The ROI on AI is obvious in these industries

0:01Hi, everyone. I'm Rana el Kaliouby, host of the Pioneers of AI podcast. And I'm here to share some exciting news. Masters of Scale Summit is returning this October 20th to 22nd in San Francisco. If you've been to Summit before, then you know that it's unlike any business or tech conference in existence. The room feels really intimate, the attendee list is highly curated, and the stage program is unparalleled. I've experienced it from the audience and from the stage, where I sat down with some of the most exciting voices in tech for a special live episode of Pioneers of AI. This year, I want you to be in the room with me.

0:38Join us. Visit mastersofscale.com slash summit.

0:50The business world is moving faster than ever. And when change hits, we need to learn in real time. On Rapid Response, you'll hear candid conversations with CEOs and leaders making tough calls about AI, human talent, responsibility, and the bottom line. How they navigate uncertainty, pressure, and high-stakes moments. I'm Bob Safian, former editor-in-chief of Fast Company, and I'll be your host as each episode breaks down what you need to know right now. You can find Rapid Response wherever you get your podcasts.

1:38We back a lot of founders, and one of the questions we ask is, are they open to advice? and in an environment like right now where capital feels like the commodity of commodities how do you assess and how do you build that trust up front so that they know this is not just a transaction there's more that comes with it what is the biggest mistake you've made as an investor um i love all my companies and anything you know about investing tells you that you need to focus on the winners. It goes counter to how I'm wired. That is a constant struggle. I've gotten better at it, but I do not do what I do.

2:24What is the one that got away? Depending on the day that you ask me, it's perplexity because we saw it many, many times and passed on it because it was really a thin layer wrapper and we were looking for more of a mold. And that's why you said depending on the day you ask me because. Because there are days where you're like, yep, I called it. Okay, great. I don't feel so bad. And there are days where you go, oh my God. That was Rudina Cesarri, founder and managing partner of Glasswing Ventures, an early stage venture capital firm investing in AI-powered companies. From AI automating prescription fulfillment to AI that's reinventing cybersecurity, Rudina has always been ahead of the curve.

3:05Today, I'm talking to Rudina about the AI investing landscape, where she sees the real value in AI and what she looks for in founders before she writes that first check. I'm Rana El-Khalyubi, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

3:30Rudina, welcome to Pioneers of AI. Thank you for having me, Rana. I am so excited for our conversation. First of all, congratulations, because you and Glasswing Ventures were named as one of America's top venture capital firms of 2026 by Time and Sadista. Thank you. That is great. Lots of hard work, but you know, backing the right founders dollars in returns out, but along the way, it's nice to see this. Yeah, that's awesome. So I want to start with a personal story. Okay. This is about six years ago now. It was the fall of 2020. So it was during COVID. I took a walk by a river that's close to where I live.

4:07And I was really thinking about it. It had been 10 years since I was running Affectiva. I know where this is going. And I was like, I just need some advice. Who can I call? And I texted you on my walk and I was like, Regina, do you have a few minutes? And you were like on the spot. Sure. Call me. At the time I was contemplating whether to raise an additional round of funding for the company or explore exits. We hopped on a call and I asked for your advice on what to do. Do you remember that? I remember. I know exactly where I was sitting as well. Yeah. I turned one of the bedrooms into an office.

4:42So I was literally in front of the computer sitting at a desk. Yep. Do you remember what you told me? I did. I said at some point, make a call. You know, you can keep going and going. But in so many words, make the tough decision and get it done. Yeah, that's exactly. That's about right. You were very, I think the thing that struck me is the clarity of your answer. Sometimes when people give advice, it's like wishy-washy. And you were like very Regina style because I've got to know you fairly well since then. And yeah, you just said you've been doing this for 10 years. It's time for a new chapter.

5:17Just get it done. It really helped clarify what I wanted to do next. And I think it was particularly sort of evident to me at the time because you had done all the hard work. It was just, you know, a question of had you taken it far enough? And, you know, it's your baby. I know a thing about starting a firm, right, with Glasswing, and it's very, very hard to let go. But you hit the success, time to call it. So do you get a lot of these calls from founders? I do. I usually get them in the context of my own founders or founders that I've backed previously that are trying to figure out what to do next.

5:53It's funny because, you know, we back a lot of founders. And one of the questions we ask is, are they open to advice? And in an environment like right now where capital feels like the commodity of commodities, where, you know, oftentimes, especially with younger founders, is a one-term negotiation, valuation. being able to sort of how do you assess and how do you build that trust up front as I talk about paying it forward so that they know this is not just a transaction there is more that comes with it and invariably you see that more pronounced in a positive way with founders who have gone through that journey first and then know that who you partner with in terms of a backer is not just around valuation and not even just the brand of the firm is actually a lot more nuanced is the person you pick and how and what will they do when they're going to get stuff.

6:47Will that VC be there for you? Yeah, absolutely. Okay, you've been investing for a long time. I am now an investor. I took your advice and now we're both investors. And I believe we're actually co-investors in one company. Yes, yes. And then some of us have a little bit of a stake in some of you. Yes, yes, exactly. You are an LP in my fund. Thank you for the support. I really appreciate it. I would love your take on what it's like to be an investor in today's AI landscape? My sense is like every company is an AI company and every investor is an AI investor. I mean, it's an incredible environment.

7:22The closest sort of sentiment to what I'm experiencing and we're all experiencing right now, it was in 2000. But even then, I think it wasn't as pronounced as it is today. there is a lot of innovation that's going on so with AI building is no longer the challenge building is not the mode we can all build I sat down two weeks ago wanted to create an app or an artifact or you know whatever we want to call it these days an agent whereby you know I could see all the audition opportunities that would come in Boston I wanted a type format done you know right that's not the advantage so when you had talent that was building you had a different type of mode today i think the advantage is more around the data and the access the totality of building systems because i think models and building in and of their own can be commoditized if we've seen anything is that you swap one model out and you know you place the next but it's the data and architecture combo and to what purpose and how deep can you go in that space.

8:29So tying it back to the founders, a lot of demand now for the infrastructure layer, particularly around GPUs and what sits on top of that, you see tons and tons and tons of those companies and they're able to raise a lot of money. You have to decide, am I in for the journey or not with very little data. But also you don't have the 18 months from now we will be able to call it one way or another because races are 50 and 100 million and everything is sort of a goal of setting up a lab. And I'm being a little euphemistic. I don't think everything is the goal, but many, many labs that are being funded.

9:11So the time for the winners and the losers to sort of emerge is going to take longer. So it's a very different model. It's also very different from what you and I do, liking to be the first check-in. How do you become the first check-in, roll your sleeves, help build the company when their first raise is 30 million? The economics don't work. So we're thinking very carefully how we tackle that. And I don't think anyone has a good answer right this very second, but I think we have ideas on how we go very early and make it so easy and bespoke for founders to build the business faster than they otherwise would, that we might be able to get into that incubation stage.

9:54You know, this is interesting because I was doing kind of a look at our portfolio, and it's almost like bifurcated into two like buckets. One is typical pre-seed and seed stage. So the valuations are under 15 million. We come in early. The companies are early. And then we've made a decision to also invest in these like horizontal AI, bigger physical AI infrastructure. Their early rounds are like, you know,$100 million, a billion dollar valuations. And we're a small check, but we still decided to do these because I think there's a lot of upside. How do you think about valuations and how do you evaluate?

10:32They're important. They're not everything. It's not really how I think about valuations. You want to be part of the bigger journey. So the way that I think of it almost like different products, we have this access check where we've gone into some of the biggest robotics companies around town and beyond. So in the physical AI space, we've invested in unconventional AI and recursive AI and a bunch of other sort of 100 million, 200 million type raises of 500 and some instances where we want to have a seat at the table because for access, not just to the deal and potentially follow on capital with our LPs, but also the ecosystems that spin out.

11:11It's the whole community, the whole machine. So that's one piece. The other piece is the redefinition of seed preceding, incubation formation, whatever flavor of the day label one wants to give it. I'm finding that I'm pretty convinced that the seed round now represents the old A. The old A was 8 to 10 million. Many, many, many, many of the seed rounds are in that range, maybe 6 to 8. And many others are just raising 30 and 50 right off the bat, right? So too early stage, it hasn't come to where the seed is, and the seed has gotten later stage, and no founder wants to take capital in between. This is a very interesting sort of dichotomy.

11:54And so by design, you try to tap into both, knowing that you better be investing in a company that's de-risked enough if you're writing a bigger check. And then in the smaller ones, that you will do a lot of sweat, blood, and tears, heavy lifting alongside founders. And those are fun. It's just that you have to do a lot more of those because the failure rates will be higher. I don't know if you saw this program TechTrack that came out of... Yes, John Warner. Yeah, exactly. I'll just give a kind of a summary of what it is. They basically selected a number of students from MIT, Harvard and Princeton and sent them to the West Coast to spend the summer there.

12:31They all lived in a house together and kind of built their startups and they came back, did a demo day. And some of them were really interesting. And we were starting to meet with them to see if we want to invest in them. I would love your thoughts on this because most of them are undergraduates or just like freshly out of school. So very little business experience. And part of me is excited to back these young kind of super ambitious talent. But I have question marks on their commitment to the project or the startup because some of them are kind of still kind of maybe taking a leave of absence from school, but they might want to go back as a parent.

13:09I'm a little biased. I would love for them to finish school. Have you backed? Oh, yeah. I mean, on my way here, I was talking to Tuna, his name from Provenance, and I think he's a junior and he's dropping out and he had like a full marriage scholarship and they're basically trying to become the platform for analytical financial models presentation. So a really interesting take with their AI. It's funny you say that because the first thing I will say is, oh my gosh, you're poor parents. So it's just that notion. Or not. I mean, if they do great, amazing. Not everybody's motivated by money though.

13:47I grew up in a culture, I'm Albanian, where education was your most precious gift because no one could take it from you, especially in a country like mine where they had communism, wealth was taken, Then what was private became government owned. So some of us are wired in a way where education is closer. I'm getting goosebumps because I have the same wiring clearly. So I put that aside. So I have a fundamental view that twofold. One, there are some people who are special that even if they don't have the answers, no matter the age, they will somehow figure their voice and figure it out. There is also a mentality that we've worked very strongly to overcome.

14:3310, 15 years ago, Boston-based VCs, I would say, they backed professors, and the West Coast was backing students. And if you think about mobile and a lot of the application ecosystem, really the innovation, the transformation was driven by the students rather than the professors. And I think that has something to do with naivete. So I always sort of go back to the Uber example. You know, the kids, I'm in Paris, it's raining. Why in the world can I not call, you know, a car or a cab, which became Uber, on my phone? As basic as that. And the barriers that came about that they had to overcome, they tackled along the way.

15:17So there is something about that naivete of I don't know too much to know that it's done this way. That has to be balanced, of course, with driving adoption and the hustle. And we go in and out of waves as backers. The downside is I hope that they know how to quickly get adopters to embrace them because there is value in the relationship. In fact, domain expertise, especially when you're talking about a vertical domain, domain expertise and depth of trust and depth of understanding of workflows could be one of the few modes that remains in AI-started companies. I'll be right back with more of my conversation with Regina, but first, a quick break.

16:38We'll see you next time. who've changed the game. It's anything but business as usual. Find Masters of Scale on Apple Podcasts, Spotify, YouTube, or wherever else you get podcasts.

16:56I know a big focus of your investment thesis is vertical AI. My definition of vertical AI is basically AI-ifying legacy industries like supply chain, manufacturing, financial services, insurance, construction, all these like unsexy industries, but where there is an opportunity to reimagine how work is being done. Where do you see AI creating value in these industries? And maybe the best way to do this is if you can give us some examples of companies from your portfolio. So I will give you a one-two-punch answer, I think, as I contemplate it. First is there is a low-hanging fruit opportunity set across these verticals, and that is to drive productivity orders of magnitude higher by way of understanding workflows.

17:43I think the easiest, and I'll do that in air quotes, path to creating value is to going and saying, hey, pharmacy fulfillment is being done in this manner. There are five different systems that have been stitched together, different software providers. Human touches the prescription, for example, in every step of the way. It gets digitally faxed. They have to take it, type it up in the other system, et cetera, et cetera. So that, to me, feels like low-hanging fruit. But the key is going to be to have access to actually the deeply vertical data that takes place to get the performance. and perhaps even more importantly, understanding the workflows.

18:27A lot of the challenges that we have in different industries is not that it can't be done, is that the workflow is in the mind of somebody or in a 600-page binder that collected dust and there have been hundreds of deviations from it. But if you understand it, I think that's a low-hanging fruit to drive quick productivity, enrich an experience, be able to actually deliver the technology in a different way than what it has previously been delivered and make the ROI quite obvious. So I've just described a company called the Safe Find My Portfolio, whereby that's literally what they do. And they give visibility also on how effectively prescriptions are being filled because little known fact, 30 % of prescriptions do not get picked up.

19:14So you've gone to the doctor, you've bothered, they've given you a diagnosis, they've given you at a minimum a treatment if not a cure, and you don't go to pick it up. So there's a whole process and a lot of ways for them that are important. One of the concerns I generally have, and that's like kind of materialized even stronger in the last six months in the vertical AI space, is the frontier labs are moving so fast, but they're also adding capabilities really fast. And so you can imagine how a supply chain company, they can build their own clod agent to do a lot of this work. And then these companies become obsolete.

19:50So I think there are two schools of thought competing right now. One is you go deep and specialize. The other school of thought is what a lot of the agents, sort of in large language models and players like Microsoft with Fabric and OpenAI and Anthropic, whereby they're saying, you can just buy us as the layer and then build the agents on top of us. And therefore, you don't need specialized tools. I think the reality will be hybrid, as is always the case in this world. So we will be, if everybody's coding internally and building their own agents, I have a lot of questions around the long-term sustainability of those agents, the quality performance compliance.

20:34That's why there's still opportunity in vertical AI, right? And not only that, but it also begs the question, how else do these companies that are vertically continue to move up and down the stack, whereby the advantage is not that the foundation model has seen it all. The advantage is the depth, is trust, is governance. There are all these other facets that are very, very important. You know, the other thought I had, and I'll use the legal industry as an example, companies like Harvey AI, right, have developed AI solutions for law firms and they have seen success. But I'm also seeing this idea of reimagining what a law firm, like an AI native law firm, right?

21:18Where do you see the opportunities? And it's an and or is it an or? I can make both arguments with the same level of conviction and be in opposing views because we really, truly don't know. I'm increasingly believing, and this is the second part, so it's a one-two punch. I'm increasingly believing that the company of your future should really start with a clean slate. So coming back to the law firm of the future, it could look completely different. So as you look forward and you look to the future, why can't I have my legal agent be there? and the lawyer just provides the judgment. I'm really focused on this judgment piece because think about agents.

22:01A model, whoever's model you pick, has been trained on all web data and we can use that as the best, at least digital proxy for society's humanities data. If a human had that kind of data, they would have been sort of a super overlord of source, right we would have been way smarter the models are not so and i think it's this notion of you know their neural nets so they're really dealing with the brain not with the mind while we humans as limited as we are in terms relative to the speed and processing and other capabilities that agents and models have we humans actually can draw a lot of decisions with very imperfect and limited data.

22:44Yeah. And that... Models are dumb. Yeah. And that judgment is still human. Yeah. I think for the safety of our species, I think we need to retain the judgment. I'll be right back. But first, a quick break.

23:24So I want to move us towards like more of the horizontal AI and AI infrastructure, but I will start with the governance and safety part. I believe that that is a very important area to invest in, but we haven't made any investments in that space because I don't know if that's something that the labs need to take on. Is there space for companies that sit on top of the frontier models to ensure governance and safe? So I'd love to hear your point of view. So we invest actively in cybersecurity. And I would say the security plus safety plus governance, and honestly, not just digital, but physical security, are kind of coming together under one umbrella.

24:04So I do think that there's high demand for governance. every enterprise is trying to figure it out. Do they go to their established security partners because they are offering a new module or, you know, governance agent? You know, what do you do? So, for example, for my own AI usage, I get so many mails. Actually, no, exactly. I get 45 to 63 mails per hour. It's incredible. And so I need some sorting and prioritizing. And in some instances pre-filling, pre-drafting, and then I go and sort of change it and press send. But that means that the agent gets to look at all of my content, all of my emails.

24:49So we have a company called D2 that's basically providing that MCP security. And we couldn't be in compliance if we didn't have that capability applied. So it's both regulation and safety and security. So I actually think that that's one of the areas that will get a lot of funding and the demand for purchasing it, whether it's delivered as AI service or as software plus service, it's almost sort of indifferent to price almost because, you know, you must have it. This is an existential question. So it's actually a nice area to look into. Yeah. Okay, great. So you have invested in Liquid AI, which is a frontier lab based in Boston, which is awesome.

25:35Tell us more about that and your thesis around like where is the opportunity in horizontal AI? So Ramin Hassani and the amazing Daniela Roofs founded that business out of MIT. Unlike the other frontier models, what Liquid AI and Ramin is doing is actually computing at the edge. so that in and of its own sort of has different ramifications for compute efficiency on the positive they also do something really interesting interesting so when you think about the pipe of any sort of agentic action that gets taken how you use the models the pipe is hungry the same whether they need that full compute or not so the layperson's analogy that i give is it's a concrete pipe and it doesn't change and you're flowing you know data and compute and it's calculating it and giving you outcomes whether it needed to do the whole thing or not that's how it's set up what's amazing about liquid is that it's flexible sort of the notion of liquid so it shrinks if it doesn't need all that compute and it expands as it needs it so by default those two facets really stand out to me because effectively it makes it a lot more efficient and oh by the way for some of the kids of ours who are anti-AI, you know, environmentally friendly and all sorts of other ramifications.

27:00So there, I think there is a real advantage, not just in performance, but what it takes to deliver that performance. So I think there is a wave of opportunity with these big models, recursive AI, again, with self-learning that are pushing into the next paradigm because at some point I do think that we will run out of sort of improvements we can make to the large language models and the next big breakthrough will be a paradigm shift and I think these frontier labs represent the potential for that next paradigm shift. We've made an investment in a company called Odyssey which is a world models company and then we're also investing in several like physical AI and also like, yeah, AI native interfaces too is an area that is so interesting.

27:47So I'm actually quite excited about our local ecosystem and what we're doing with physical AI. I was going to ask you about that because obviously we're both Boston-based. We invest all around the U.S. I think it's the same for you as well. We have, we've always had more of a tendency to stick to the East Coast and we do West Coast investments, but more selectively. Yeah. So what is your point of view? Because there's a kind of a narrative that a lot of AI is happening in the Bay Area. But what's your view on what's happening on the East Coast? So I think we need to acknowledge and then we need to own our strengths.