#124 - Robert Smith: Enterprise Software, AI, Agentic Execution

0:15I have the honor of having Robert Smith on the show today. Robert is the founder, chairman, and CEO of Vista Equity Partners, which he founded 26 years ago. Vista manages$107 billion in assets as of year-end 2025 with a focus on enterprise software. Robert has been named to the Time 100, Time Magazine's annual list of the 100 most influential people in the world, reflecting his impact across business, technology, and society. Today, we're going to explore how his engineering roots shape the way he thinks about systems and value, how enterprise software and AI are redefining how work gets done, and how to lead and invest thoughtfully through periods of profound change.

0:53Robert, welcome to the show. Thank you, Alex. Thanks for inviting me. Excited to be here with you. We're excited to have you. So let me ask you something that maybe goes to the core of who you are. You're an engineer at heart. How did that early training shape the way you think about risk, systems, and value long before you ever thought about investing? You know, we all are a product of our experiences in many respects. You know, I had the great fortune of starting my career in the early days at a place called Bell Laboratories, which helped me understand a few things. One, the impact of technology on, you know, society at large.

1:33And then as I finished my degree in chemical engineering, I went to Cornell, you know, I started understanding the importance of systems and processes and processes at scale. You know, engineers are focused on, you know, taking discrete activities and creating unit operations that can scale kind of in some respects, I don't say infinite, but in some cases they are, but and how to build systems that are not just scalable, but controllable and can and minimize waste in essence in those unit operations. And so when I think about investing when I think about building things. I want to build them to scale that have sustainable infrastructure, but also have the ability to evolve.

2:25And it's funny, I was actually having a conversation with the senior leader of a consulting firm who was also a chemical engineer, and we were just talking about the nature of feedback and feed forward mechanisms that we are trained on. How do you make sure you get signals back from processes? And then how do you drive signals forward from processes? And to a great extent, investing really is trying to understand changes in an environment and what's the next equilibrium state of that environment and evaluating it. So these frameworks, even though in some respects you think about them in a context of a process engineer, but any market is really an environment that has some equilibrium system that changes over time.

3:15So that's a dynamic that I call it creates a framework on which I use every day. And a lot of it's kind of innate now as opposed to conscious and deliberate, but it's kind of the nature of how I've been built all these years. Yeah. You mentioned chemical engineering, and we know mostly right isn't good enough in chemical engineering. How did that standard influence the way you later evaluated businesses, processes, and organizations? That is a great question, you know, and it's kind of, and it's an interesting one, especially in the world of AI, which I know we're going to talk about today, you know, you start thinking about risk tolerance.

3:53And to the extent, if you, you know, have, you know, parameters that are outside of certain bands, those risk tolerances actually can be catastrophic. And so, you know, you have to understand the nature of the environment that you're in and, you know, to what extent you need to design systems that can operate safely in those environments. You know, in the world of AI, you know, these systems, these LLMs, these, they're by their very nature, they are probabilistic. So they give you a probabilistic output. But in business, often you need deterministic outcomes. Some cases, probabilistic, It's OK.

4:31But you don't want your wire transfers to be mostly right or or your diagnosis for a medical condition to be mostly right. And so you have to in understanding that you have to think about what are the conversion mechanisms to go from probabilistic where you might get massive amounts of efficiencies through utilizing, call it, you know, digital systems, digital workers, AI to deterministic outcomes that can be actually applied in a real world where it matters. And so actually understanding that framework, understanding the framework of your customers as well as an essential part of designing systems.

5:09What is the tolerable outputs and errors that you can that you can actually handle in an output mechanism? And to what extent can you control it? And if you can't control it, do you end up with, you know, systems that can fail because, you know, your probabilistic systems hallucinate or they put put forward a an output that actually isn't either real or isn't tolerable in the output. And so you have to really be thoughtful and conscious about that, specifically in an enterprise, you know, you know, to take it up a level in a consumer environment. If you and I decide to go out to dinner, use one of these models to pick a place that we want to have Thai food in book reservation for aid or whatever it might be, you know, 93 percent of the time it's going to be right.

5:55And seven percent of the time they have a Thai dish on the menu or the restaurant's been closed for two years. Right. And as a consumer, you may not be that challenged by it. But if you're, you know, performing a specific operation in banking or finance or insurance or, you know, in health care or, you know, logistics or whatever it is, that outcome can be catastrophic. And so that's why you have to really understand the nature of the systems that you're operating in and build, of course, you know, processes that adhere to the call it the constraints and the output of those systems. Earlier, you talked about building things to scale.

6:32And I know you learned very early about the impact of technology on productivity and scale. What did you see long ago that helped form your long-term vision? And how has that vision played out over time? really at its core of what our company does and what Vista does is we actually look to manage what I call the transformations of enterprise software businesses. You know, I always maintain that software and enterprise software particularly has been the most productive tool introduced in our business community over the last, you know, 50, 60 years now. And likely will be even this new technology, AI, guess what?

7:13It's software, right? And it's an enablement software. And with all that said, you have to say, well, the nature of value creation is often moving from one state to another. And how do you do that systemically? And so the design points around Vista are, okay, what are the critical factors for success for an enterprise software company to get from state A worth X to state B worth 3X, 4X, 5X. Often there's, you know, a precision that needs to be brought to the operations of the business so that you eliminate waste. Eliminate waste in production, eliminate waste in customer support, eliminate waste in go to market or your selling activities, eliminate waste in the value capture for, you know, what is the value that you're providing your customers and then call it, you know, feedback, which is, you know, what's the information through serving these customers that can inform you to deliver them a higher quality product or service or offering and then build that and deliver it back to them.

8:21And that's kind of the, you think about it, that's the virtuous cycle that enterprise software has enabled for decades now for companies. And so designing an organization that can take that into account and have the design points as opposed to reacting to changes in the environment, you're actually driving changes in different states of that company. so that it is going through a transformation of value creation, eliminating waste and providing products or services that enhance what I call the economic rent opportunity for your customer base. And if you can provide an economic expansion, economic rent for your customers, then you should be entitled to your fair share of it as a provider of those services.

9:12So that's the design point of Vista and what we focus on, I call it critical factors for success being under our control, things that we know how to do. And so when there's an introduction of newer technologies, on the one hand, going from on-prem to cloud, we said, all right, let's build a factory. And that's what we did in 2014. So we need to build a factory to convert businesses from on-prem to cloud. And, you know, the raw material that was needed was compute. And there was only one organization that had compute in the early days, which was this little company called AWS, part of Amazon. And I knew the CEO by the name of Andy Jassy.

9:55I said, Andy, no, look, we want to build this factory, but I need guaranteed compute. And, you know, there was compute scarcity. I said, well, if we convert the businesses, we will put them on your platform. But I need the technical resources to enable my factory to get up and running. I also need guaranteed compute. And so that's what we did. We converted more businesses from on-prem to cloud than really any institution on the planet. And those businesses, on average, are about a three times expansion in multiple as software rerated, becoming more efficient from on-prem to cloud. Okay. Now we this new technology.

10:33And then as more compute became available, you know, through Azure and others, I cut similar deals with Satya and others and said, OK, we'll make those conversions. And that's what actually ushered in an expansion of, frankly, the consumption of enterprise software and the efficient consumption of it, which created the re-rating of software in the marketplace. Now we have a new technology, AI. And so what we did again was build another factory, right? Now we're going to go from SaaS and some cases on-prem to agentic, okay? Implementing agentic solutions in these workflows and data sets and building out the right partnerships to make that happen.

11:14We have access to compute and low-cost compute. So all of that in my mind is really, you know, these frameworks of building institutions and institutions being ours that can scale and absorb newer technologies, manage those technologies to deliver it to our stock, our existing portfolio companies, then it'll flow, but we can underwrite new portfolio companies to that motion. And that's really what the motion is. And that's the design point in which Vista is built. The way you describe that sounds so obvious and simple and straightforward. But when you look at organizations that struggle during periods of rapid change, what would you feel that they most often misunderstand about evolution and adaptation?

11:59Every evolutionary environment has different components you have to think about. Often executives are focused on the instance of change as opposed to the process of change. Here's what I mean by that. You know, senior executives often run one company. And so they're focused on, OK, let me solve that one problem for that one company. You know, we have the unique advantage of having an out, you know, 90 plus software companies today. And we've always had, you know, dozens to 30 to 40, 50. And so you get a chance to see a landscape. And so as a result of that, that informs you as to, you know, the various elements that you now need to understand and anticipate in making change.

12:55And so you just have a different perspective. So where I think executives often make a mistake is they get wedded to a framework, thinking that's the only framework, as opposed to, you know, we get, I call it the opportunity to survey a landscape that says here are different frameworks that can be effective in different environments. You know, let me give you an example. You know, people talk about AI today, and it's like agents. Well, because of the way we built our factory and how we now, I think we have 54 of our companies through the factory. We've got another 20 going through now. We've identified there's kind of four archetypes, main archetypes of agents that likely will, I'll call it, you know, persist in an agentic world.

13:37You know, one's called the fixer that never sleeps, for instance. That's a time and always on agent that acts very different and has a different consumptive color requirement than what we call the orchestration agent. The orchestration agent is orchestrating other agents to do things, which is different than what we call a workhorse agent who actually has to do some very complex things. You know, often those are for governments, et cetera, with broad implications. And then you have what I call the sidekick, which is someone who's an agent that's working alongside a human, right? And each of those have different patterns, different what we call inference requirements, requirements to operate, and have different requirements in terms of what the outputs need to be from them.

14:22If you're only working with one company, you may only identify, oh, gee, that is the one agent that we use or could use, as opposed to knowing and solving a problem for a company, you might need two or three types of agents that do different things. And how do you optimize those for the type of agent as opposed to this is the one thing that I know. So we have the advantage of size, scale, and now capacity across, you know, our, we call it, you know, Vista Agentic Labs and our Vista Agentic Factory to evaluate these things and optimize them across each one of the portfolio companies. And, you know, I would say that those are the advantages that will be long.

15:02They have the effect of compounding the insights that come from a bigger laboratory as opposed to singular instance of experiences that will limit the insights to only the frame of the problem that you're looking to solve then. Something that you just said, which I think is really insightful is, you know, my experience, and it sounds like your experience as well, is oftentimes people tend to zoom in to the problem that is right in front of them. And they don't really have the perspective to zoom out, see the bigger picture, and recognize that there's multiple paths to the solution rather than trying to force their way through the path that they think is the only way there.

15:47I think you have that perspective because you work with so many different companies that are successful in different ways. So you see different ways of achieving success rather than trying to force your way through a framework that you think is timeless and universal. Yeah, that is, I think, a great observation and is spot on. You know, I was talking with the CEO of a very large global bank, global banking platform. And I've been friends for a while now. And we were talking about the efficacy of them looking to implement, call it agentic solutions in their organization. And we were just comparing notes and KPIs.

16:26He's like, man, we're just no one near seeing the productivity. And he said, I'm spending billions and I'm not seeing the productivity. I said, I understand. I said, you have to understand I've got 90 laboratories. You have one. Right. And because I have 90 laboratories, I have the ability with my team to glean insights that actually can be informative in other parts of the ecosystem. and that you know if you think about you know structure of of revolutions you're often better with 10 design experiments than one right um that can provide you with again these these insights that then you can you can replicate uh and expand much more rapidly so in some respects you know It's just the design dynamic enables us to do that.

17:22And now, which leads, frankly, and all honesty, what that does is it actually entices partners. Biggest hyperscalers in the world we are partners with, and they reach out to us because they know we can test and we know we can evaluate. We can come back with real-time information about either their model or their deployment or their technology or whatever it might be as it relates to either a specific industry or a specific type of an agent. And we can do a contrast to compare. So it's kind of a really unique place to evaluate that for the enterprise, because ultimately it's going to be the enterprises that will drive the profitability of those hyperscalers, because they ultimately will need to provide inference, call it the power, the compute to these agents to do the work that leads to the productivity that's been touted in the marketplace.

18:18And I guess if you have great success, that can also hamper you because you can become dogmatic in your approach because it has worked for so long. And then when you live through a world of rapid change, it can be potentially more challenging to try to find a different path to success. That's why you have to always have as part of your culture the capacity to evolve. You know, you mentioned a little earlier about what are the things that in some respects, you know, paraphrasing, you know, inhibit executives from evolving or progressing or getting dogmatic on it. It's often because they say, well, this has worked, and so I'm going to stick with that, as opposed to understanding that you have to have a posture that evolution is the natural state of things, not fixating on a certain paradigm that does not evolve.

19:19Like I said, those paradigms will often work in a certain equilibrium state, but those equilibrium states change all the time. I mean, last four months, everything's changed in and around the perspective of enterprise software. OK, that doesn't mean that enterprise software has changed in its utility, but there's been a perspective change, which we actually have to know that, OK, that evolution. And you now have to take an approach to inform, like we're doing now, the investing community that guess what? Less than 1 % of these models have been trained on enterprise data. Less than 1 % of enterprise data is accessible by these models to be trained on.

20:05So, and Satya Nadella talked about this at Davos this year when you're saying, listen, companies have to understand that if you take your data and your workflows to these models, those models are designed to absorb, in essence, your intellectual property, now you're leaking enterprise value, as opposed to bringing the models to the data in your environment, your unique workflows. And even though two companies might be in property and casualty insurance, they underwrite differently. Okay. And so they have different ways of doing things, different policy management, risk management, you know, fraud detection.

20:45But if you let those models in essence, you know, understand how you do all of that in a general sense, then you're giving up enterprise value as opposed to bringing those models and operate in your environment where you can compound your capability and your competitiveness, as opposed to diffuse ultimately your, you know, what is your intellectual property. But what I'm saying is you have to really understand that you have to focus your culture on embracing evolution as a natural state of being and build, bringing the right sort of people in at the right time or partners in at the right time and making sure that your teams, uh, don't fixate on the only way of doing something as opposed to evolving to more natural states of change and natural states of operations.

21:36If you look at the companies that have survived through time, they're the ones that have evolved as opposed to the ones that die a natural death because they stick to the old model that eventually becomes outdated. Absolutely. You know, in the technology world, of course, and everyone is littered with companies who fail to evolve. Um, you know, I, we do this every now and then it's been, it's actually interesting, you know, so look at the S and P or fortune 500, fortune 100 companies over the, and just take a snapshot every 10 years going back. And wow, wow, wow, wow. How I remember that company.

22:12What happened? It's amazing how much it changes. It is, Alex. And you, and then you double click in it and you're like, well, what happened? And often you found there was a senior executive team that got fixated on a way of doing things in the world evolved around them. And they didn't adopt the technology or adapt to the environment in a way that would help them remain competitive. And in fact, often eviscerated what was a competitive moat that they had potentially been known for. And so look, you as an advisor and us as an investor, it's our job, our role to understand what those changes are and to inform, to educate and to anticipate what those changes, the implications of those changes are in a way that we can best inform our clients.

23:05Investors tend to panic and often overreact during major technology transitions. We've seen that in the past. Are there patterns you see repeat every time a new platform fundamentally changes business models in enterprise software yes you know we're going through one of those now it's it's it enterprise software goes through re-rating cycles uh when we went from you know initial mainframe to on-prem okay and there was a re-rating of, okay, well, wow, you know, this big hardware stack and with Moore's law and the ability to distribute compute, um, created a re-rating and saying, wow, this package software, um, can be actually quite valuable.

23:53And so those companies started to, to understand there's an economic rent advantage, uh, that can be captured by selling this software on premise to customers. And you now have to go do that. And then it was another re-weighting. And you remember this when we went from on-prem to SaaS. Remember, SaaS is going to destroy software, right? Remember that. These SaaS companies or these cloud companies are going to destroy software. And every public company said, I'm going to move to a subscription model, went through this massive J curve in the public markets, which gave us, of course, the opportunity to go buy those businesses, manage the transformation and then capture value on the back end.

24:34And then that actually ushered in what was the greatest expansion of enterprise software utilization really in the history of that industry. But it started off with people panicking, right? Oh, my gosh, this cloud thing will just destroy as opposed to it enabled that software. And it also enabled a different business model. I remember on-prem was a licensed model. You licensed software and you got maintenance and there are episodic pulses of, call it economic rent capture for these companies. And when you moved to SaaS, it became seat-based licenses. And, you know, you got a little more linearity in terms of the business model.

25:16And that's what people started thinking, oh, man, this creates a terminal value that I can calculate based on seats. And great. Great. And then, of course, now this AI thing hits and people are like, well, wait a minute, is that going to eliminate the terminal value? Well, for some companies, it will, but for others, it's going to expand it. And so if you think about it, you went from software as a product to software as a service, now software as a worker, right? So agentic systems actually operate within workflows and they are workers. And so you can now start to price against outcomes, not seats.

25:48And if you're priced against outcomes, you're now able to capture an economic rent or value that was here to forward delivered by either services or labor or enhanced activities of what you're, you know, market that you're currently addressing. And so your total addressable market actually expands for those agentic companies. But the market doesn't hasn't seen enough proof points of that just yet. And so you're going through another re-rating. Right. And then in every rating process, there's a downturn, which is where we're in now. And then ultimately, it'll be like, well, wait a minute. You can now capture the services part of the spend or the labor part of the spend that can now be multiples of what you were going to get in a terminal value in a SAS model only.

26:38So we're in that process. I call it the early part of the messy middle of this transformation that on the back end of this should yield massive amounts of economic rent. Multiples, in fact, of what we saw in the SaaS models for certain companies. I think that goes back to what we talked about earlier, which is people think of it within the old model and they see a change and the old model doesn't work anymore. But they don't really have the foresight to see what the new model is going to be because they're too zoomed in. Yeah. And, you know, and you can't blame them on the one hand because they don't live in the space.

27:14And so they have to rely on those who are in the arena. What exactly is happening? What is the utility of this new technology? What is its capability? What are its constraints? You know, the constraint of probabilistic outcome. And what does that really mean? OK, and in an environment where you need deterministic results, you have to now figure out what is, for lack of a better word, the algorithm to go from probabilistic to deterministic, which is housed and encased by these enterprise software systems because they are designed to deliver deterministic output. And so that's why AI enables enterprise software, in essence, to now what I call expand their TAM by eating services.

28:06And that's why the two together are what create the massive opportunity. You've got this new cost, which is called inference. You know, before you had a cost, which was labor and services, but now you've got to actually enable these models to work by feeding them with power or compute or inference or, you know, tokens that affect that. And that's a dynamic that you now have to think about. How do you manage as a purveyor of these solutions, these agentic solutions to your customer base? And I think that's an explanation of what you described earlier, which is, you know, there's a common view that AI is going to eat software and you're saying, you know, it can actually feed software.

28:48Absolutely. Now, you're going to have three states of enterprise software going forward. You're going to have this agentic state where agents are operating within these workflows and creating massive amounts of economic rent capture in the environment. The second state is one where you use this to enable the delivery of software at a much lower cost, i.e. co-development and customer support and, you know, administration, including sales administration and go to market. And so you can actually double, you know, triple, you know, the expansion of the margins of those businesses that may not be agentic.

29:31If they are agentic, it's even more right. You get a bit of both. And so those are kind of states one and two. And the third one is you don't have a right to exist. You know, if you don't actually have sovereignty and dominion over workflows and data sets in an enterprise software solution where your enterprise solution was really designed to gather information from either unique or non-proprietary places, repurpose it into a customer's hands. You know, these LLMs that have been trained on that same information are likely going to be more efficient. So those companies don't have a right to thrive.