Full episode transcript: AI and PE: The Future of Value Creation

Guest: Adam Silverman | Host: Kyle Roemer

Adam Silverman: What are the applications my team’s using every day that, if I were to build an agent on top of, it would be a 10X improvement in my business? If you think through this lens, I think you’re set up for success for the next decade.

Kyle Roemer: Public software stocks have lost over a trillion dollars this year. Three hundred billion vanished in a matter of days. Everywhere you look, the question’s the same: Is SaaS dead? Welcome back to AI and PE: The Future of Value Creation. Today we’re talking about whether AI is fundamentally disrupting the SaaS model and what private equity firms need to do right now. I’m joined by Adam Silverman, one of the AI leaders in the space. In these episodes, we usually focus on three things you need to know for AI and PE, but Adam and I spend a lot of time in the market, and as we sit here in March of ’26, there’s one story that’s dominating the AI and PE landscape. That’s AI’s disruption of SaaS businesses. We’re going to spend a lot of time talking about just what is happening in the market today, both in the public market and, importantly, in the private market.

So let’s start with a little bit of background here. The question we’re getting from clients every day is: Is SaaS dead, and how much is AI going to disintermediate the products that we’ve been building and investing in over time? And I think that’s the right question. Is SaaS dead? But I think it’s more nuanced than that. Adam, what are you actually seeing in the market right now?

Adam Silverman: By no means is SaaS dead. I think what you’re going to see over the coming months is that there are going to be companies who just go completely AI first. And when I say AI first, I don’t just mean having a chatbot on your website or something like that. I’m saying: Are you going to be able to have an MCP? Are you going to have an API that’s agent-accessible? Are you going to be able to take your data and connect it to the bleeding-edge companies like OpenAI and Claude and all these different companies that exist out there? When you think “SaaS is dead,” these are the ones who are not going to die. But if you have legacy players who are putting up gates, making it very hard to access their data, and making it very difficult for people to build and use AI on top of their data, that’s where you see huge opportunities for disruption in the market.

And those are maybe some of the SaaS vendors I would consider more in that dead category. So it is really this transition. Across every industry, people are adopting OpenAI, they’re adopting Claude, and they’re really trying to see how they can revolutionize their workflows. The reality is there are legacy systems that hold a ton of data, and if you can’t access this data, make sense of it, and actually do something with it, that’s a problem. As a person who’s selecting a new vendor, that’s something you should be thinking through: Is this company going to allow me to go and utilize this amazing technology directly against the data that I’ve already been storing within a legacy system for years?

Kyle Roemer: I think that’s right. If you’re a SaaS business, whether it’s horizontal or vertical, and you’re not allowing your customers to get access to data, you’re dead. I think that is absolutely right. And it’s so funny, you see this in vertical software over the years, whether it’s construction, healthcare, et cetera. So many of those vendors that had sizable market share wouldn’t really let you get at the data. It was their defensibility, and it made it much harder to switch. In today’s world, if you do that, you’re going to have a hard time maintaining that customer base, given how quick it is to spin up some of these newer applications.

Adam Silverman: One thing I see on Twitter, everybody’s like, “Oh, do I need Salesforce anymore? I could just vibe code my CRM.” And the quick answer is, unless your CRM is absolutely tiny, please, if you’re listening to this, do not do that. But what I do think through is that there are these new incumbents coming into the market.

Kyle Roemer: Yeah. I think it’s fascinating with some of the upstarts. On the point of “are you going to vibe code your replacement for your ERP or CRM?” Sure, you could vibe code something, right? But building something with a minimal amount of features was never the hard part in software. It’s all the other stuff. You’ve got to sell it, you’ve got to market it, you have to support it. But importantly, it’s what you build. Defining those things and making sure you’re building the right thing that, to the point you’re making, is data-accessible, API-first, LLM-first. Those are going to be the differentiators moving forward.

Adam Silverman: For sure. And I think it’s interesting. On one extreme, there are startups in San Francisco who are just live coding everything, going at everything like, “Okay, day zero, I’m going to have my agent connected to my own custom CRM and my own ERP.” And I’m like, that’s great. It’s very cool that you’re able to do this, and do it really quickly. But at what point does the scale actually get to a certain level where this stuff just breaks completely? So as companies are exploring this technology, we should be realistic. As great as the user interfaces look, Google AI Studio just came out with this way to build visual interfaces. You look at it and you’re like, wow, these are really beautiful websites. What used to be a $10,000 website you can now generate for a matter of dollars.

And it’s crazy to see, but as great as it looks, what is actually going on behind the scenes? Some of these players emerging in certain categories who are saying, “Hey, day zero, we’re going AI native,” and have been building the infrastructure behind the scenes to support more enterprise-grade workloads, that’s where massive opportunity exists. And that’s where I think the biggest disruption over the next six months is going to be with SaaS overall. Companies are saying, “Hey, we are going to allow everybody to connect to this as much as possible.” So I’m really excited to see how buyers and different firms start influencing their decisions based on these metrics. And I think you’ll actually see it baked into contracts too. When you’re going through your contracts: “Hey, am I allowed to query this with an LLM? Am I allowed to build my own layer on top of it?” Because every company has specific needs, and if you have great engineers internally at your company, or pay for consultants to build custom solutions for you, you want to make sure you have access to that data.

Kyle Roemer: Yeah, that’s absolutely right. It’s one of those things today where there are a lot of questions around the defensibility of these businesses and what your moat is today versus your moat in two or three years. Those are all reasonable questions. I think the reality is that fundamentally, revenue predictability is changing. It’s going to be easier over time to switch vendors, especially as these AI-native vendors you’re describing allow access and allow switchability. But the other piece is just how you price these things moving forward. The idea of seat-based pricing feels like it’s gone at this point. That’s one of those things, like “SaaS is dead”: well, seat-based pricing is dead, and if it’s not dead today, it’ll be dead sometime from now. So I’m curious: there are product features that are much easier to produce today, but then there are the economics of these pricing models, especially in the age of token usage and all of the foundation model support in these products. What are you seeing that’s interesting right now from a pricing perspective?

Adam Silverman: I’m going to be a bit contrarian here. I don’t think seat-based pricing is necessarily dead. If you want access to software, at the end of the day, software costs money and has fundamental hard costs associated with it. What we’re starting to see is this big movement on the variable cost piece. You see it with Notion, you see it with Perplexity, you see it with OpenAI, with Anthropic. All these folks have this usage-based pricing. A lot of people who are probably listening to this have access to Claude, maybe pay for the $20 a month plan, and the second you run out, there’s this little thing that comes up, right when you need it most, that says, “Hey, you can actually start paying per token or per message.” And I think there’s going to be more and more of this. I’m using AI to the nth degree possible. Every day I’m really pushing myself on how much I can use it. As people adopt it more, you’re like, “Wow, I can’t imagine actually working without it.”

So I think you’re going to have this ability to charge way more for usage, and you will see these power users emerge. Where you might have had people logging into a SaaS product 10 or 20 times a month, you might now have people logging in 10 or 20 times per hour, really pushing as much as possible. It’s been interesting to see how companies are changing, where they might have had 600,000 queries in ChatGPT a month six months ago, and now they’re at six million queries per month. So you are seeing usage change quite a bit, but I think there will be some type of blended form. Some companies are going to the extreme end, just paying for usage, and there are definitely business models to be made there if your product’s that sticky. But I don’t think seat-based pricing is going anywhere. There will just be these new levers to pull in a lot of regards.

Kyle Roemer: Yeah. I think a fascinating emerging area is: How do you sign up for outcomes or efficiencies or cost savings as a software vendor? That’s a bit of a new paradigm for a lot of these players. If agents are going to be doing X amount of the work moving forward, based on the software product and the series of agents you provide, how do you actually price that? Is it based on usage, or is it based on some quantifiable outcome generated by the actual products themselves?

Adam Silverman: Yeah, everybody’s thinking about this. I don’t think there’s a silver bullet quite yet. It’s going to be exciting to track over the next year or two: Is this a moving target? Are we paying for outcomes? You can have an incredible research report created by Claude in a matter of minutes, but what do you do with that? Is it actually going to be useful? As great as it is to create these artifacts, you still need somebody to implement it, and you still need a team to put it into action. As great as agents are, and all these things coming out now on the open source side, there is a very big need to still have humans who say, “Okay, great, I have this information. Now I’ve got to go do something with it.”

So a lot of people are trying to do outcome-based pricing for AI, and it’s very hard, because at the end of the day, if you don’t have a human as an actuator to take this information and do something with it, oftentimes it’s just going to live and breathe within your chat window. This is something we’ve got to continue to watch and see how the world shapes up around it.

Kyle Roemer: Yeah, agreed. The level of customization, people monitoring and actually consuming this stuff, I think is important. Let’s talk a little bit about moats in general, and maybe tie moats into how SaaS businesses are rethinking finance and back office as a part of this, where there could be pretty substantial cost. For software providers and product teams, where do you think defensibility is moving forward? Is it their data? Is it their openness, to your point earlier? What do you think it looks like?

Adam Silverman: Well, the great thing about all the legacy players is that they have the distribution moat from day zero. So when people say SaaS is dead, it’s like, no. To be blunt, Salesforce is not going anywhere. What we need to realize, though, is that defensibility long term will come down to: Can I replace this with a lower-cost solution that now allows me to access my data freely? If you have developers on your team, or people who are really trying to push on the AI side, the long-term opportunity comes from saying, “Hey, I’m not locked into just their ecosystem, and I’m not just locked into the types of agents that they’ve prescribed and said I can use. I can really develop whatever I want on top of it.”

It’s going to be very interesting to see how this plays out. The moats are eroding in a lot of regards. If you have good scale, good distribution, a good product, and have made it really easy for developers to build on top of it, there’s going to be more competition than ever. The cost of building software is coming down, there’s no beating around the bush there. So we’ll see how this all plays out. I also think the moat comes in regulated industries. Anything legal, healthcare, finance: these are things you just can’t vibe code quite yet. Long term, some of these players that have all the compliance certifications, the ability to scale, and the ability to ingest this type of data, that’s actually one area that’s very, very interesting. For anyone listening, if there’s going to be an opportunity for expansion, it’s asking, “Can we go build something that becomes an incumbent in a very regulated market?” That’s a huge opportunity right now, because you’re not going to have some 20-year-old kid hack something together on Lovable or Replit or Claude Code to compete against it.

This is one thing people should be thinking through as they look at their vendor stack right now: Is there anything here that could pose a risk to my company? Is there anything here where, if we were to vibe code or build something custom from the ground up, or go with an incumbent, it could put our company at risk? That’s the way I’ve been thinking through it.

Kyle Roemer: Look, we’ve talked about the risk here for SaaS businesses and some of the upstarts, and where the opportunity is. But let’s talk about upside more broadly for these businesses. A lot of SaaS and tech businesses today are scaling without nearly the same amount of headcount that they had historically. Where do you find some of the biggest opportunities, whether it’s in finance, back office, or elsewhere, for these businesses to scale without the same cost burden they had historically?

Adam Silverman: My biggest recommendation is that if your team is not using AI every day, that’s a problem. I saw this amazing tweet, and I hope I can find it, where this founder talked about how he just turned off Claude, and basically anybody who didn’t message him within that period got fired. Obviously that’s a very rash way to do it, but if your team is not dependent on these tools right now and not using them every day as part of their workflows, that is one of the biggest red flags to me. Everybody I know is so ingrained, trying to run as many prompts as possible, trying to get as upskilled as possible on this technology. The biggest risk is that you have employees who are just saying, “We can go as is, the world hasn’t really changed.” Because fundamentally, the world has changed. The way people are using their computers every day has changed.

This is one of the biggest things companies need to start thinking through: Do we have the education plan and the transformation plan in place so that my employees are empowered as much as possible to use AI? And that they’re not scared of it? They’re not looking at it as, “Oh, if I tell my boss I’m using AI, I’m going to lose my job.” You need to have a company that’s very, very receptive to this type of feedback.

Kyle Roemer: Yeah, completely agree. It’s an interesting tactic to say, “Hey, if you haven’t messaged me with Claude turned off, you no longer have a job.” But look, if you need people to start using this stuff, that’s one way to do it. And I think the more you have it embedded into your workflow, the more effective you’re going to be. What you can do now with a small team versus what you used to do with a large team is completely different today if you’re AI native in your workflows. Well look, let’s wrap up with a little bit of rapid fire. You’re advising a PE sponsor. Where are you having them focus in one of their SaaS businesses tomorrow?

Adam Silverman: The biggest thing is just making sure their product has the ability for AI to query it, the ability for AI to connect to it, and making sure developer docs are very strong. If you’re a SaaS vendor right now and I’m querying an LLM to build a new product, I want to make sure your docs are readily accessible so I can build integrations directly into your platform. A great example of this is Netlify. They allow you to publish your website, and almost every LLM will say it’s the easiest way to host your website. That’s because their docs are so machine-readable and accessible by the LLM. People talk about LLM SEO, and this is one of the biggest things companies can be doing: making sure your documentation is in order and making sure you have ways to connect to your data.

This is very low-hanging fruit. If you have a product team right now, you can get them on this today. There’s nothing stopping you. There are no technical breakthroughs that need to happen. Just make sure your docs are LLM-readable, there’s a robots.txt file in place, and there’s something people can ultimately build on top of.

Kyle Roemer: Yeah, that’s great. This world of LLM-based purchasing decision-making is so different than just going to certain search engines in the past. Well, great. In wrapping up, I think one thing we’ve agreed on is that SaaS isn’t dead, it’s just evolving, right? The winners here are going to be the companies that are data-first, accessible, and building AI features that leverage the best of LLMs. What are some parting thoughts for the audience on where SaaS is today and where it’s going?

Adam Silverman: The biggest thing is: go look through the contracts you have right now with every vendor you work with and see, are they allowing me to plug all the data they’ve been storing for years or decades into LLMs? That’s a good conversation starter with every vendor you’re working with right now. If you want to set yourself up best for success in the next couple of years, you’ve got to make sure your data warehouse is perfect. You’ve got to make sure your data is in a place that is machine-readable and accessible, and you’ve got to make sure your vendors are allowing you to do whatever you want with the data you’ve been storing there. That’s my biggest recommendation for everybody going into this. And for any CTOs listening, literally think through: What are the applications my team’s using every day that, if I were to build an agent on top of, it would be a 10X improvement in my business? If you think through this lens, I think you’re set up for success for the next decade.