.

Defensibility & AI Investment
---

audioYaakovZar11197339708: I think what, I think what you said, Zach, about the, what do you guys invest in as a, as a founder, it's even more crazy. It's like, what is, what is the value of the soft- of the software you're building? What is the value of the company that you're, you're building? And it's, it's a big question.

Like, there, there was a time, you know, a few years ago [00:10:00] where if a team came to you and they're like, "We built this thing over the weekend, and we're live now, and it's going crazy, and it's like we figured something out." It's like, that's amazing. Like, yet they figured out this niche, they're first, they're growing like crazy.

If someone comes to you today, I mean, I don't know, I don't invest. But like, if someone came to me today and they're like, "We built this thing over a weekend and it's awesome," I'd be like, "Okay, anyone's building, everyone's building stuff over the weekend, and a lot of them look really cool." And a lot of them, to your point, could be distilled to a skill.

And in fact, like my hot take is that some of these companies are actually worse than using Claude out of the box, right? Like some of the skills that you might find or copy or try to come up with without the diligence that your team probably went through putting that together are just like telling Claude wrong things or conflicting things, or it was written for like a, you know, half-ass.

And that ends up creating a huge question on just like where is the value creation for software companies on top of this? And I think for [00:11:00] enterprises it's a big question as well. Like, I meet with a lot of these companies and they're like, "We wanna do blah, blah, blah." I'm like, "There's nothing for you to buy from anyone.

Have one of your guys sit down and make a skill and give people Claude access." things are moving fast.

audioEdwardCohen21197339708: Absolutely. I mean, the, the speed of things is, is what, also I gathered from the folks I spoke to at the, again, Real Estate Gala, weakest flex that I can drop right now. the speed, uh, that the things are changing. I mean, a lot of people may, thought that they were early to the AI, boom a couple years ago, and a lot of folks built stuff that now it's not needed because we have much better integrations.

Now

MCP, Speed of Change & Industry Fear
---

audioEdwardCohen21197339708: we have MCP, frameworks. MCP, you know, for those who aren't familiar, are basically the, the protocols that Anthropic and Claude and now the big LLMs use to transfer, send data back and forth from services, uh, that you may wanna use, like Zillow, for example, or Lev, uh, that are offering clients to integrate and actually be able to do stuff and not just use the LLMs as an [00:12:00] enhanced Google search, but actually do workflows and connect with their data.

but yeah, so hearing about you know, how fast things are changing, but that's, Zach calls it excitement. I do generally think people are excited about, you know, in, in, in the past, they maybe wanted to do stuff, but they didn't have the technical abilities to do it themselves, and now they do.

On the other hand, I do think a lot of it is fear. Not only fear of missing out, but fear of, you know, your competitors beating you and, to a certain point, fear of your job or y- you as a person becoming i- irrelevant, uh, which I think is, uh, an extreme take. but yeah, I mean, necessity and fear may be the mother of all inventions of the drive innovation.

now I wanted to ask you, Yakov, i- do you get a feeling that, uh, when you speak to prospects, whether they be real estate brokers, investors, lenders, are the majority of them in experimentation mode, uh, or they're actually already looking to, okay, changing workflows and, and ultimately decisions?

audioYaakovZar11197339708: yeah, I haven't seen [00:13:00] much of people with actual AI tools or agents at scale org-wide on the enterprise side. I have seen a lot of people in, like, random pilots and some in many pilots. You talk to some, to some of these large enterprise groups who've always been, you know, they've always had a pilot process and, technology buying procurement process, but now they're, like, running 30 pilots at a time on stuff, and I'm like, "That's not even...

That's insane. That's difficult to manage." so a lot of them are in that phase of sort of like figuring out what they wanna do, and a lot of people are, you know, doing individual things. Like, people might be a-able to buy a seat or two for a variety of services that are out there. But I haven't seen much happen at scale yet.

I haven't seen much scale production use of the tools yet beyond, you know, gi-giving them a Po- Copilot license and a Claude license maybe

audioEdwardCohen21197339708: I spoke the other day with Chris Ressa, the COO at DLC Management out of New Jersey, also a fellow podcast host. I recommend to listen to [00:14:00] it, called Retail Retold. And he had a point that he, on the operator side, he's feeling the, the demo fatigue. Zach or Yakov, is that something that you've sensed already, even though we're still in the early stages of AI adoption?

Real Estate's AI Reckoning: Demo Fatigue & Pilots
---

audioZacharyAarons31197339708: Yeah, I mean, 20, 2026 are the years of the, the demo and the years of the pilot. And what happened, at least at large organizations, is there were mandates from the C-suite, and there was a lot of, uh, a lot of leeway, um, a long rope, uh, given to, mid-level executives running various departments to pilot and test, they wanted.

Uh, there was sort of more, appetite for that than I've ever seen, in my 15 years i- in this business. I think what we're starting to see right now, [00:15:00] started to see the first evidence of it, last quarter, I would say, or even this quarter, was companies, not just feeling the, the pilot test demo fatigue, but, but companies actually, uh, leaning in, and selecting one or two or three, uh, main providers.

typically in this case, I'm talking about large organizations and really making a commitment to working exclusively, uh, with them. And so I think the era of just nonstop piloting, is gonna end this year. I think, a year from today, you're going to see, Real estate companies locking in to their, their providers of choice, you know, for the next several years most likely. and that's gonna be some combination of foundation model usage coupled potentially, if they're large [00:16:00] enough, uh, with a layer on top of that. So I've seen a couple real estate companies rolling out stuff like Glean, for example. And then the third level is the, the real estate specific AI tools that still can't be achieved with quality output from exclusively the foundation models. that's the stuff that we've been trying to invest in the most. We obviously can't predict where the foundation models are going, but there are certain workflows and tasks, one example being, material takeoffs like our company, uh, H2Corp, or, construction documentation peer review is in the example of our company Light Table. these workflows are, are not easy for the foundation models to do, and who knows three to five years out. but we at least think we're safe, safe with them for the time being. So you're gonna see [00:17:00] large organizations probably have three different forms of exposure to the agentic workflow universe. Mid-size companies, probably won't have that middle layer. They're probably going to go directly to foundation model and sort of vibe coding off that for certain s- skills and tasks, and then engaging with the more... for the more complicated stuff, engaging with, uh, the, the real estate and construction specific AI technologies. and then really small companies, who knows? you know, you made the point, are there any companies actually Changing their workflow, I have yet to see that. I, I have seen for the first time companies that have fully rolled it out. Across the org, everybody's using it, everybody's turbocharged. However, and I think I've made this point before on the podcast, it's the same workflow. It's just [00:18:00] an AI agent doing it. I've yet to see an example where someone has said, where there's been this kind of eureka moment where someone has said, "Hey, the AI is so good at X that we can actually change how we do business." If you look at our skills library, every single one of those skills is something that a human was doing before we-- the, the skills were possible. It's not a radical re-imagination of how to conduct business. And so that's a super interesting topic. We should try to have maybe some genius academic-type person on the pod.

But, like, I can't even-- My puny brain can't even really think that way. I look at AI, and I say, "Oh, this AI can do this task with much [00:19:00] more accuracy than my human brain and I can do it. It can do more of them quicker." but I, I don't see it changing the way I think about doing tasks that I need to get done during the day, just agentifying them.

Like, I'm talking right now. As I'm talking, I'm writing a LinkedIn post. I'm not writing it. My marketing agent's writing it, right? But that LinkedIn post is not being written any differently than if I actually got off this podcast and sat down to write it

Real Estate's 'Move 37' Moment
---

audioEdwardCohen21197339708: That reminds me of, an amazing documentary called AlphaGo by the team at DeepMind by Google, there's a Chinese game, 2,000-year-old, Ch-Chinese chess, call it, but it's called Go.

been played for 2,000 years. And the team at DeepMind came up with AlphaGo. I'm not gonna spoil the end of the documentary, but, they ended up beating the best humans in the world at Go. Again, a game that we've been playing for 2,000 years. And the [00:20:00] shocking part of it is that the AlphaGo, made moves that humans had never made, made plays that made absolutely no sense.

This game, we've been playing it for 2,000 years. So my question to you is, uh, Yakov and Zach, s-sure, right now we're, we're in the early stages of AI, but are we gonna end up making different decisions based on AI, in what we do with what to invest, what to develop, what to lease, how to lease it?

audioZacharyAarons31197339708: You're asking when does real estate have its Move 37 moment?

audioEdwardCohen21197339708: Ba- basically

audioZacharyAarons31197339708: We haven't had it yet. Yakov, we have, what do you think?

audioYaakovZar11197339708: I, I think it's a good question, and I haven't thought about it. I-- and I'm curious if how the sort of modeling of, AlphaGo works versus how LLM sort of logic and output works, and whether it's, better suited to be able to figure out new things. I think that there's a lot of small versions of that.

I-- there is one, one thing which is, if you think about the types of activities [00:21:00] that any company can do and the amount of stuff that they're not doing because it's impractical or takes a lot of time or is too burdensome or they don't have the person to do it, n-those are net new things. it's not the same like we should have built a, building in the sky that AI figures out how it stays up or whatever.

But the amount of new or more activity that organizations and people are able to do that we haven't done before is for sure happening. And I know for us, for example, like we always were, you know, really bad at sending newsletters when we send, when we launch new features. And now we built like a whole agentic flow that's amazing and creates beautiful emails and sends them out and is super productive.

Uh, some of the clients we talk to talk about reviewing documents and ver-verifying certain information within documents that we now have agents that are doing that for them that they weren't able to do in the pa- in the past. So I hope we get there of brand-new things and, that we could have never imagined.

But I think we're, we're quite a ways away, and in the meantime, we're sort of just doing a lot [00:22:00] more than we thought could be possible.

audioEdwardCohen21197339708: That's true. That's true

audioZacharyAarons31197339708: Yeah, I think, Edward, I, I saw an interview, I saw Demis Hassabis, who's the, the founder of, uh, DeepMind, on an interview with Lex Fridman, and he was saying that games because, are great testing grounds because you can say that, well, if the AI does move 37 and Lee Sedol beats it, that's not good, right?

That means the AI is not doing anything revolutionary, it's just playing like a bad human. When you have a game with constrained parameters, even though a game like Go is theoretically, I've never even played it, but it's, but people say it's the most complex game in the world with the most amount of permutations of potential moves, right? But that said, you're still operating in a fully constrained environment. So it's a very fertile testing ground [00:23:00] for an AI, 'cause they can do a crazy move and someone will say, "Oh my God, that's crazy." But then if they win, you know that that is revolutionary. In real estate, what does it mean to win? Does it mean getting a 30% IRR?

Does it mean successfully buying out a tenant? Does it mean entering a sub-market you wouldn't, weren't thinking about entering? Like, what does it mean to win, right? It's

kind of vague and opaque, and winning might mean something different to different people, even within the same organization, let alone the real estate industry writ large, which employs, depending on how you think about it, you know, millions and millions, tens if not hundreds of millions of people globally, right?

So it's hard to measure, even if a move 37 type event does happen with the, at the [00:24:00] intersection of AI and real estate, it will be hard to look at it retroactively and say with conviction that this was a special moment in the, history of this field,

audioYaakovZar11197339708: I think there's also I don't know anything about Go, and I haven't seen this movie, but the amount of variables on a game board, if, if someone else is in the same, know, board setup at that time, this move working or not is, to your point, there's a certain amount of constraints that exist.

And in developing a building, you're not reproducing that sample again in the time, market conditions, property, exact location. There's a million factors there, and you could get close, but it's, not necessarily apples to apples w- on any, any two deals

audioEdwardCohen21197339708: All good points, all good points. And without going too much, too deep into Go, just for people that don't understand, Go [00:25:00] has roughly 2X to 3X more potential moves or positions than chess. And, move 37 that Zack keeps quoting was a move that the machine did that in the moment it was deemed a mistake by professionals.

But then once the machine beat the best player in the world, Lee Sedol, uh, that move, uh, move 37 was recognized as a, quote-unquote, "God move of creative genius." for real estate, yeah, like buildings in the sky, data centers in, in Mars, like sure, that's, that seems like a, a move 37 one day. But, you know, we don't have to go that far.

It could be AI recommending you actually to keep your, your corner retail vacant until you get certain rate or till you get certain tenant. you know, those type of like unintuitive like, oh, I'm paying taxes and, maintenance for this property that is vacant.

And maybe the AI is like, "Sure, keep paying for it because you'll get a better tenant in one year." I don't know, that could be a move 37. let's talk about more practical stuff. what practical AI [00:26:00] applications, are being implemented now or are being looked into, to create value in any meaningful way?

Practical AI Applications in CRE
---

audioYaakovZar11197339708: we think at Levelot of the day-to-day workflow of commercial real estate professionals, which is pretty much ubiquitously, whether you look at lenders, investment sales brokers, financing brokers, sponsors, developers, their leasing tenants, leasing brokers, the, the work that they're doing is extremely similar in terms of the desk work, administrative work that happens, and then they are all, to varying degrees, building relationships in the market and sort of working their unique magic in those relationships and finding them, sourcing them, developing them, et cetera.

And I think that there's a gigantic opportunity, and sort of historically, when you think about like a really good salesperson in software sales, like they're very diligent with their CRM, they're very diligent with their pipeline, they know all their numbers, they know how to put certain, you know, weights and, and et cetera onto their [00:27:00] pipeline.

And, they have all these structures, Dant and I dunno, all of them. When you look at commercial real estate, like great brokers, very few of them have a pipeline of any sort. Like, they're managing their to-dos in their e-email inbox. Yes, they have a to-do list, and they're not dumb.

They're extremely smart, and they're super diligent and focused on getting jobs done. But they're not-- they don't have the systemic support. They have like a guy in the office that they're calling and yelling at to do something and, you know, they have a support infrastructure of people. And the thing that's exciting to me is Having the best of those people at your fingertips at any moment with the s- automation and speed that you could get with these AI tools, I think is incredible.

And especially, at lev.com, what we're, what we're building is these sort of agentic flows and workflow tools that are purpose-built for these types of commercial real estate professionals. And whatever your job is, if you can build the, the [00:28:00] support infrastructure for you to deliver faster, more completely, have easy access to the tools, I think that's a, big differentiator for people.

So you have like basic things like, of course, I want someone to do an underwriting model for me, I want someone to research comps for me, I want someone to, you know, do a market survey for me, I want someone to put together an OM. Each of these things, if you invest the time as an individual into skills, to a- to, to Zach's point, and you make amazing skills, and you spend a bunch of time setting up connectors into all of the tools that you use, and you, you know, spend a bunch of time figuring out the right way to talk to your agent, you can do a lot of that with work, with a bunch of work and prep work and setup.

And what we're trying to do at Lev is build a out-of-the-box system that does all that and more, and also has the user interfaces to be able to support the more structured sort of CRM or, or deal management, type activities. I think it's exciting to think of what the future of these professionals is gonna look like, and how we can, as an industry, get more time for those valuable, useful relationships, [00:29:00] structuring deals in creative ways, finding the right people.

'Cause deal making is about quantity of that, right? If you have a high quantity of that, you're more likely to get deals done, but you need the sort of support infrastructure to be able to do that

Leapfrogging Legacy Tech with AI
---

audioEdwardCohen21197339708: We're gonna dive deep into Lev in just a second, in China, most, people never got laptops, but every single person now has a smartphone. Could we see something similar within real estate where there's people that are great at relationship building, they're great at deal-making, but they never took the tech part too seriously?

Or as you said, you know, they're just working with their email, spreadsheets, a couple docs. Could we see a similar leapfrogging from real estate pros with AI?

audioYaakovZar11197339708: I am 100% sure that that is what, what happened. If, if I think what you're saying is that the laptop is like Salesforce or some, you know, last generation SaaS. A, a trillion percent. I think the reason commercial real estate has not adopted the sort of software 2.0, infrastructure, is that it requires putting [00:30:00] structured information into certain fields and building well-defined workflows.

And software 2.0 is amazing at that. Using Salesforce for a call center operation or for a customer service operation or for a high volume sale that is very, linear sort of or, or well defined is incredible. And commercial real estate and sort of deal making in general, and especially if you look at like investment banking and deal making and a lot of stuff within finance, commercial real estate, they are totally not that.

And most of the larger companies in commercial real estate spent millions of dollars trying to build Salesforce to do that, they're like, "I need software. Am I gonna build software or am I gonna buy software? Well, I don't really wanna do either 'cause I want it to work for my business.

Okay, well, I'm gonna go with Salesforce 'cause it gives me this platform on which I could basically have custom software, but without all the annoying stuff of of building software from scratch." And then it's like, now I need to create a concept. Let's talk about a financing broker, for example.

Now I need to create a concept of just, if you [00:31:00] just wanna store a term sheet in Salesforce, you need at least 200 fields in Salesforce, at least eight or nine custom objects, and an engineer to figure all that out. really smart people to figure out how to get into Salesforce, engineer to figure it out.

And then you need an army of people who are taking the term sheets and putting it into that 200 field form because it's extremely difficult. And with agents it's like not, it doesn't matter. It doesn't matter. the data structure might be the same, but the complexity of implementing it is significantly reduced, and the compli- getting 100% compliance is also significantly reduced, or almost zero.

audioZacharyAarons31197339708: I'm gonna take the other side of that. I think in order to get AI tools to work for you, you need data, and you need it to be organized, and you need it to be digitized. And so if you haven't been using data platforms [00:32:00] and property management software solutions for the past 10 to 15 years, and all your stuff is still in the filing cabinet, can you leapfrog by digitizing all your data really quickly and then putting it into the AI?

Sure. But how many firms are there that are really going to do that? If you haven't embraced any Web 2.0, whatever you call it, SaaS data, digital data platforms over the past 10 to 15 years, are you likely to be the type of firm that magically mass adopts, agentic

AI workflow? Unlikely, right? So, like, people make this as become almost a cliché, but, like, the AI output is, is only as good as the data you feed it in, right?

So if you're only feeding it, either publicly available data or data that is, that sits outside of your organization and your historical best practices and... It's not gonna be great. [00:33:00] So you might, leapfrog in terms of your usage, how you're spending your time digitally, but I'm not convinced you will leapfrog in terms of the results you're getting out of it.

I think a lot of the most forward-thinking with respect to AI real estate shops I see are the ones who did lean heavily on the last wave of technology and have a really good understanding of their data, and of their processes.

audioYaakovZar11197339708: I think it's a good point. not to argue with my generous host of having me, but the... Thanks for having me. But I hear exactly what you're saying, and you're right. you need some level of structured data to, to get this sort of second and third order of effects. I think there's two factors.

One is, If you take Claude out of the box and you enable the Microsoft Outlook connector, and you prompt decently, you're [00:34:00] able to get a tremendous amount of, And it will cost a lot of tokens and take some time. You're able to get a tremendous amount of really great output from that.

O-one Of the things that we built to exactly Zach's point is, like, people don't have a system to import everything. c- If you set up- sign up for Lev, there's no, like, Salesforce to plug into.

If you have Salesforce, we could plug into it, pull all your data in. If you don't, where do you get that data? So we actually have built a tremendous amount of really complex, agents that are able to... When you're signing up, you're able to plug in your email, and we will go through and recreate the entire history of deal flow that you have.

It's insane. It's like a swarm of agents that go through your inbox and help you effectively recreate the deal in your CRM as if you were using it at that time, and put in the notes of what happened, and interpret the emails and find the attachments, and parse all the attachment files. Now, again, that's costly, time-intensive, but the output is back to that structured data.

And I think that [00:35:00] th- as the cost of tokens go down and the intelligence and, and sort of capabilities of agents go up, you're-- we're gonna end up with the cost of that continuing to fall, and that, barrier between structured data and unstructured data is collapsing at a rapid pace. And you're, you're not gonna have to think anymore.

And I think one of the things that's r-really interesting to think about is- A lot of these businesses, the quantity of unstructured data that they have is tremendous, and it's also private data, meaning like you're getting an OM from some guy on an off-market deal, or you're getting an underwriting model from someone on an off-market deal. You have all this not that it's secret, and it's not even necessarily yours, but it's data that you've gathered over time. And AI being able to help you unlock that through this like ingestion understanding pipeline that I think is, you know, more and more common, is a huge unlock for the amount of, knowledge an org could have.

Selling into Commercial Real Estate
---

audioEdwardCohen21197339708: so in commercial real estate, great, products don't automatically win. so Jaakko, wanted to ask you, what have you learned about [00:36:00] selling into this industry? I mean, you started being well-known as the, financing solution or solution for, for lenders, and you've now evolved to address critical workflow issues and efficiencies for lenders, brokers, investors.

So, what have you learned about selling into this industry?

audioYaakovZar11197339708: you know the sort of like, "If you, if you could, would you do it again?" kind of thing? I'm always torn. when I started working on Lev, The problem I saw was I helped finance a building for a nonprofit that, that I'm very close with, and that $4 million loan took six months.

Broker, lender, many lenders, attorneys. I'm like, "This is mayhem. This is crazy." Even just getting a quote was three months of work, and I wanted to fix that. The stupid thing I did was we decided to start that as a, as a brokerage, and my premise was let's be a tech-enabled brokerage that's digitizing the financing transaction.

It was amazing 'cause we learned a tremendous amount, but it was stupid because I'm still trying to shake that image off. We still have people who are like, "You guys are a [00:37:00] broker. You're gonna compete with me." And like, other than giving them pin-pinky promises or those blood things that little kids do, I don't know, I don't know what to do to promise that we're not.

That's not our business. We're a software team. Our team is built around software. Our engineers, salespeople selling software. Everything we do is, is around building great software for people. That's what we're good at. That's how I, grew up in the world of software. And the thing I learned is the importance of the nuances of their workflows are so important to them and how they understand and execute on deals, and you need to build for that.

And even if it seems stupid or doesn't make sense or you think there's a better way to do it, they need to understand that. And I think the other thing is, and this applies universally across software, people wanna understand what the software is doing. It's amazing that you have AI magic, like cool. But if you're telling me that I should reach out to this contact or go to this lender for this deal or underwriting this deal like that, I need to understand what did the software do to come up with that and why.

Because that's [00:38:00] contributing to how you're executing on that deal-making, right? Like we said, the deal-making is this sort of artwork that you're putting together. You wanna know what's going into it,

audioEdwardCohen21197339708: What parts of the commercial real estate transaction process feel more ripe for disruption or more ripe for, innovation and improvement with, with AI?

audioYaakovZar11197339708: it's countless. I think it's countless. I think that there's a lot in the closing process that's, pretty, a-agent ready, let's say. I think there's a lot in the underwriting process that's agent ready.

I think that there's also a lot when it comes to, big organization operations in commercial real estate of loan servicing, insurance verification, closing verification. those parts of the business usually have a lot of headcount and are time sucks. Okay. And there's a lot of sort of operational effort being put into systematizing those parts of the business.

And that's the example of, like, we build crazy Salesforce flows to be able to support it, and you can now put an agent that's way more effective at it, [00:39:00] costs way less, much less headcount.

audioEdwardCohen21197339708: I like that term agent ready or agent friendly.

What CRE Pros Are Overestimating About AI
---

audioEdwardCohen21197339708: What's one thing commercial real estate pros are overestimating about AI right now?

audioYaakovZar11197339708: I think about this a lot. So the, the pattern I see is like someone says, "I'm interested in figuring out stuff about AI." Usually what I do is I get on a Zoom with them, and I'm like, "Let's install Claude together and do something. Like we'll-- let's make an underwriting model or let's, create an OM for this deal."

And you do it with Claude out of the box. I show them skills. I'm like, "Okay, dude, have fun. You know, have a good time." And then they call me back in a day or two. I-- Literally yesterday, I got a call. I, I didn't even look at it. This guy sends me a video. he's like, "Can you call me?" on the phone, he's like: "I need to send you something." He starts sending me these WhatsApp videos, and they're marketing videos that he created for a property. This guy's amazing, longtime client. He's incredibly smart, very capable. I didn't see the videos yet, but I'm sure they're actually pretty good marketing videos about a property.

I think that the delta between a vibe-coded thing from someone [00:40:00] who fig- who didn't write a line of code before to production-ready, like live at you name the enterprise, commercial real estate company, is light years away.

Light years. Like, there is no connection with that. And I'd even suggest that even a two-person shop should be very careful about using these tools that are, that were developed in the past, eight months without technical oversight. I think it's probably not the right decision. And it sounds funny, but, like, they're spending money on ads and they're spending money on tokens, and they're, and people are actually signing up and giving credit cards.

I see a lot of this stuff on LinkedIn, and I would send them in our Slack. Like, here's some, you know, like my wife's uncle's brother was a real estate broker, and I, like, built this solution for him. Or, like, I was an analyst at, like, whatever for, like, six months, and I, like, started a real estate underwriting company.

And I was sharing them in our team Slack, and now I go to, I go to Codex and I'm like: "This is the company I just started. Please do a full penetration test to make sure that there's nothing, blah, blah, blah." And the stuff it comes back with is insane. Like, there are holes the [00:41:00] size of, you know, whatever you want in these, quote-unquote, softwares, and that scares me.

It's, it's, There's two things to be aware of. One is, like, if something was started by someone who did not build production software before and started in the past year, text me first, I'll just make sure it's legit before you give them a bunch of money.

And the other is the stuff that the LLMs do well, they do so effing well that it is really hard to justify spending any amount of money on other services to do it. And for example, like OM generation was like a whole thing for a long time.

Create an OM, upload your documents. We built a bunch of stuff around it. Ours is really, really good. Claude Design is so freaking good. Like, really good. and even before that, you can create a skill that's unbelievable, and with Claude will generate a branded, high quality, fully, diligent OM for you.

before spending money on these tools, I think it's important to, to push the bounds of the foundation models to the [00:42:00] limit and make sure you actually can't do-- they actually can't do it out of the box, or you have a unique interface around it, or you have a unique collaboration thing around it, or like something unique that is actually, you know, making it valuable

audioZacharyAarons31197339708: don't quit your day job, right? If you're seeing a lot of these real estate professionals, they

build a skill, they start getting stars in their eyes, they see the valuations of all the AI application layer companies. "Oh, I'm gonna be

we're gonna build the next, Harvey or for legal whatever, you know, 11 Labs," you know, one of these big, breakout hits.

And, and I'm just like, " in the real estate business," 'cause yeah, this is a big gap between vibe coding a skill over the weekend that you use yourself or internally with your colleagues and something that you're actually, uh, commercializing

Rapid Fire & Wrap-Up
---

audioEdwardCohen21197339708: Yakov, if you could choose one person, historic or living, to do a partnership with, who would it be?

audioYaakovZar11197339708: I think Steve Jobs from a business [00:43:00] perspective. I'm sure it's a classic answer. I think I wanna see what the, energy that created this product output feels like. I like to think that, like, we're good at design and we build great products, and I wanna experience it from, like, the best of the best. Jony Ive, I would love to work on designing a product with those people or just see them in a room.

I was like, what... It's one thing to, like, nitpick the radius a phone, right? Like, okay, that, that makes sense. And, like, how deep in the details does it get, and sort of what sort of justification do you need of my opinion, someone else's opinion?

audioEdwardCohen21197339708: I keep wondering what Steve Jobs would have thought or s- done with, AI. Imagine that he was just like, "Screw this. I'm, I'm, I'm all for humans. I'm human-centric," and, like, he just retires and goes live, lives in the woods or something like that.

audioYaakovZar11197339708: Uh, I think if he would've seen early prototypes of ChatGPT, he- Apple's agentic products today would be leaps and bounds. Because i- if, not him, but, a, an amazing product designer [00:44:00] understands the potential of where that software can get to, and with the resources there, like, I think it would've been really interesting

audioEdwardCohen21197339708: Yakov, where can Tangent listeners and commercial real estate brokers, lenders, investors, learn more about Lev and connect with you?

audioYaakovZar11197339708: If you wanna see a lot of human-written LinkedIn spam, which is rare, most LinkedIn spam today is AI. Follow me on LinkedIn. lev.com, zar@lev.com is my email. Happy to chat with anybody

audioEdwardCohen21197339708: Yakov Zhar, founder and CEO of Lev. Thank you so much for coming to Tangent today,

sharing your exciting company and your insights with us

Muchas gracias. If you learn something new or enjoy the conversation. Text a friend the link to this episode right now.

You can find links, resources, and ways to connect with us in the show notes