Artificial intelligence is moving the mortgage industry beyond simple automation and toward a fundamental reinvention of how loans are manufactured, decisions are made, and capital moves through the market. In this week’s AI Update, David Lykken and Pavan Agarwal unpack recursive self-improvement, the importance of keeping humans in the loop, and why AI systems need safeguards built into their architecture from the beginning. They also tackle a much bigger question facing lenders: what happens when decades-old processes and legacy technology can no longer compete with AI-native platforms capable of handling complex loans in minutes? From retraining employees and rethinking operational models to creating greater transparency and trust for capital-market investors, this conversation explores why the mortgage industry may be approaching an AI tipping point—and why adapting may require much more than simply adding AI to the technology lenders already have.
[David] Listeners, we’re back with Pavan Agarwal with another AI update. Pavan, good to have you back.
[Pavan] Good to be here, David.
[David] Well, I know you’re sitting there outside of a hospital. Kudos to your mom and a quick recovery to her. That’s awesome that you take time. Sitting out there in Southern California sunshine, doing a beautiful during that beautiful day and enjoying and sharing with us some thoughts. Last week, Pavan, we talked about how Dario Amade of Anthropic and Sam Altman and Elon Musk curiously agreed that we must, it’s absolutely necessary that we quote unquote slow down the development of AI. And then in a YouTube video that you sent me over the weekend, which I thought was excellent, by the way, listeners were putting a link to that YouTube video by Elle and she talks about the explanation that’s being given is something is recursive self-improvement or RSI? And what I wanted you to do is come on and explain. That is giving a lot of consternation and a lot of motivation behind these recommendations. At least that’s what they’re saying. She did a good job of kind of debunking it, but what the heck is RSI recursive self-improvement? And why is that not applicable to the AI, Angel AI?
[Pavan] Okay, so this starts from a we talked about this earlier, there was a paper MIT issued at the end of December, early January. They call it recursive language models. Okay, and the idea is is that you’re you’re constantly feeding back into language model and improving it and essentially it’s just like Excel doing a goal seek, but do that in a hyperscale. So and now with AI agents and you basically can doing tens of thousands, hundreds of thousands of these agents running at the same time and delivering a result that is acceptable to a human, right, the models get really good at being human like. If I speak to you like a human, right, and I react to you like a human, you can’t tell the difference whether I’m a human or not. Okay, and that’s what that’s what this is all about. and is that conscious? No, I don’t I don’t I don’t I think that’s a really big stretch to say is conscious. Is that artificial general intelligence? No. Right? It’s not general intelligence. it’s not like a being that can come and take over. and also the other interesting thing about all this is Colorado passed a law two years ago and it goes into effect January 2027. And one of the requirements in that law, which is super interesting, okay, one of the requirements of the law is that when a user is talking to an AI, they need to have the option to request a human oversight. Or like, hey, can a human help me out here? Okay. that that is one of the core and tenets of that law. Okay.
[David] Do you think other states are gonna adopt that law? Something similar? It would seem to be in the consumers’ best interest, especially for the suspicions that are being raised.
[Pavan] You know, I don’t know but let’s see how this rollout goes. But it’s very interesting because this is something that we at Angel AI has done from day one, which is, you know, we call it there there’s area of research called augmented intelligence. Right. And it also so what we do is that we go one step further. The consumer doesn’t even have to proactively ask. It basically based on the consumer’s feeling, we’re detecting how you’re feeling as you’re interacting with Angel. And based on the feeling, we automatically loop in a human. So that’s called empathetic technology. Okay, so we’re it’s interesting that we’re already in com we’ve been in compliance with this law for years. and because it’s the best way to do things. And the other advantage of having that kind of an architecture is that it is i it’s there’s like it’s an automatic fail safe it’s automatic kill switch that way. If the AI from the ground up is trained to loop in a human at the right moments, right, then the there is a kill switch in the AI, so it can’t escape, quote unquote escape. Right. and so that means that your AI models if you it this is not some this is a very difficult thing to put in as an afterthought. Your model has to be built to support this fundamentally. Okay.
[David] And that’s what and what you’re saying is Angel AI has already built has this already built in because of the human support that is built around this about angel.
[Pavan] Exactly. But the human support that’s built around it and you have to, right? Because this is real money that’s moving around. Right? Real lives at stake. So we from day one when we rolled this out we we had to put fail safes in in place. We c we cannot let Angel AI escape and go out and start wiring money from our bank accounts on its own. That’s our doomsday version, if Angel AI escapes and starts spending money. Right? Which is and if you think about it, how is that any different than the Terminator nightmare of mil of a military AI escaping and starts shooting up people? It’s the same problem, right? It’s you know, whether when you when the AI pulls the trigger, pulling the trigger to shoot Yeah, yeah,
[David] Whether it’s a loan decision or literally a weapon, yeah, I get your point.
[Pavan] It’s a trigger to wire money or is it a trigger to shoot a weapon? It’s it doesn’t it’s just data, it’s just it’s just yes no. It’s do I shoot or do I not shoot? Right. And your model, the foundational model has to be prepared to from has to be built from ground up to have that human oversight. Okay, and this is this is a model that we built from scratch and we and we have you know this patent pending process and this Colorado law is super interesting because it basically means that the whole world is going to ultimately need our patent to to make this work correctly.
[David] Very exciting, very exciting. I want to go so that kind of answers the RSI question, the recursive self-improvement question and why this has been such a big issue going on. We’ll see that how that plays out. But there’s another call as a result of these AI update calls that we’re doing and podcasts we’re doing weekly, Pavan. We’re getting a lot of people saying, I really need to have a solution. I’m intrigued by what Angel AI does. But we’re locked into this, the LOS we have, or this system, or that system. And it seems like we’re trying to we’re trying to figure out how to make a migration, and we’re stuck. And there’s a lot of lenders that are stuck in this. What would you say to them? I mean, they’re reticent and understandably so, to just throw out everything they have and all the things that they have and leaping into a complete new ecosystem. So I get it. But If they don’t, at some point, they’re gonna be at such a disadvantage that it’s gonna be difficult. And so it’s like do it now or do it later. What would you say to them?
[Pavan] Yep. I say give me a call, I’ll help you do it. It’s is but you have to be prepared. It is none of the old stuff works. None of your old processes work. And I get it, I get how hard it is. Like you spent you spent your you know, life, right? You know, decades. Many lenders have spent decades in this business. And you built these processes and you built these teams and they’re finely tuned and you’re getting your deals done and now you wake up and say, everything I built can be replaced by a GPU. How is that you know, what do I do next? Right? It’s just like that that article I sent you from the Wall Street Journal about SASCopolis. Right. And that CEO, that tech CEO saying, What’s the point? I’m yeah, he he’s like, Okay, I’m gonna let my company go because what’s taking me years to build cannot be built in a in a day with AI, so what’s the point? I need a new business model. Right. And that’s the reality. I mean, come on, I hate a I hate to be the bearer of bad news, but it’s actually good news. It’s not bad news, it’s good news that the industry’s becoming this efficient now. That’s so much money is being saved and you can you can you can deliver approvals in minutes, right? One and done, like in a couple of minutes is actual no I’m not talking about prequal. I’m not talking about pre I’m talking about actual loan approval.
[David] Full on, full credit pulled, income verified, all the things, all the blocks that need to be checked, and they’re all checked. And they’re
[Pavan] Right, and right, and I’m not talking about you know, people that are listening to this like saying, yeah, what’s the big deal? I can get a DU approval on a on a seven eighty credit score borrow or two years W two in in one minute too. No, I’m talking about, you know, five fifty credit score, ten job changes, you know, money coming from every which direction and credit problems and all this stuff. Getting approvals on that in a few minutes.
[David] Yeah, kid that that’s not gonna happen at at all. Yeah.
[Pavan] Yeah. Right, right. And with right, with all kinds of different paperwork and right, this is the kind of stuff that requires a lot of manual processing right now. It’s no longer needed. This is why we get our cost per loan down so low.
[David] Yeah. A lot of people I remember that article that was recently published by Brian Vieaux about the cost. And I thought it was fairly neutral. Brian’s a good friend. I really do appreciate him. And he heard it from me directly, and he’s heard it from others that he talks about the cost originate that you have compared to others. And people just look at it in utter disbelief. They just can’t believe that is possible. But that was the same reaction to Amazon. I think it was one of those articles that we were talking about the last time. And they said, and you said, yeah, do you remember the day, Dave, when Amazon, when they said Walmart sells more in a day than Amazon does in a year? Now look at where it’s at today. And that is a really good metaphor. We can’t be looking at where things are at today. They’ve got to look at because you are the Amazon of the mortgage industry for the future, because of what is happening. You can’t look at the Walmart, you know, metaphorically look at a Walmart equivalent in the mortgage industry today. And there’s many good companies that have done a lot in developing some things, but it’s changing. It’s not just an incremental change, it’s a complete tectonic shift.
[Pavan] Yep. By the by the time Walmart woke up and realized that they needed to have an e-commerce platform, it was too late. Amazon had already taken over. And even then it still doesn’t work as well as Amazon’s. Because it there’s a there’s a fundamental cultural shift that’s required in order to deliver on a AI loan production, loan manufacturing system. Right. The way you look at a loan has to change. The way you look at your customer has to change. Okay. And the old culture just doesn’t work anymore.
[David] No. Well, then there’s those that have said, I mean, I’m thinking one particular one of the major players in the mortgage space that says, Yeah, I’m gonna throw two hundred million at this and I’ll catch up. This, I mean, that I wanna go call the guy up and say, please don’t throw your money, flush your money down that fast. I mean, it’s just I mean, you’ve been doing this for twenty plus years, twenty-two years now. Is it what it is? Twenty one, twenty-two years.
[Pavan] Yeah, yeah, yeah. Yeah, something like that. Yep. Yeah.
[David] And you have been doing this with loans, and it’s incrementally been done over and over to the point these have been tested. And I think though you gave a great example for Fannie and Freddie, the cream puff, those shouldn’t need any human intervention at all at this point, anyway. And you can get those decisions quickly. But the market, I just interviewed Bill Dallas recently, and Bill’s a good friend of both of ours, and you respect him to talk to him. Bill was saying, we’re moving into a market where there’s gonna be less conforming, less conforming paper being done, more non-conforming, and to the point where like the Japanese are bringing in through your fund funds that will fund that are not Fannie and Freddie, and they’re gonna be bringing in while you tell us what the plan is there, what you see the vision is if to the extent that you can. I know there’s some stuff that’s you’re not ready to talk about yet, but I know what it is. But it’s huge, and it’ll be done. But it’ll they’ll only do it through Angel AI because of the what you have created, it’s proven.
[Pavan] Yep, the capital markets AI has changed the whole capital markets game as well. and you know, there’s literally hundreds of billions of dollars sitting on the sidelines. People who are burned in 08, and they s basically swore off mortgages, swore off swore off you know, CMOs. Because they’re basically the rating agencies are worthless, their conclusion. I don’t care if it says triple A or triple X, it they don’t really care. Right. They don’t want to put their money in those products. Okay. Now on an AI produced product, right, completely automated manufacturing, all they have to do is audit it once, audit the model once, and then you know that every single product that’s produced is going to be produced correctly. Okay?
[David] And we’ve talked about this before in some of the podcasts you’ve done, because you have a complete audit trail. Every decision, any state audit, any rep you have a complete rendering of why you made the decision down to the finest details of a single loan file. And
[Pavan] Exactly.
[David] And we don’t have that with any other system. And I think that’s one of the reasons when you start these investors are looking at it. And I think that’s why you’re very spot on by saying what you’re saying about the capital markets are shifting and they’re noticing this. And this is gonna have that Amazon, Walmart flip that you talked about as an example.
[Pavan] Right. Because ultimately the biggest lenders are gonna be the ones that have access to the capital markets. Right. Is who does the actual money trust? Right. And is the is the money is a pension fund gonna trust some mortgage that was underwritten by a bunch of people and the documents prepared by hand, who have you know, and the people who are working on this stuff don’t have the investors’ best interest at heart. Right. They don’t have they’re you know, unfortunately they often don’t even have the bar’s best interest at heart. Right? They they’re only their only interest is making their commission for the month or the bonus.
[David] Looking up the name of the book, The Speed of Trust, and I try to think who wrote that book. It’s a book that was written some time ago, but it’s a really good book because what we’re talking about is the capital markets, there is trillions of dollars that one invest because in housing. They got burned in the last cycle of 08, and they’re going, we’re not going back in until we know that we have a system that we can trust. And that’s coming down to and their data, they’re about data. They’re wanting to know how we do. The rating agencies is what really they lied. They flat out lied about what they were looking at. And now because they weren’t verifying what they said they were verifying. So I’m we’re really at a unique place in the market that I just can’t stress to people and I’m talking to them. They’re calling us, calling me as a result of this. And I do encourage our listeners to get on the phone and call you. You’ll take phone calls. You’ll talk to anybody. You’re so accessible.
[Pavan] Yeah, yeah. There’s I mean the rating agencies you had a bunch of hundred fifty thousand dollar, two hundred thousand dollar analysts making decisions on, you know, ten billion dollar bonds and they were being bullied by Wall Street bankers making millions of dollars every year. Guess who wins who’s gonna win that? Right? Exactly. Yeah, yeah.
[David] Yep. No, Wall Street people every single time. But the end investor is the one that loses in that, and that’s what happened. So the point of what we’re trying to say here, listeners, is there’s lots of money wanting to come in through means other than Fannie Me, Freddie Mac, Ginnie, Freddie, I mean all the various secondary capital markets that we have that are approved through the right now through the federal government. We’re gonna see that come in and that interview that I did with Bill Dallas, by the way, listeners will put a link to that in interview. And Pavan has captured Bill’s attention as about as because of the decisioning that you have built and others, there’s money coming in to and talking to you. Literally, whole countries are talking to you about setting up and the housing finance system for them using Angel AI. And so if it’s good for a country, why would it not be good for others and that’s why I’m wanting I’m a big fan of what you’ve built Pavan. I’m a I know you personally, I know your family personally. I know the level of integrity that’s there. And I’m begging people please before it’s too late for your business because there’s people that are employed at your companies that quite frankly are going to be out of work if they’re if you’re not making this shift. And it’s not an easy one. Again
[Pavan] That’s a really good point. That’s a really good point ’cause it’s not if you don’t make the shift, they’re gonna definitely be out of work. If you make the shift, a lot of people don’t make the shift because they think, you know what I’m gonna do with this person, this process or this LOA or whatever, right? But actually here’s the fact. If you make the shift, those people just get retrained and retooled and they’re more productive than ever before. This isn’t this isn’t a this or that game. This is this is make the shift. Do more business, do better business and your people make more money. You don’t have to you don’t have to let your people go because they just they shift into sales and customer relationship building.
[David] Yes. No, it’s really so true. Pavan, good discussion. I’m so grateful that you took the time while you’re there at the hospital. Say hi to your mom. Wish her well. Continue recovery. So glad everything’s going well with her and your family. It is a joy. And these updates are getting listened to by tens of thousands of people every day. And so I really appreciate you taking the time to come on and be with me. Thank you.
[Pavan] Thank you, David. Cheers.
Important Links
Pavan Agarwal is a renowned leader in the mortgage lending industry and a pioneer in bringing artificial intelligence to the financial markets. Agarwal serves as the President and CEO of Sun West Mortgage Company and Celligence International, LLC.