Episode 3: How to Think About AI at Work
In this episode of AI at Work, host Rob May sits down with Rudina Seseri, founder at Glasswing Ventures. Glasswing is a fund that only invests in enterprise AI companies, tune in to learn more about how to think about AI at work.
Rob May: Welcome to the latest edition of the AI at Work podcast. I'm Rob May. Today I'm here with Rudina Seseri from Glasswing.
Rudina Seseri: Hello, Rob.
RM: Welcome, Rudina.
RS: Thank you for having me.
RM: We're going to talk about a couple different things today. Glasswing is a fund that only invests in enterprise AI companies, and I believe they're the only fund that is focused on that. So, this is very relevant to listeners today to talk about where that thesis came from and what some of the opportunities are there. Rudina, you and I ran into each other, was it summer of 2016?
RS: Yes, it was.
RM: We knew each other socially, we ran into each other down on Third Street in Cambridge. You told me that you were raising a fund that was only focused on enterprise AI, which at the time, most VCs were still figuring it out. What were you seeing that you thought was a good trend that made you want to focus on this?
RS: By the way, you told me you were building a billion dollar company, as well, just for the record. The whole focus that Glasswing has around AI relates to the fact that dating back to late 2014, early 2015, we were noticing a transition from AI, and more specifically machine learning-type techniques, going from research in academic institutions to industry.
Something was happening in the shift from advanced analytics or the big data-type terminology with a number of startups starting to focus on utilizing deep learning and other techniques, but not in the way of, “oh, I have this cool technique, now where do I look for a problem?” But very much the I'm an enterprise play for cybersecurity, or martech, or just IT, and I think I can achieve a better outcome by leveraging some machine learning technique for better predictives or whatever the focus area.
We were starting to see this pattern around problems being enables and better solved by leveraging AI. So, we went deep into that area, formed a more formal thesis, and that became the genesis of Glasswing.
Said differently, our view became that any enterprise play that was not leveraging some form of, narrow AI, granted, but AI, was going to be legacy from the get go.
RM: Because it was so focused, how did your limited partners in the fund respond to that? Did you have a lot of convincing and selling to do, or were people pretty interested in it? What was that experience like?
RS: You got to love tech, right? We started fundraising in mid 2016, and I kept talking about, “there is an AI wave coming. It's a wave of disruption, truly transformative”. I think at first, prospective LPs were looking at me as if I had multiple heads, "Really, is AI a thing?" and then I love tech, because it's so quick and so transformative. Within 12 months, they're like, "Of course, AI is a wave!"
Fortunately for us, we raised a $112 million fund. It's actually the biggest first-time fund of all of the recent funds done in Boston and surrounding. By that measure, we've done well. I think the opportunity is quite big.
Let's get to the substance, which is what got LPs comfortable and committed. They saw the opportunity. This is not a little narrow research exercise, as I said. This is all-pervasive across the enterprise, in cybersecurity and platform. It's multi-trillion dollar markets that are getting penetrated. It's like saying, will the internet be a big thing? Well, we know that answer.
RM: I totally agree. You and I are co-investors, me as an angel and you as a fund in a couple of deals, and one of the things that I've seen across my portfolio is that these AI companies are building a little bit slower than SaaS companies build. There's a couple reasons I've seen for it. I think the buyers are figuring out what they want. There's workflow behavior change.
What are you seeing on the ground in terms of building AI companies? Are you seeing a similar thing? If you are, do you think it's still going to be worth it from a valuation perspective, at the end of the day?
RS: That's a very good question. Let me discuss the input, and then let's talk about the output. On the input side, what I think is different about enterprise startups that are leveraging AI, are fundamentally three variables. One, you no longer just have the business lead, typically the CEO and the technical co-founder. You have this, I call it a third leg to the stool, the researcher or the data scientists. There is scarcity of that talent. Fortunately for startups, they're not always motivated by the largest source of compensation. They're actually largely motivated by creative work. So a lot of startups are able to draw that talent. But, nonetheless, scarce talent, it's hard to find and there is lots of competition.
Secondly, also related to that piece, they're typically researchers focused on perfectionism. It's a very different mindset when you're trying to get to the minimum viable product. The dynamics there have changed.
Thirdly, the need for data, especially any form of machine learning, whether it's reinforcement learning, supervisors, whatever the case may be, getting the datasets matters. It's not just about the size of datasets, it's about the right dataset for what you need. Getting the right algorithms and neural nets built is the third piece. Those make it seem as if it's taking time.
One way to think of it is that the initial set-up takes a bit more time and effort, but once you have built the neural nets in the product, good luck to a competitor to compete with you, because you've got the machine constantly learning and improving and iterating vis-a-vis a static product. At that point, you have all the advantage in the world. It just comes down to execution, as everything else in life.
As I look at the output, I think the adoption of AI in the enterprise-- and by now, enterprise more broadly, the buyers of tech, software, and products-- I think we've learned to filter. This is why I think having the right VC that knows the space matters. Because if you put a call, it's a normal company, as a startup, you have a very hard time getting someone on the other side if you're trying to sell something. If you're an AI company, no matter what you are doing, you're going to get a lot more leads. Those are false leads, because everybody has a mandate to go figure it out AI.
But it's also a market in the making. So it cuts both ways. So how do you filter through that to get to the right people? But the hunger for AI-powered software is quite there.
RM: That leads to another good point we've seen at Talla. We see a lot of the people that come in and you get on the phone with them, and you say, “Hey, so what interested you about Talla?” and they say, “Well, our bosses said that we need to deploy AI”. And it's like, “Great, well, what kind of AI do you want to deploy?” And then they'll say, “Well, we don't know, tell us what you do, and we'll see if that's something that can help”.
I think it's a little difficult for some of these companies to decide where to start. Do you have any guidelines or anything that you've seen work? If you're a big company and you're thinking about how to dip your toe in the AI water, what do you do?
RS: Well, I think you're a big company and you're trying to solve certain problems. Every company is. Whether you're going through a digital transformation, some sort of data, or whatever the challenge may be. You say, “Okay, I'm going to solve this. Let me see if there are AI-enabled solutions that will help me with that, rather than just look at the legacy players.”
Don't go looking for AI for the sake of AI. It's like me funding a company that has AI for the sake of AI without solving anything. I think that's misguided. If you're going to go through a transformation, if you're going to have a different approach to cybersecurity, if you're going to launch a new product, how could machine learning, how could predictives, who's the best vendor that's leveraging AI, you will see the advantage that way.
Or, if you're trying to build an internal team and capabilities, again, what business problem or technical problem am I solving? How could I do better if I leverage AI in a certain way? Who are the right people to build that with? I'll start with what you're trying to solve first, and then look at AI as an enabler.
RM: I wrote a post at Inside AI a while back about this concept of the Matthew Effect. They normally use this when they talk about children reading, which is children who learn to read earlier, they read more at any given age than their peers, which builds on itself, it becomes this cumulative advantage, they become better and better and better, and then they're just better readers as adults.
I have this hypothesis that AI is going to be very similar, which is even if you're an insurance company or a manufacturing company or whatever thing that you do, if you adopt AI earlier, because so much of it is new and because there is this learning component, not just in the sense of what you traditionally learn when you do something new in a company, the skill sets and everything else, but actual learning on your data, understanding your business, making predictions about things, that it's going to be hard for the laggards to catch up. Is that something that you've thought about?
RS: I fully concur. In fact, I have this expression, “It's very hard to retrofit AI”. That goes back to the crux of our investment strategy at Glasswing, which is, amongst other things, all else equal, if you're an AI native or AI-enabled company, it takes a bit of effort to set up, but you will win at the end of the day, all else equal, because you've gotten a jump start, you're constantly improving, you're able to make better predictions, whatever the case may be.
By the way, that's why I also think that incumbents are struggling a bit with this. Because you have a startup come from behind, you have all the incumbent advantages and disadvantages and dilemmas. But the startup can leverage AI to actually disrupt you.
Applying this to big enterprises that are trying to build their own AI capabilities, I would say go for newer products. Whatever it is on their pipeline, go AI native, rather. And then perhaps try to figure out what to do with your legacy. It's much harder to work with a legacy product in retrofitting.
RM: It's interesting because I think a lot of buyers of AI forget that when you buy an AI tool, the worst it will ever perform is when you first deploy it, right? It's learning over time. It's getting better. I think having the workflows and the setup to be able to do that is not something that a lot of companies are ready for.
RS: It's particularly noticeable with startups. You don't have the data, you just build a neural net. It's going to take time to train it. And that first customer is both going to learn a lot, but also get the lowest level of the product, and then it improves from there.
RM: What are you seeing in terms of startups in the ecosystem? Are there opportunities out there that you wish you saw companies attacking that you don't see? Or do you feel like we're on the other side of the market, which is, there's just people attacking things that aren't real opportunities? Are there too many AI companies? Are there not enough, they're missing opportunities? What's the status?
RS: My joke of the month is, “Thank God for blockchain!”, people have stopped calling their companies blah, blah, dot AI, to get on the AI bandwagon. Now everybody's on the blockchain bandwagon. But, no, that joke aside, I think we're in the early days of AI and we do see a lot of opportunities. Some are real, some are not. I'm hard pressed to find a particular category or market where I don't see the opportunity for disruption.
I'm seeing companies in the martech, sales tech category, around data, even just the ability to connect data to each other. Right now, they sit it silos and have linked experiences. I'm seeing it around RPA, iPass, I mean, across the board. Even in fields that I don't invest. Look at what you can do in life sciences, in drug discovery and gene mapping.
I think the opportunity is here to stay. I think we're seeing the velocity increase, as more research comes out and faster and faster adoption at an accelerating rate of research into industry. I do think you have to worry about some of the large incumbents, the Amazons, the Googles of the world, and what they are doing. More prevalent around consumer category than the enterprise category, but I'm not worried. I'm in the business of separating noise from reality, that sits on my shoulders in terms of what companies we back.
I'm quite bullish. I'm bullish also at the convergence of what's happening around AI with decentralization, with, all kidding aside, with blockchain, and then in some of the AR / VR capabilities.
RM: I know you're a speaker at the conference we're having in a week, probably about the time this podcast goes live, at the intersection of AI and blockchain. That's going to be an interesting conference. There definitely is a lot-- because, as you mentioned earlier, data is such an important point of this-- data control, data sharing are huge, huge things for enterprises. Blockchain and federated learning approaches and stuff like that a lot of times can help, I think can help find companies ways to share data to train models to make their AI better together.
RS: Again, I think on that front, I believe there are certain use cases and market categories where, whether it's security and identity protection around data, especially in light of GDPR and some of the European regulatory forces that have come into play. And then there are others where we still need to solve, for example, how does blockchain take advantage of the cloud, because it's inherently kind of counter to it. You're not calling one node in the cloud, or multiple nodes making calls into the cloud, you're storing multiple databases.
I still think there are some roadblocks for blockchain to become all pervasive, but there are certain use cases where the opportunity of combining blockchain and AI is truly powerful.
RM: Now you're looking mostly in your fund at sort of enterprise applications and everything else. Are you looking further down the stack? Are you looking at AI infrastructure? Are you looking at AI hardware? What do you think the impact of those is going to be on the application level stuff?
RS: I mean, as it has always been, look at what's going on with Nvidia, both stock price wise and what they're doing. And we're also seeing an influx of chip and hardware startups that are specifically targeting AI. So they're trying to go from a world of ASICs on the semi side to actually GPUs that are way more powerful and can handle AI. So I think that's definitely an area of opportunity.
Hardware is tough. It's capital intensive. So at Glasswing, we do make hardware investments, but very, very selectively. The bar is quite high. Not to say that we won't see disruption there. I think we're already seeing it, just at the sheer volume of innovation that's going on, both within the startup world, as well as the large incumbents. But not necessarily something that I expect to see Glasswing investing a lot of.
RM: And because you're in the AI space, I always like to end with this question. There's been a big public debate between Elon Musk and Mark Zuckerberg. Many other people, but those two sort of are best representative of their specific sides, Elon Musk saying, “AI's the biggest risk to future. It's out to kill us.”, Zuckerberg saying, “Ah, I don't think we have a lot to worry about. That's a long way off, if it's a possibility.” Do you have an opinion on the Musk versus Zuck debate?
RS: So for what it's worth, I think Elon Musk's views were actually shaped by Nick Bostrom and his book, Superintelligence, which is a book worth reading, even though it's quite disturbing and spooky in certain ways. And it makes individuals like me be even more thoughtful about the steps we take and what we fund.
My view is, I am not quite on the Zuckerberg, it's all hunky dory type extreme. But I am positive and optimistic. I think when you have “Responsible AI”, I'm coining the term. But what I mean by responsible AI is look, fundamentally AI will bring a lot of currently augmentation and then automation. We will see entire categories of professions either disappear or change so much in nature that individuals in those professions, whether it's legal, medical, who are highly paid and highly trained, will find themselves in a bit of a gap. The new generation will be fine, because they will be properly trained. And then we will see new jobs created that we can't even imagine now.
That shift is an important shift and it, I think, can have societal ramifications. Again, it's 10, 20, 15 years from now. But we need to be thoughtful about-- because when you have socioeconomic classes get displaced, particularly-- and we've not seen this before-- when they're highly trained and well paid, sort of the white collar-type category, there are consequences. So you see discussions around universal income, basic income, and others, irrespective of where one is in the political spectrum, and I'm not trying to weigh in on that. I'm merely saying it's something that we need to be thoughtful about from that point of view, as well as what machines can do on their own. But I think this doesn't mean we shut our eyes and don't take advantage of AI. I think we need to be responsible in what we fund and on what we build.
RM: Got you. If there are entrepreneurs listening in, they're interested in talking to Glasswing, what's the best way for them to get in touch with you?
RM: All right. Good. So Rudina Seseri, thank you for joining us. That's all we have for today. A little bit shorter show than normal, given my international travel. So stay tuned. We have a bunch of other interesting guests coming up. We'll see you next week.
RS: Thank you, Rob.
Subscribe and share AI at Work with your friends! If you have feedback or questions, we'd love to hear from you at email@example.com or tweet at us @talllainc.