Paul was MIT's first hacker in residence and has since taught, mentored, and advised thousands of entrepreneurs around the world. Paul designed and now leads a new advanced course he developed to help student entrepreneurs build their businesses, called Venture Creation Tactics. Additionally, Paul teaches hundreds of MIT and Harvard undergraduate, graduate, and PhD students each year in the historic New Enterprises course. His teaching materials have been used by educators around the world. Prior to his work at MIT, Paul was Co-Founder of Oceanworks and Work Today. Oceanworks is a for-profit company with a mission to end plastic pollution. Work Today is a venture-backed digital staffing and recruiting company. Paul has also built technologies for and consulted with BlackRock, Prudential, Mazda, and many, many other companies. He serves on the CNBC Disruptor 50 advisory board, the MIT Sandbox funding board, and judges the U.S. Chamber of Commerce's CO—100.
He is the author of Disciplined Entrepreneurship, Startup Tactics, as well as his just-published book, No One Works Here. Paul is a graduate of Bentley University and lives in Boston and has 37 AI agents helping him run his life. What did those 37 agents do this morning before you started to join me?
Sometimes it's hard to keep up with them, Willy. Sometimes it's hard to keep up with them. As best I try, they're off doing things all through the night, which I find to be absolutely wonderful. But real talk, I just saw, I just got a notification that the agents are down because I hit my usage limit with Claude. So I think I need to give them a little bit more budget for today to get them back in action. Regardless, I'm glad to be here with you, Willy. We're going to make the most of it, regardless of whether they're working in the background or not.
I think it's important, one of the things you just mentioned is one of the things that I want to demystify as part of our conversation today about these chatbots and how really, if we look forward to the future, it should be about a lot more than just chatbots if we are to achieve what I will refer to as the infinite speed limit for any organization. There's plenty we can cover on our way there.
Let's go to 30,000 ft at the beginning because I think the narrative on AI has shifted dramatically between 2025 and 2026. Many of the frontier model CEOs, as well as others, whether it's the hyperscalers or others, have sort of changed their tune, Paul, from AI is going to take away jobs, it's going to some degree, cause all sorts of social issues in our world, to no, it's going to be used to augment what humans do, and it's going to actually be a job creator, not a job eliminator. As you publish your book, "No One Works Here," is your title more appropriate for 2025 than 2026?
I actually think it's more appropriate for the here and now, the present and the future. I want to also make sure that we understand what that really means, because what meets the eye is not necessarily what happens underneath. Also, for the record, I didn't name the book. My followers on LinkedIn decided what the book would be called by voting in a poll. Where they landed was "No One Works Here." No one works here-
What do you think your- Hold on a second. Hang on a second. What do you think your 37 chatbots would've come up with if you gave it to them rather than your LinkedIn users? Think they would've said the same title?
You know, I don't know. That's interesting. I think, given that the technology is fundamentally probabilistic, I think it would really depend on how I framed it for them. If I said to a large language model, "Here are three options for the title of my new book about X, Y, and Z," I think there might be one most probabilistic answer. I also think if I had shared the context that I want to get it in the hands of as many executives as possible, it might go in one direction. Whereas if I said something along the lines of, "I want it to be as provocative as possible," then it might have landed with No One Works Here. With or without the framing that what No One Works Here really means is not that people lose their jobs.
What it actually means is that work has changed, and it is continuing to change fundamentally. What we think of traditionally in terms of work actually looks very different now. It is less about doing the thing, and it is more about setting the intent and the goals for the agentic system to go and execute on our behalf. That is a fundamental mindset shift as we think about leading an organization, and it does require a redefinition of that word organization, something we talk a lot in the book about. Ultimately, what it comes back to in terms of what the agents might choose to title the book, Willy, is probability. That is the way the technology works, and my framing of that question to a large language model is what will ultimately determine which option it chooses.
Before we dive down deeper as it relates to how companies ought to think about using it, establishing it, leading with it. NVIDIA came out with monster earnings last week. It was the most profitable quarter by any company in the history of American business. The history of the world business. It is just stunning. Yet the market reaction on the day of the earnings was very muted, like flat, and then it was up quite a bit the following day. What does that say to you, Paul, as it relates to where we sit today on the future of AI? There are plenty of concerns today as it relates to the cost of tokens coming down significantly and also open platforms replacing the closed source frontier model platforms. What is your take on those two big issues now from a 30,000 ft basis?
Yeah. The way I think about it, and even coming back to one of the things that you said earlier, Willy, is is this going to create problems in society? Yeah, sure. Absolutely. We cannot ignore that. Is it going to also create tremendous opportunity and all sorts of new jobs? Yes, absolutely. I believe both of those things are true. When I think about what is happening, NVIDIA's most profitable quarter ever, I think that is fabulous. I think that is an indication of demand, commercial demand for this technology, which is phenomenal. I think one of my reflections on that is that it is still early innings. While that may be a historical quarter in the economy, I think that is really just the first of many to come, and that is not a prediction of what happens to the stock market overall.
It is more so my outlook that the way in which this technology is being adopted is something that will take time. It is something that is slow, and that is not because of the speed of the technology. Rather, it is the speed of the people implementing the technology. Right now, one of the things that I see from my vantage point is that the technology is advancing faster than human habits are. Changing a human's habits takes a while. Changing an organization's habits takes even longer. When I see that milestone moment, I think, "Wow, this is phenomenal," but really this is just an indication of what is to come as human habits catch up or pace with the technology advances. When I think about that, I think, "Wow, we still have plenty to come," and we should be watching that very closely.
But at the same time, what you highlight in terms of open versus closed models, things along those lines, I think that's where a combination becomes extremely important. Looking at the ways in which we leverage those different types of models together. One of the things I think about when I consider NVIDIA is that NVIDIA's demand will continue to rise regardless of open versus closed models, given that those models run on top of NVIDIA products. Now, when I think about the frontier labs, I think they're developing some of the most advanced models. Those are extremely valuable in such a wide variety of use cases. But there are also a lot of use cases where perhaps we don't need a frontier model. Perhaps we don't need the latest and greatest, and I have many of those running right now.
I think, wow, from a cost perspective, we need tokens to come down, and I believe that they will. But even as the cost of tokens comes down, there will still be a number of considerations that would lead an organization to choose to use open models, instead of using one of these closed proprietary frontier models. So that's where I think the integration of the two becomes extremely important for any organization who's trying to drive economic impact moving forward. One of the things I'd ask you, Willy, is like you see that, you're clearly tracking it quite closely. What was your reaction? Especially, when you think about Walker & Dunlop and you think about the firm you run and many of the folks that you work with.
Yeah, what do you think of when you think about where your organization goes, knowing that that's what's being reported in the market today?
First of all, we use the frontier models today, and so focused on whether we would move to open source and bring it back onto our own servers. Data security is obviously something that's very, very important. I think there was a sense at the beginning of 2026 that if you went and partnered with Anthropic or OpenAI, you were doing the right thing because they were so cutting edge and moving so fast. I think now that is flipped around and there are a lot of companies saying, "Hold it. You give them your data and you've given them your business." So there's a move to say, hey, data security, data integrity is super important.
I would also say, Paul, your comment on the implementation of the technology and the need for either open source models that allow you to continue to do things more efficiently, more insightfully, yet does not need to be bleeding edge. There are plenty of companies that you work with, and I am assuming in the Cambridge, Massachusetts area, you have met with biotech firms that are really at the cutting edge of the forefront of science and technology, and they need to be using the bleeding edge model that says that we are going here to find a cure for cancer, for instance. In my business of processing mortgages, I like to joke, I will use the University of Colorado as an example because they are just up the street from us.
Sure.
We have someone at Walker & Dunlop today who has got a University of Colorado degree, and he or she is doing X, Y, and Z inside of Walker & Dunlop. If I can upgrade that person's educational capability from a University of Colorado to an MIT degree, that sounds pretty great, but I do not need the MIT degree to have 15 PhDs to be able to do the work that we are hoping the technology will do. I think the other challenge to that, Paul, is making sure that we as an organization are taking that person who is at the company today, who graduated from the University of Colorado and is doing something here that is very important to what we are doing, and we find a way to take that person's knowledge and apply it in a different place inside of Walker & Dunlop.
We are only a firm of 1,500 people. When I look at our competitive set of these firms of hundreds of thousands of people, I see the opportunity for us to move into the opportunity in a dramatic way and move into new verticals and do new things. Whereas I think that some of these larger companies are really faced with the challenge of some of our big competitors, if you will, they have a banker on every corner.
Sure.
They have hundreds of thousands of employees. For them to use the technology and continue to innovate, they do not have a lot of white space to move into if they stay in our industry. They are faced with the challenge of how do I either move into different industries with my company and my brand or just scale down, whereas we have the ability to scale into it. That is just the way I am looking at it now.
I love it. There are so many things that you just mentioned that touch on a variety of thoughts I have going on, and you are right. A lot of the biotech firms, they need those frontier models. They need the best of the best to uncover new potential therapies or what have you. At the same time, one of the things that I find folks often confused by when I share with them my AI-driven enterprise model, which is a model for business growth that I developed at MIT, building on a couple of traditional business growth models that we have taught for many, many years. A lot of people think, "Oh, Paul, that model of rapid scale with limited upfront resource requirement to get started, that is for tech companies or software companies or AI companies." I say, "No, no. Let me be very clear.
An AI-driven enterprise is any business in any sector of the economy that leverages and deeply embeds AI in every functional area of the business. Think marketing, sales, engineering, design, operations, HR, finance, you name it." That is where even in a biotech firm that might need those frontier models for, let us say, R&D, for example, there still may be many other areas of their business that may be best served with an open model. That is where I think really the combination, even within one firm, can be really powerful. I love your comment about a University of Colorado student working at Walker & Dunlop, because I feel like that is one of the things that becomes so important is we have got folks in the workforce. How do we help them upskill and be their best professional selves, right?
I always talk about how I am not trying to use AI to cut costs or displace workers. I am using it to help make every individual do their most impactful work, to do the work that brings them the most joy, so that they can offload work that they might not be able to AI, or the work that, I do not know, feels more mundane to them. Ultimately, that upskilling is what helps to build the affinity for the individual to the organization. It is our investment in them saying, "Here is the time, space to go and learn what your role may look like in the future," which I think is being so incredibly powerful.
I want to double-click on that for a second because there are two things that I think about when you bring that up. The first is that I met with a team here in our offices last week talking about the use of technology and how they are using it to either grow our top line or shrink our middle line.
In the growth of the top line, we talked about using AI to make them more knowledgeable, more insightful. Then I asked how we are actually using that knowledge to win business. What was very interesting about it was that everyone around the table was clearly using AI to be more knowledgeable. But when asked, how has that helped us win a new client engagement, win a pitch, there was one example of, oh, well, we took this view of the market, put it into a pitch, and it opened the client's mind to a new way of looking at the asset, a new way of looking at what they could do with the asset, and we have now put it in every pitch subsequently.
But I was interested in that we have to be very careful as it relates to what is the return on the investment in this technology to make us grow the top line. The other one, then I am going to turn it to you to dive in on both of these. The other one as it relates to the middle line-
Yeah.
-the costs. One of the things that we have found that I find to be fascinating is that, so we have bankers and brokers across the country, and the way that we have our banking and broker teams coordinated is that there is a banker or broker who has these great client relationships, and then below him or her, they have a team. In some instances, it is just one analyst. In other instances, if they do a lot of business, there might be three or four people on the team. They are dedicated to that banker or broker. So when a deal goes to X banker at Walker & Dunlop, he or she takes it in. Their client gets to use that team at Walker & Dunlop.
One of the things that we have been focused on recently is that if we stay in that org structure and you do anything to create process improvement on the ingestion of data on deals that we are underwriting from a banking or brokerage standpoint, that it might make that team more efficient, but if the inflow of deal flow does not change because that banker or broker is still going to do X number of units a year, go do 30 financings on the year, or go sell 15 assets on the year. Unless that person steps up their sales capability, all of those efficiencies that you have gained by using AI do not actually benefit W&D. So if you stay in that sort of, I will call it a one-to-one, many times it is one to three, one to four.
If you stay in that type of an org structure, you're not going to be able to derive the benefits. The only way to actually do it is to convert into some type of pooled organizational structure that allows for the technology to create latency, which we're going to talk about in a second because that's what you think that corporations need to really focus on. Once you've created that latency, you now have the deal flow coming from a broader origination sales force that then picks that latency up and allows you to actually benefit from the efficiencies. Dive in on those two things because we're right on top of those two things.
Oh, I love it. I love it. We got to start with ROI. The return on investment with AI, it's something that everybody's looking at and thinking, "Where's the short-term ROI? We're not seeing it. Are we looking in the wrong places? Are we doing the wrong things?" The thing that I always come back to, Willy, is the fact that when a high school student decides to go to university, and they say, "I'm going to go do," let's just say for sake of example, "a four-year undergraduate degree." They're making a significant investment of time, let's say four years, and money, the cost of tuition, right? When I think about that, I think they're doing that, they're making that investment, which in the grand scheme of their life to that point in time, their age, that's a pretty significant investment.
They're not looking, however, for the ROI working the on-campus job their freshman year in the cafe in terms of their hourly wage, right? They're looking for the long-term salary multiplier to ensure that they are truly competitive as a job seeker in the marketplace moving forward. They're looking for the long-term salary multiplier. I believe AI is very similar. If we're looking for the short-term ROI from the investment that we're making with anything and everything related to AI, then we are in effect looking for that hourly wage increase in the cafe. I think of this as less about are we going to see the short-term ROI, not to say that we shouldn't, but I think if we're looking for needle-moving ROI from a transformative technology in the short term, that is shortsighted.
The way I look at it is it's not just about a person who knows and uses AI replacing a person who doesn't in the workforce. It's more so taking that long-term sustainability view as it relates to the organization that an organization that deeply embeds AI in every functional area will replace one that doesn't in the economy. That to me is the bigger picture risk that every organization should be thinking about. Not to say that we shouldn't be measuring ROI, it's just if we're looking to have a significant impact on the organization's overall ROI in the short term, we might not find it. That doesn't mean that our people aren't learning, that the challenges that we encounter as an organization won't be able to be solved even faster by these individuals.
We're really building a capability within the workforce in our organization that says we are going to be competitive. We are going to compete with whatever new market entrant that emerges or whatever our competitors do to steal market share. When we started, Willy, you referred to the young whippersnappers. I like to think that they are not out there just to profiteer, but really to have an impact in the world. One of the things that I see from two different perspectives of working with the AI native startups, starting and scaling from nothing, and the S&P 500 and their C-suites, looking at what they're doing, looking at their competitors, looking at what they're doing internally, everybody is trying to increase the clock speed or the velocity of the organization.
Clock speed is one of those things that says how quickly can we move, but it's also how quickly can we learn. Right now, where we are in this AI adoption, we're largely in a learning phase. How do we help each and every member of our workforce learn as much as possible? Studied the role of AI literacy at the board level, at the C-suite level, and we know that AI literacy is one of the things that correlates to increased AI implementation. Really, we should be thinking about this as an investment in learning, teaching the workforce, helping our workforce learn, but also ourselves. That's what I would say on ROI, thinking very broadly, and I think it's just so important to keep that in mind because too many people are getting overly frustrated in the short term, having not seen it show up in the P&L.
That's what I would say on ROI, but I recognize that you brought up this idea of if we are going to see it at the top line for the organization or the middle line, then really there's other changes that need to take place. Willy, one of the things that I always say that I think really kind of translates back to what you just shared with me about how you think about that banker and their team is this idea that AI tools plus AI training does not equal AI transformation. AI transformation requires a fundamental organizational redesign. That organizational redesign, from my perspective, says this organization needs to look different. Not that we're removing people from it, but that we are reorganizing, changing our operating model, and that all starts with the redefinition of the word organization.
That word has traditionally meant a group of people working together in an organized way for a shared purpose. But I reframe it to say it's a group of nodes. Some nodes are human, some nodes are AI agents, and if we are not redefining the word to reflect how those human nodes and AI nodes come together to work in an organized way with shared purpose, then we've effectively left what will become a significant portion of our workforce. Again, not to say that we're taking jobs away from humans, just to say that we are going to build more and more and more agents that will help the organization thrive. Then we don't have shared understanding of what it is ultimately that we are building to remain competitive in the marketplace moving forward.
I would love to dig into some of the details on how might we redesign that, but I think it requires rethinking the roles, rethinking the workflows like you highlighted, but also rethinking the skills that we need for each and every one of the folks on that banker's team. In my book, "Startup Tactics," Willy, we said a lot of people talk about founder-led sales. When you start a company, the founder is kind of doing the sales. They are doing all the sales up front, but then even as they hire in sales talent, they are still closing the deals at the end of the day. I made the argument that really what we should be looking for is founder-led everything.
In the early days of starting a new business, the founder needs to be able to do a little bit of design, a little bit of marketing, a little bit of sales, a little bit of engineering, a little bit of design. Kind of needs to be extremely well-rounded and able to go do this, that, and the next thing. I think that becomes even more true where when you reference that banker and their team. Perhaps their team might need more deal origination. They may need the skills that make them more entrepreneurial, where they can actually go and pull in those deals in the short term before that overall reorganization has happened. The best news is that AI can help them do that in a much more efficient manner than having to go hire a bunch of folks who can do deal origination.
Those are some of my thoughts on ROI and on how that org structure needs to change. I do not know, maybe we will wind up going down the path of state-of-the-art org charts, which I think need fundamental redesign.
A couple of things on that. One is you talked about boards of directors and C-suite. I think the stat that I read in one of your articles was that something like 78% of boards do not really understand AI, and something like 92% of C-suite executives overstate their understanding or use of AI. So there is definitely a curve that established businesses, their leadership, and their boards need to get on, if you will, to get an organization to react and move quickly enough in what is coming. What are some of the either tactics, tricks, hacks that you have seen scaled organizations take to be able to get both the board and the C-suite on that learning curve?
Yeah. It's a great question, and I think the board level literacy is one of the most important components to this because realistically, part of their role is to manage risk for the organization, but the organization now has a new type of actor, if you will. And that actor is not human. It doesn't work the way many of the actors that they have led or managed over the course of their careers act, and it introduces significant risk. The best way that I've seen is to really take a few days aside, and I would say for the first time in a very long time, we are seeing entire executive teams and boards of directors taking a day, two days, three days away from all of what they have going on in their business, and they're coming and spending it with us at MIT.
They're coming and spending with us at MIT saying, "We need to upskill. We know that." Instead of feeling like they don't know and they put their guard up, they say, "No, we don't know, and we need to know." So they embark on an AI literacy upskilling program. Now, this is something I'm seeing a lot amongst chief human resource officers is their need to upskill not just the executive team, but also the entire organization. The non-negotiable that I have and that I know is one of those I don't even want to call it a trick. It's not a trick. It's really just a best practice, is getting 100% buy-in on this transformation, this business transformation leveraging AI from the CEO.
Now, I don't embark on an AI transformation program unless I've sat down with the CEO, looked at the whites of each other's eyes, and we said, "Yes, we are all in. We are in this together." And I make a CEO commit to investing their time personally to going on that AI upskilling journey together with the rest of their team. And the reason why is because unlike many of past technological advances, there is a need for rapid upskilling amongst the entire workforce in an organization, and it's very difficult for every individual within an organization to justify taking time out of their day on a regular basis to say, "I'm going to go learn AI," which sometimes may not be productive for the project deadline, deliverable due date that they have coming up.
The only way they can really justify that is if the CEO is doing it, too. The CEO is doing it, the executive leadership team is doing it, the senior leadership team is doing it, their manager is doing it. The individual employee can then justify taking time away from what is expected of them. So that's the biggest thing for me, really. But coming back to what I said before, Willy, AI tools plus AI training does not equal AI transformation. I think it's one of the many mistakes that many organizations can run into in deploying an AI tool and saying, "Here's the training. Have at it." Because really this is a matter of that habit change that we talked about earlier. How do we actually build programmatic structures around that training to ensure that the culture of the organization shifts as well?
The way I think about that is much like I run educational programs. I run a variety of different AI upskilling programs. It's like, how do we get people doing the training, but also build in the accountability mechanisms where they have to actually show off what they've done, whether it worked or whether it didn't? I literally go back really to the idea of elementary school show and tell. I have companies do show and tell every week, both amongst small individual teams, but then also nominating folks to go to all-team all-hands meetings. This is something that is actually embedded in the culture and the way of work, the way of doing that work shifts not just for an individual and not just for a team, but for the organization overall.
These are small steps on the path to building out a more agentic organization, which requires that state-of-the-art org chart. But we have to do these things to get everybody comfortable with it on our way there.
So much in there.
How do you think about it, Willy, when you think about this idea of getting a team upskilled? How do you think about it? What are some of the things that you've seen others do or that you all have done, or also if there's things that you think about and you're like, "Oh, wow, this seems great, but it also presents this concern or this risk." What are some of those things that you all think about at Walker & Dunlop?
Well, there is a ton that I think that all We are a relatively scaled company, thousands of employees, billions of dollars of revenues. When you have an existing client base, when you have a certain trust environment, turning that over to technology to either interface with the client or enter your trust environment. A perfect example is that we underwrite thousands of loans every year. We lend tens of billions of dollars every single year.
One of the ways that we are confident in lending the money the way that we lend the money to the clients that we want to lend money to is our underwriting processes and the human interface of, "Well, that looks like a good loan, t hat does not look like a good loan." If you look at the commercial side of the business versus the single-family side of the business, the single-family side of the business is all algorithms. You can apply technology quite significantly there purely because of the size of the bets.
Sure.
Okay? When you are doing single-family mortgages and the mortgage is a $200,000 mortgage, you do not want it to go bad, but you are pooling it in with enough mortgages that if one of the $200,000 mortgages goes bad, you can handle that. When you are doing a $200 million mortgage that goes bad, that is a really bad day. As a result of it, our business is much more bespoke, if you will. As you try and drive efficiencies through that, you try and use technology to get to a point where you can actually make that credit decision. The way that we have done this business for decades, not quite centuries, but getting close is to rely on the human interface of, that is the data.
The data is, A, good data, B, that person has the skill and capability to do it, and we are applying technology against that to make that decision quicker and more insightful. At what point do we allow the technology to actually make the decision? That will be a hard question when we get to it, of okay, we do not need a human interface to it. Computers land airplanes every single day. There is a pilot up there because we all feel better having a human person sitting behind the actual controls of the plane, but we all know that the technology is actually doing it. The idea of making a decision on a mortgage should not be It is not a life and death decision. The technology should do it.
Yet it's that transformation from a company standpoint of how do we ingest the data, how do we then collate the data, how do we then analyze the data, then how do we make tens of millions of dollars, hundreds of millions of dollars credit decisions based off of doing it? Then at the end of the day, how does our client continue to interact with us? Because one of the other pieces to it, Paul, is that the reason money is the greatest commodity in the world. There's no difference between our money and Goldman Sachs' money, Wells Fargo's money, or anybody else's money. It is the absolute commodity. You can actually see a little bit of difference between wheat that comes from Iowa or from Kansas. You can find differences between the wheat, even though wheat is a pure commodity.
Sure.
Money is the purest of all commodities, right? So our money is no greener, it's no more valuable than a dollar from Goldman Sachs or Wells Fargo. So what differentiates it are people, the way we structure deals, the way we make credit decisions. Anything you do to standardize that, anything you do to say, "Oh, it's all the same, it goes into some great AI calculator and it comes out with an answer, and that's how much we'll lend to it, and this is the loan-to-value ratio," you are diminishing the value of that human interface. You are diminishing the service that we provide. So one of the other things is we drive to get efficiencies in here. As we try to standardize, there is something in the nature of it which diminishes the value of the service you're providing.
That, in and of itself, is a huge challenge, and I'm just talking about commercial real estate lending. You can run this out across industry after industry, after industry in the services space. You look at the accounting firms. The accounting firms are using AI tremendously, and you sit there and you say, "Okay, we're a publicly traded company. You can go look at our proxy and see how much money we paid KPMG last year." We pay KPMG a fee for the number of human work hours that go into them processing our accounting every year. As AI does more and more of that, should our bills from KPMG go down commensurately? Lots and lots of people are asking for just that.
So in a services environment, you run the very real risk that the, quote-unquote, "value add" you've added is going to diminish the revenues you can therefore charge, therefore your entire business model.
The one question you asked, I'm going to let you take up with KPMG directly. But in all seriousness, I love your framing around money being the ultimate commodity, and I think it's really important to look at it as just that. But I also think within the context of a business like Walker & Dunlop, it's really important to weigh all the opportunities that we have to implement AI. One of the questions I get most frequently from executives is, what do we do with AI? Not because there aren't a million and one opportunities, but because prioritizing them looks a little bit different than prioritizing the opportunities to implement emerging technologies from the past.
You touched on at least two of, if not more, of the four key considerations that I ask every executive to evaluate, to build a mental model around anytime somebody says, "Let's use AI for that," to ask themselves the questions, first, is this feasible to do with AI? Second, what is the business impact? Is this going to move the needle for the organization? But the other two that I think are really important, especially within the context of your business, are what is the risk associated with using AI for this? Like you highlighted, for single-family or multi-family mortgages versus a commercial loan, those are two very different ballparks with different levels of risk that should be addressed appropriately in terms of the implementation of AI.
The fourth is actually one that was added to the list after it started out as three rooted in a conversation that I had with a senior executive at a financial institution. That fourth question is should AI do this or should a human do this? AI probably could do it, but it may be more beneficial for our organization to have a human in the loop for that. That's one of those things where I think a lot about the organization's core values, but also what their stakeholders expect of them. Now that might be the customer, it might be regulators, it might be investors, it might be any number of those different stakeholders and their expectations. We need to make sure that we continue to meet them.
If they expect that that is done by a human, for example, that human relationship with a client, then we need to keep that part human. What happens for another opportunity to implement AI, maybe to supercharge that human is another question with a different set of evaluations on those four key characteristics. But I think those are four that I think about all the time and that I always want people evaluating in their minds because just because we could do something with AI does not mean that we necessarily should.
You counsel a bunch of super smart people at MIT and Harvard University and other places who come to you and say, "Hey, I've got this great business plan. I've read your book, I've taken your course. I'm ready to bound into it." The software- as- a- service industry or everyone who is a software- as- a- service provider has been under a lot of pressure recently from the public markets-
For sure
-basically saying that AI is going to do what they do. I thought Salesforce's earnings last week were interesting in the sense that Salesforce, I view as a horizontal provider of software- as- a- service, where they go across industries. They have these incredible records. Walker & Dunlop, we use Salesforce. It's not like we're going to move off of Salesforce because we've got everything embedded in Salesforce, and it is our deposit of record, if you will. We take all of our records, all of our client records, they sit in there. We need Salesforce, and as Salesforce innovates on AI, it's going to sit underneath Walker & Dunlop, and I think lots of other companies.
You must have students who come to you and say, "I want to do a SaaS model of some great software in X vertical." Let's just use my business. "We're going to go in and disintermediate Walker & Dunlop because all those things that Willy Walker talked about on the Walker Webcast, we think all of those can be done away with by technology, and we're just going to step in and go do it.
Willy, you said it, not me, but yes, okay.
No, but one of the things that I have great confidence in is the fact that two MIT students who are super smart with a business plan, they might be able to create the technology, but they do not have the client relationships, and they certainly do not have the $148 billion servicing portfolio that we have that allows us to go and take technology, train it on the servicing portfolio, and identify where the market is going, what we ought to be presenting to our clients, and why our clients want to continue to do more with Walker & Dunlop rather than two MIT students with a business plan.
Sure.
My question to you is, how are you advising those students as they come in and say, "Hey, here's my disintermediation business plan on commercial real estate, and we're going to run Walker & Dunlop out of business"? What are you saying to them? Go at it or, hmm, careful. You do not have a $148 billion servicing portfolio. You do not have the client relationships, and those pieces of the puzzle are extremely important to have or to be able to scale off of to be able to win in this space.
Yeah. Well, first of all, I say do not go disintermediate another company. Do not make that your goal or your intent. Go find somebody you can help. And if you can help them, you will have a place in the markets. Now, with that in mind, the things that I hear most frequently are, "We built the thing. Can you help us get customers?" It's like, whoa, hold on. Distribution is the biggest bottleneck you have? Well, then we need to take a step back because my mindset on entrepreneurship is start by testing the market, validating that somebody really wants what you think they do before you go and build anything.
Right now, because the cost of creation has dropped so significantly, and at a macro level, the cost of customer acquisition has continued to rise, we see this interesting point in time where anybody can build whatever they want, but getting those first customers is extremely difficult. I say make sure you have a path to market. If you don't have a path to market, we need to go back to basics and establish that. That market testing as one of the earlier phases of the entrepreneurial development process I think of as being extremely important, something that's always been important, but today cannot be overlooked. That's one of the things that I say most often to established firms. I say, "You've got the distribution. Leverage it because now the cost of creation is near zero." That is a wonderful place to be in right now in this moment.
It also comes back to one of the core steps in the entrepreneurial process that I teach at MIT, Bill Aulet's Disciplined Entrepreneurship, and in that process, we talk a lot about competitive positioning, cores, and moats. Moats can dry up. Moats are great in the short term but may not last in the long term, and that's where the core comes in. That's one of those moats that is unique, important to the business, and that grows over time. I say, what's going to make sure that you have something that gives you a competitive advantage compared to whether it's another startup or incumbent firms or what have you in the marketplace? And I emphasize that a little bit more so now than we used to. We used to say, "Look, go find your niche market segment where you can win.
Make sure you nail that because there's enough customers out there in the world, even if you do have a competitor in the marketplace. We'll figure out competitors after you know that niche market to get started." But today, they need to understand what it is that's going to give them that unique competitive advantage moving forward into the future because if they are to compete with, let's say Walker & Dunlop or any established firm, they'll need to nail that first initial market and then scale from there.
Their ability to develop those moats, that core, that advantage upfront becomes even more important now that the cost of creation has dropped to zero and many established firms could go build whatever they want to build in far less time than they would have previously with small teams that ultimately can have a significant impact on the established firm overall. So that's generally how I think about it, but that also doesn't mean that there isn't a threat to the established firm, because there most certainly is. Simply because most firms today, when I look at what they're doing, they're still working in the same way. What that means is that latency topic kind of comes back top of mind. It takes a long time to get anything done.
The amount of time that it takes to make any decision to have it go between the different people or committees or what have you, that need to sign off on something existing. Additionally, the way that the AI-native startups are operating, the way that they're working today looks totally different from even what an organization would've looked like a year ago, two years ago, three years ago. That efficiency that they are gaining means they may be able to build some wonderful moats with whether it's capital or client relationships or what have you in the short term because of the speed with which they're able to get work done without compromising on quality or the actual effectiveness of those decisions that they're making. I think, it's interesting in that speed has increased.
The question is, can the established firms keep up with what I refer to as that infinite speed limit?
I think about regulated businesses and the ability to-
Oh, yeah
-for, one of the big topics right now that is coming up is Anthropic trying to basically create regulatory capture of Dario sitting there saying, "Oh, we need the federal government to come in here and set up some guardrails here," because he's already in the space and he wants to kind of keep everybody else out. I sit there and think there's a reason that Boeing is the only airplane manufacturer in the U.S. It's called regulatory capture. You can't start up a new airplane manufacturing company and anyone who walks up to you and says, "Hey, Professor Cheek, I got this great idea. Boeing's really slow at going about how they build the new 737 MAX," or whatever the plane is.
I'm going to go use AI to completely disintermediate what Boeing is doing and create better airplanes faster." You're going to look at them and say, "Good luck," because it takes four to five years to get the Federal Aviation Administration to approve anything and your student's business plan, no matter how compelling, is not going to go and compete with Boeing tomorrow. I think about the regulatory capture, like in banking. Someone comes to you and says, as I think about banking and making decisions, we work with lots of banks. I won't pick any of them out as it relates-
Sure.
-to who's really slow and who's fast, but there's plenty of latency that exists in the banking system. But good luck trying
Some of it for very good reason.
Some of it for exceedingly good reason and good luck trying to knock JP Morgan off of its mantle as the largest bank in the U.S. from a regulatory standpoint, from a process and control standpoint, from all sorts of other things. I think about that issue as it relates to what businesses and what areas of businesses are highly susceptible to faster, quicker companies. It makes me think about Cursor. Cursor had what, a $100 million ARR with 20 employees? I think I am right on that.
Yeah.
That makes all the sense in the world in that-
In software.
In software. Exactly.
Yeah.
That makes all the sense in the world in software. The question is, I certainly can't sit in my seat and put my feet up on the desk and say, "No one's going to come along and try and knock us off," because I'm certain there are plenty of. Oh, by the way, I've got competitors trying to do it every day themselves.
Exactly. It's not just about those new market entrants, for sure.
But I guess the question I'd have for you is, you talk about AIDE, an AI-driven enterprise. How do you, beyond taking your executive team and great plug for executive education at MIT Sloan, I know your team will be very happy that you plugged that in there as well.
I'm sure they will, yeah.
Other than taking a team up to Cambridge and sitting around and doing a three-day, let's actually, the other thing I was going to think about that is most people do retreats to turn off their devices and engage with one another.
Yeah.
This is the antithesis of that. This is go use your devices, understand how the technology is going to enable you all.
Yes.
Beyond doing that, how can somebody focus on creating an AI-driven enterprise or an AIDE as you use the acronym?
Yeah. It does all start with literacy, regardless of how the executive team gets that literacy. The way I think a lot about it is saying, we are going to, first of all, invest, not necessarily money, but time specifically, time, that investment of time to say we are going to reimagine our business. Some of that will have to come in phases because as any executive running a large organization thinks about it, they can't just flip the business on a dime overnight. It's a matter of first developing that literacy, doing the tools, training, culture change, all that, getting people to the 10% productivity boost that comes with using the chatbots, if you will. That is a stepping stone.
While that is happening, the thing I always recommend, what I literally just documented in my AI transformation playbook is as that occurs, doing an opportunity assessment to evaluate every opportunity to implement AI in the existing organization based on those four questions that I mentioned earlier. If we do that, we come out with a prioritized list of the things that we should implement AI to do. With the right level of AI literacy and fluency, we should be able to do safely, ethically, all of that.
That is where we start to reorganize, do the organizational redesign that looks at how do processes and workflows change, how do roles change, and ultimately, what does the org chart look like when we are mapping not just humans, but also those AI agents in terms of who they report to, who is accountable for them, whose money they spend, who covers the cost of inference for those agents to run, and what data do those agents have access to? These are all of the different components that I think of as being necessary in this overall transformation. It is not to say that it can be done alone just by the ELT. It needs to be done as a team effort with everyone in the organization contributing. When I consider all of this, it is a matter of literally just sitting down and getting started.
A lot of that can be AI-aided, but it cannot be done by AI alone. This is a people problem more so than anything else. The first thing I always recommend is to sit down, begin that AI literacy and upskilling, whether that is in person with educators or trainers or whatever, wherever, but also to say we are going to go and actually set aside a day. Willy, you know what? Let me do what is probably new to the "Walker Webcast." Willy, I would like to give you one day off next week. You get to pick the day. But that one day off next week, Willy, that day is for what I call tinker time. That tinker time is what anybody can do.
Set aside a day to say, "In my personal life, I am going to go explore all the ways in which I might use AI to learn where those risks, where those failure modes show up," not in the high-stakes environment of like you described, massive commercial loans or otherwise. Rather, in something super low stakes, something going on in your personal life that you could use AI to try. So that when you come back to work on Monday morning the following week, you are really well prepared to see how you can apply what you learned in that day for tinker time to many of the different needs that the organization has to increase the top line, to manage the middle line, or otherwise. That day for tinker time, that is one of the first things I always recommend folks do.
Of your 37 chatbots that are running in the background every day, and you've clearly done plenty of tinkering, what's the one that has transformed your You have them doing different tasks, I'm assuming. You have one that you're saying, "I want to stay up to speed on the latest research on AI," or, "I want to make sure that I'm tracking the frontier models to make sure that when somebody steps forward and there's a review of that model, that I know exactly what's going on in it." That's one. As you think about all of those, what's the one that has transformed your life most dramatically?
Well, there's a few that come to mind, but the one that's simplest, easiest is, whenever I end a meeting, I tend to go right into the next meeting. But I also know that there's a wide variety of things that I probably need to do coming out of that meeting. And I used to just keep a list, and then I'd try to get through the list whenever I can. Now, that work gets done in a different way. When I end a meeting, AI is analyzing everything that happened in that meeting, figuring out what needs to happen next, whether that's a follow-up email that needs to be sent, whether that's a document or a proposal that needs to be put together, whether it's any number of other things.
It's going and at least doing the draft of all that work so that I actually do follow up on meetings. That, to me, has lessened the mental load, more so than anything else, of making sure that I actually follow up on what I say I'm going to. That, to me, is one of the most important things, that I follow up on what I say I'm going to do. And if I can create more certainty that that actually happens, that, to me, is transformative. That's a productivity boost, not a 10x efficiency gain, but that's really where the second brain and knowledge graph that I've developed for all of my data comes into play. That has been truly transformative.
That has given AI far more context on anything and everything across all of my different data sources, anything that I've seen on my computer, any text message I get. All of that can be taken in as context to inform the output of anything that AI does. That has helped me to get much higher quality, not just more speed.
Where do you draw the line, Paul, on what's you and what's technology?
Yeah.
Let's just say that you've got it set up that it's going to send a follow-up email. You're going to come out of this Walker Webcast, and you come out and you say to your chatbot, "Write Willy Walker a note saying, 'Thanks so much for the Walker Webcast. That was a great hour, and I found it to be really interesting.'" A, does it show it to you before it sends it to me? B, when I get it, does it make a difference whether I think that it is just auto-generated by a chatbot that's just writing the email for you, and you did it in 30 seconds, or whether you actually went back, thought about it, and added your personal touch to it?
Yeah. Oh, it definitely matters, and I think that's one of the most important things. I would say two clarifications. One, I'm not telling it to go write a follow-up to you. It knows that it needs to do that. That takes one huge step off. At the same time, I don't let AI write emails from me. It can't send them, I should say. It is not allowed to send them. Nothing gets sent without me looking at it first, because if it's coming from me, it needs to come from me. That comes back to that question, can AI do it? Yeah. AI can send an email from my account. Should it do it? No, it shouldn't. That's where-
It's really interesting because-
That's where I draw the line.
You may have seen it, but Stan Druckenmiller, who's one of the best macro investors of our time, wrote an op-ed in The Wall Street Journal last week about how Secretary Bessent's bond buying strategy is sort of silly. I'm using the term silly. He's basically saying it's not going to do a whole lot. We've got a spending problem. We're spending trillions of dollars more than we are bringing in. If we don't do something about the $40 trillion of debt in America, a bond buying program that has the Secretary of the Treasury buying billions of dollars of bonds isn't going to do anything. The big debate around it is that Druckenmiller has said that AI helped him write it.
He hasn't said whether AI wrote the entire thing or whether he just had it kind of come back and proof it, but there's a lot of question marks about, is this actually Druckenmiller or is it AI? As you kind of peel the onion on that question, it's sort of like, well, obviously he would've used a computer to write it if he'd written the entire thing. Should he have put written on a PC? If he took it and had the computer spell check it, do we want to make sure that Stan Druckenmiller checked the spelling himself, or is the spell checker okay? If he used Grammarly to check the grammar on it, are we okay with that?
But the idea that he hasn't said, "I wrote the whole thing, and AI might have it," to some degree, some people are sitting there sort of going, "Man, I want to know if it's Stan. I want to know if that's actually Stan or whether it's just AI." Clearly, if it were AI posing as Stan Druckenmiller, it has less weight than Stan Druckenmiller saying, "I'm going to put my name on it," and it's in The Wall Street Journal. I think that whole frontier is such an interesting line, going back to what I said previously as it relates to services organizations.
Oh, totally.
Like I talked about KPMG and their fees. One of the reasons we use KPMG is because they put a stamp of approval on our books that says it's KPMG. It's not a bunch of chatbots. It's not a bunch of technology coming into Walker & Dunlop and running roughshod over all of our numbers.
Sure.
It's that KPMG stamp. So if their books are done 90% by technology, and it's just the partner who comes in and goes, "It's there," do I feel good about that? Or would I prefer to know that there's more human involvement? I just think that this whole standards issue as technology takes over more of our world is something that's going to be very interesting to watch and define as it relates to what we put value in, what we pay for.
No question. I see more and more people paying for outcomes over time. Outcomes are the thing that, moving forward, people will want to pay for because this latest and greatest generation of AI, agentic AI, what it really provides is finished work. That doesn't mean that there's not a place for services-based businesses, because there absolutely is, and I would say services businesses actually have a phenomenal opportunity right in front of us. What I would also say is if that's what the latest and greatest generation of AI really provides, then we need not forget those previous versions or generations of AI. Because right now I see everybody jumping to, "Oh, let's use generative AI for that.
Let's use agentic AI for that." When the fact of the matter is that many problems people are trying to solve with the latest and greatest generations of AI don't need the latest and greatest generations of AI, and they're actually far more expensive to do it that way. I always encourage people to think, first, do we even need AI for this? And second, what is the earliest generation of AI that we can leverage to effectively produce whatever outcome we're trying to create? Because that will keep costs down, it will keep risk lower, and ultimately, those are all things that we can generate once with generative AI or agentic AI, and then run in perpetuity at a much lower cost.
I love the excitement, but I also don't want us to over-index on this latest and greatest AI development, knowing that much of what needs to happen in an organization can be done the earliest of generation. That's barely even AI at all today.
Paul Cheek, I'm super appreciative of your time. I'm having you back because this topic is moving so fast that I want to have another bite at the apple on this one with you six months or a year from now to dive in and see what you're seeing happen. Your new book, which was just published, "No One Works Here," to those listening, I would strongly recommend it. I'm just super appreciative of you taking the time in person with your chatbots off in the background-
Oh my God.
-to come and join me on the Walker Webcast.
Thanks for having me, Willy.
It's been great. Thanks everyone for joining us today. Have a terrific day, and we'll see you next week.