Transcript: Yaniv Golan | lool ventures | What to Build Now? Which Companies Will VCs Back?

Eyal David: [00:10] Hi Yaniv, what's up? Good, come on, how are you? Great, so nice that you came. Honestly, after our opening conversation, whether it's great or not, but I enjoy being in your company. It was a bit of an apocalyptic conversation.

Who Yaniv is and what lool ventures does

Yaniv Golan: [00:21] Yes, I've heard a few of those already, a lot of them run through me too, both in my ears in podcasts and between my ears when I think, but yes, a thought-provoking conversation, I think. Uh... Want to tell us a minute about yourself, so people know who's with us? Sure, happily. So first of all, thanks for the invitation, it's fun to be here. Uh... So I'm Yaniv Golan, I'm a partner at lool ventures. lool is a venture capital fund. Uh... We invest mainly in... or only in the early stages, pre-seed and seed. We've been active since 2012, it's a fund of founders, operators, we've all gone down the path one way or another. Uh... And we're a very curious fund, and that leads us to invest in a very wide range of fields. In complete, uh... contrast to what we advise the founders we invest in, male and female founders, uh... to keep focus, uh... as a fund, uh... we make a point of keeping, uh... very high diversity in our investments, which means that in the portfolio you'll see companies from standard, uh, software, uh... B2C, B2B, Enterprise, fintech, uh... etc., through, uh, IoT companies, through robotics companies, through VLSI companies, uh... Medical Devices, Digital Health, uh... Materials, uh... and more and more. Uh... a truly very wide range. Uh... that comes partly from what I said, that we're curious, but also because we really, uh, believe less in being investors who put some thesis on the table and act according to it, and believe more in, uh, interaction with founders who are in the field, discovering problems. We get most excited when it's problems, uh, in fields we don't know at all and have never heard of, and... they come with solutions that come out of that knowledge. Of course they need to have the right team and the right economic potential and all those things, but that's where the ma... that's where the magic happens, right? In the less familiar places.

Eyal David: [02:14] Wow, that's fascinating. How do you know how to... since 2012 too, so it probably works, how do you know how to assess risk across so many different industries when they're at such an early stage?

Yaniv Golan: [02:24] I don't think we know how to assess risk, we try to assess risk. We try. Look, there are, of course, patterns that repeat themselves no matter what the industry. Ultimately a team is a team, and people are people, and that spark in the eyes, which you already know how to recognize, of a founder who's come to change the world, those are true no matter what the industry. And when they're there, they're there, and when they're not, nothing will help, no matter the industry, and no matter the market-size calculations, and what the ACV will be, and the TAM, and the SAM, it doesn't help. Either it's there or it isn't. That's a fundamental thing, I think, in that calculation of risk, as you call it. After that there are all the usual startup mechanisms, looking at the market size and its dynamics, and at the consolidation happening or not happening in it, and when it happens, and what we think will happen in the future. And always in the end, certainly in the more specific industries, let's call them, there's also consultation, of course, with professionals, with experts, and with customers. But it's much harder, I think. My experience, right? In my past I'm, first of all, a technologist, okay? Many years as a technologist, and then a founder, really quite a few years now in all kinds of stages and iterations. But that means I come, at least in my own eyes, okay? with a lot of knowledge and perspective in technology. In the software world, okay? In the software development world. And I find that precisely in the fields where I know best what's going on, with many years of perspective and also with my finger on today's pulse, that's exactly where my decisions as an investor are the least good, okay? As investment decisions. Because it creates bias, preconceptions, all kinds of things. Precisely in fields I understand less, okay? those are the places where I can look at it cleanly, from an investor's point of view and not as someone with a passion for a certain field. Yes, is there one investment where you really didn't understand anything, and that's really what... can you take us through an example like that? I can tell you the opposite, okay? I really think I can, I won't list the anti-portfolio, but there are quite a few examples of companies really from the fields I understand a lot about, okay? where I said no, this won't work, this won't go, this won't... and reality proved I was wrong, and wrong by orders of magnitude, okay? Very respectable orders of magnitude. In fields where we understand less, yes, you see our portfolio is full of examples like that, right? Like NeoLogic, a VLSI company, in chips, okay? that produces an alternative technology to CMOS, or actually quasi-CMOS, which is basically the basis of... today I can already talk about it because I've had time to learn, right? But certainly when I met them I knew roughly what CMOS is, but I had no idea of the implications and its importance within the world of chip design and electronics, and what a technology node even means and what an interactor is, I didn't know any of those things. And that's what lets you look at the people, okay? and see in front of you two founders with a very clear passion to change the world, this world, the world of circuit design, but it's very, very clear, very clear that they come with the credentials, very clear that they come with the knowledge, very clear that every question you raise has an answer that's knowledgeable, that's considered, that addresses the whole picture. Now, that's where the journey starts, right? But it lets you see it clearly, this team that came

Eyal David: [06:01] to really change the world. Okay, we'll continue from there, I just wanted to jump in to state the episode's topic for a second, really: you gave a talk that was the basis for our conversation, which was basically, let's put it this way, about types of companies now, and how the world is changing around the AI revolution, which is basically already fully here. It can also really be connected for a second to how you see companies and which companies come in, and also whether it's just in that eye, whether it's one pair of eyes or more, I really don't know. But come on, take us for a moment, before we start, really, tell us who you presented it to, how it went, and we can start like that. We're basically trying to sum up this deck as the basis for the conversation now.

Where the framework came from: no moat on top of the model

Yaniv Golan: [06:42] So the specific deck you're referring to, I presented a few weeks ago, I think, at a conference of an accelerator called Genesis. The person who invited me to speak there was Dekel, and the title of the event was Building in the AI Age, and he basically analyzed it from several angles with several different speakers. But actually the deck is also the result of thinking and of an article that was published in Forbes, and before that, it actually came from a certain distress we had as investors. Meaning, when this wave started, when the companies started being born, when ChatGPT 3 arrived and changed the picture, and companies that are Powered by started coming in, there was this difficulty of saying, wait, how do we analyze this? You've basically been in this situation for two years already? Of course, yes. And we had to define for ourselves some criteria by which we start analyzing it in a more orderly way. Meaning, I think one of the very clear things that happened is, you meet a company, you meet an amazing company, again, I'm saying, when the wave started, okay? Today it's much healthier, I think. You meet an amazing company, and it does amazing things, okay? And you open the wrapper a little, and you say, one second, but actually the amazing part of it, okay? is built on, there's the mountain of capabilities of, doesn't matter, at the time it was ChatGPT 3, right? There's the mountain of capabilities it brings, and ultimately what you do as a company is some small house on the edge, which is amazing, yes, in terms of value to users, but as an investor I look at it and say, one second, but if this whole mountain exists, with the house on the edge fitted onto it, then basically anyone else can come to this mountain and build their own house. Meaning, basically there's no, as it's called in investor speak, there's no moat, okay? No moat, yes. And then suddenly, it starts like that, after a few companies like this, you say, one second, this is the kind of company that's amazing and can succeed and make lots of money, but I can't invest in them, I just can't, it doesn't fit our model.

Eyal David: [08:39] So you start asking, which companies do fit? Something very significant is happening here, you can't say, I have to ask you for a second, so what happened during that period in '23, say, like at that moment, did you take time to understand it? What was it like? We argued a lot,

Yaniv Golan: [08:52] we really argued a lot. I was the bad guy all the time on the team, you know, like, someone brings a company that they're very passionate about, because it's really very significant. And I was the bad guy sitting there saying, well, but like, anyone can do this now, and there's no moat and there's no moat. A lot of frustrations like that, okay? Really, it was very frustrating for all of us. But you know, eventually you, it starts, the picture starts to clear up, you start to understand that actually they're not all the same, and there are places where it's significant, by the way, even if 99% of the value is the model and your value, what you bring, is something very small at the edge. If that very small thing is based on knowledge that's very unique to you, and very few people know it, okay? or on data, okay? that's very specific, yes, to a certain industry, and even very hard to obtain, then that 1% isn't so easy to replicate anymore. Look, let's give an example, for instance, off the top of my head, Lemonade,

Eyal David: [09:50] so they have a data moat around insurance that they've accumulated over the years.

Yaniv Golan: [09:56] Yes, but I'm not sure that's a very accurate example, because they started before the LLM wave, okay? No, no, I mean, but in the end they have data that, like, good luck getting it now. Absolutely, absolutely, absolutely. Meaning, they basically added the data moat on top of the moat of their specific models, which were born even before LLMs, yes. And certainly since LLMs it's much easier for them to deliver value much faster. So you basically sat in a room for a period we don't know how long, you were the bad guy? No, we sat as a team, and we argued, and we talked, and we thought together, and slowly the picture gets clearer, and then we also started, you know, putting some names on things, saying, okay, if so, let's call it this. I think the whole industry also went through this process, and the names kind of crystallized at the industry level. We as a fund also discovered that we're very drawn to a category that's exactly the opposite. Meaning, when you see all the amazing activity enabled by AI, you ask yourself, wait, but what's going on with the rest? Not everything is, not everything is software, not everything is AI. You also ask yourself, in this world where models come out every, every, literally every day now, yes, and technology changes so fast, are there areas that aren't so affected by this story, that are protected, you could call it, or maybe at a level we'll call risk diversification, okay? I don't want to depend only on what happens in the AI world. So that's how another category was born, at least internally for us, which we call AI Resilient, okay? Those are companies that, honestly, don't care which model comes out, because it's simply not relevant to them, okay? A company like Chiral Energies, okay? which develops nano-coating with very special properties. It's a company doing Deep Tech, okay? It sits in the lab pouring test tubes from side to side,

Eyal David: [11:45] and gets annoyed when the measuring instrument's fuse burns out, and they don't depend on the model. That also really brings us back to what you said, that you're curious and you know how to get in fast, so okay, that's probably also connected to that four-year research thing, of course. Okay. We'll also say we'll put a link to this deck afterwards in the show notes. One second before we continue, click the link to the Product Builder WhatsApp group in the episode description and join the community. There you can continue the discussion, ask questions, and be the first to get updates on new Tachles episodes. We're waiting for you there. And now, back to the episode. So do you want to maybe start telling us about the different types of companies? The deck actually, sorry, let me sum it up for a second because nobody can see it, we didn't do video either, it actually starts with hypes, right?

A revolution or just another hype?

Yaniv Golan: [12:32] Yes, yes. I think really, we ask ourselves and everyone asks, okay? So, honestly, all this AI is cool, but we're already old, we've already been through several hypes like this. Specifically about AI, by the way, this is basically the fourth time, okay? that there's an AI revolution. I really remember as a young kid, a nerd excited about the field, that my only source of information was "People and Computers", the magazine. It's still around for sure, they do conferences I think, but a bit tangentially related, right? But back then it was really the only source of information you could get to. And I remember myself reading about the expert systems of the '90s, right? those wondrous decision-support systems and Prolog that would run on them and replace the human doctor and all the... and of course that didn't happen, right? And then it happened again, we went through it when machine learning started rising, and we started to understand the... In short, it's not the first time, but there were many other revolutions too. There was the dot-com revolution, and the Java revolution, which was supposed to run in every shoe even, I remember descriptions, in every, yes, in every component we touch. And the revolution I was even part of, the Web 2 revolution, right? Web 2.0, which changed the world for a moment, and the Cloud, and e-commerce, and IoT, where, yes, every plant was supposed to come with a small sensor built into it, and blockchain, of course everything has to be powered by blockchain, whether it makes sense or not. And we still have remnants here of the metaverse revolution, the third time already, I think, and quantum, which is arriving any second now, I don't know how many years. So are we now, let's call it a revolution, the GenAI revolution, okay? Is this another one of those hypes where we'll wake up in a year and say, well, it was nice, many companies made money, what's next? You know, what's the next hype we'll market to each other? And that's a question, it could be. If we look at the Gartner hype cycle, we all know what it looks like, okay? and how it feels on the way up, and really everyone is sure, or sells themselves and believes that this is the next thing, and then comes the stabilization, the decline, it reaching its right place. So is that the situation? Could be. I think not. I think this is a revolution that's fundamentally different. I think we crossed a line. We crossed a line, I'll be presumptuous for a moment, I think we crossed a line in human development, or the story of humanity, I don't even know whether saying humanity is the right word here, because we don't know what the next part will be, but we crossed some significant line.

Eyal David: [15:20] I think that... sorry for stopping you, we even talked about it a second before, if you'd like, about GPT-4, the emotions,

Yaniv Golan: [15:28] you want to say? Yes, yes, I think one of the things we touched on in the conversation before was when OpenAI released GPT-5 with great pride and said, that's it, the mess is over, there aren't all those models anymore, and now there's just one model. A day later, the whole world came and said, what have you done? Give me back my 4o, I have a relationship with it, it holds all my secrets, and we have a real dialogue between us that I can't recreate with your 5, which is a bit too cold and a bit, you know. And it took, I think, a few days, not many, I think it was something like a week, and they simply brought 4o back to life, and put it at first as an option, and then it appeared there in the menu, and it's still there. And actually, when you think about it, this is, I think, the first time that not out of a person's design, meaning, in products we'd call viral or addictive, engineers and product people and psychologists sat and did everything they could so that Instagram would be addictive and bring me back there again and again and again, and the same for each of those products. This is the first time no engineer designed it, but the product took care in advance, in some sense, of course it's not, technically it's not like that, but the product took care in advance to create that hook, that attachment to the user, that the user isn't willing to give up, and the user will fight for its life, and the user really brought it back to life, it was already in oblivion, it had already gone down to the backup servers,

Eyal David: [16:53] they pulled it out of there, did a restore, and there it is, back to life. Okay, so we totally agree, we could really dive into these aspects, and it's also super interesting, but we understand that there really is a different revolution here, and we also see it in the business world, we really see it in all the headlines. We won't go into it, but let's, let's really continue with the deck, so we agree there really is something here,

Yaniv Golan: [17:13] it's not... Yes, I think we agree, and probably many others agree with us too. The people leading this revolution, I think they don't hide what they think and know about what's about to happen in the world, and they're getting us used to it slowly, because it's going to be a significant shock. And I think the question ultimately, since we really had control over it, okay? most of us, is what do we do in the meantime, okay? And here, as I said, in the meantime we continue to create companies that benefit from this industry, and also companies that are protected from this industry, and also to create the revolution, and also to accelerate it, and also to enjoy its fruits. And is that the right thing to do? I don't know. Yes, maybe the right thing to do at this stage is to retire and grow vegetables in some corner, but this is what we chose to do, and this is what we enjoy doing. Two days ago, when I spoke with some company, they talked about how hard they're working to get an LLM to identify PII, personal information, inside data files, right? And how much they're struggling to get it to identify, there are some emails and email addresses, whatever.

Eyal David: [18:25] And I'm listening, I'm listening, I understand, because it's very impressive, the amount, how you managed to improve performance, but a regular expression, at least knock out 90% of the obvious cases, and now use the LLM for the truly hard cases, like, but this doesn't make sense. What you're pointing at is really interesting, and it's really interesting whether there'll also be this moment of sobering up, as you describe. We also recorded an episode here at the time with AI21, and like, we also really talked about this direction,

Yaniv Golan: [18:56] but do you think there'll be a sobering up? Meaning, because people, companies already went and developed it to... I think that... specifically, I think AI agents as a headline will follow the classic Gartner hype cycle. Meaning, there'll be lots and lots and lots of development, lots of money, lots of companies created, lots of promises about agent frameworks and new paradigms for building software, and in the end it'll come down to its right place, and be balanced by the fact that writing code itself is becoming much, much cheaper. I'd even say economically, okay? The moment the cost of writing, of producing code, okay? classic code, as they say, okay? mountains of modules in Python or whatever, that simply do the work without an LLM. Since the cost of that keeps going down, and there's some kind of race to zero almost, okay? happening somewhere, it'll simply be much more economical, okay? for AI to produce reliable code that goes through testing and is checked, and that's it, it passed approval and from now on it runs forever, rather than relying too much on LLMs, which are very expensive, right? They're very reliable by rule. To do the work that code knows how to do by itself. In the end there's a combination, a healthy combination.

Eyal David: [20:05] So okay, that's one hype, which I also connect with. But you know, let's, maybe let's talk about the... leaving the hypes for a moment, because there really are a few more hypes we could... but if we talk about the types of companies, I think maybe,

AI Native: companies where AI is the essence

Yaniv Golan: [20:21] we touched on it earlier. I think the topic of classifying into types, first, it's a basic human desire, okay? to put things in their boxes, and to know that I know which box to put each thing in. I think it's also very useful, it's very useful for investors, okay? so we know how to analyze each company and what to look at. It's useful for customers too, for buyers, for whoever buys these systems, to understand what they're buying here, okay? Are they buying a revolution or are they buying an improvement, and if they're buying an improvement, what does it improve and by how much? Are they buying unique knowledge, each of these is different. It's also useful for founders who come to create a new company, both to improve the product they're building and its fit, and to know how to talk to investors and how to get that essential resource, which is money and investment. It's also very relevant for product people, obviously, to know what kind of company they're in. Right. I think the first category was the clearest category and also the most exciting in a way, which is AI native: those companies being born now, and they're basically built for the world of tomorrow. Some of them live in the world of tomorrow, some in the world a year from now, some in the world three years from now, and they're companies where AI is their essence. Of course the most extreme example is OpenAI, a company whose entire essence is the models and their progress and the ability to run them at scale and so on, and also our own AI21 Labs, who do it with great heroism, it's very impressive.

Eyal David: [21:56] You put Base44 here, Base44, and I'm kind of torn. Yes, I think being torn is right,

Yaniv Golan: [22:03] because many of the examples I think of could be either here or there. But I think if we go by the definition of AI native in the sense that the entire essence of the company is AI, without it they have no ability to exist, no right to exist, but also no ability to exist, and it also creates the future, okay? by its very existence. I think Base44 is an example, it's an example of a product that's really native too, in the sense that it's not let's take an existing product and improve it with AI, but really a product built by a person you can see lives the revolution and sees the future. Another company that I think also demonstrates this way of thinking is AI Work, a portfolio company of ours, a company that looks ahead and understands that we're about to work in companies where some of the employees are flesh and blood, fewer and fewer, and some of our co-workers are digital workers. And it also doesn't treat them as LLMs or as AI agents, okay? but rather treats them as collaborators, as employees who are full partners in the company's processes. They know how to take tasks, know how to perform tasks, they know how to consult, they know how to learn, they know how to help other employees, human or digital. That's the world of tomorrow, it's a world that's possible, only the engine behind it is an engine that's truly smart enough to enable it, and it'll probably reach the intelligence we need within the coming year.

Eyal David: [23:44] How do you look at companies like this as an investor? So I think these are companies that represent the highest risk,

AI-Embedded: adding magic to an existing product

Yaniv Golan: [23:53] gain profile, okay? The risk in them is the most significant, many times it's not clear at all whether it'll work, many times it's possible, okay? They need to push the technology forward, sometimes beyond what's possible today. On the other hand, they have the opportunity for the biggest gain, for profit, for the most significant return, because if they succeed, they change the world. What kinds of founders come from companies like these, who come to you actually, what does it look like? I think it's very clear when you meet them, there's the... we talked about the spark, there's that spark. There's that spark where a person, where you, you look at the people, at the team, and you see they didn't come to build a company that'll do 500,000 in the first year and a million after 12 months, that's not their story, they came to leave their mark on the world. They came, they think very, very far, very far ahead, and no, they deal less with, it's obvious we'll set up the company, it's obvious the money will come, it's obvious the pre-seed and the seed, we're looking at what happens afterwards, how we, we don't have time, we're now, now, there isn't even a piece of code written yet, right? Now we already want to be in the biggest companies in the world, and talk to the most significant companies, and influence them, and start the POC, when there isn't even, they don't have time, they came to change the world, and it's very clear when you see it. And I think it's surprising, but there are quite a few of them. Meaning, there's no doubt it represents the cream of the crop, okay? of the founders we meet, but it's not like it happens once every ten years. It's, if I put some number on it, I think once a month, right? Oh, wow. Me, once a month for sure, I get excited to meet a team like that. Okay? A team like that, a team of founders, who really, it's clear they came here, it doesn't matter if we invest in them or don't invest, because we also can't always, but it's always exciting to meet them and understand that what's in front of you is people who will change the world. Let's move on to the next type of company, okay, we'll decide, which is AI-Embedded. Yes, AI-Embedded. When you really look at this split between types of companies, what we call, what the industry already calls, I think, AI-Embedded, are the companies where there's some existing product, okay, and AI comes and adds something significant to it, hopefully significant, yes, there's a product, there's a service, there's some infrastructure that already exists. It's basically this move of taking something that already works well, okay, and adding magic to it, and through that magic suddenly making it much better, hopefully, of course, it'll be a successful product. I think examples of this are companies like classic companies, like Notion for example, which have been with us for a long time, and lots of people really loved it, but suddenly the moment Notion added that AI magic inside, intelligently, smartly, the whole product got a different life it didn't have before. Suddenly it helps me much more, it's much more, uh, at work, yes, it also produces the content, and also searches the content, and also organizes the content, and in general it's much more helpful. I think also the moves Adobe made in Photoshop, you can argue about their success, some of them I think are very, very successful, some are hard for them because of the size and weight, but in some cases they added the magic. What's interesting is to see how young companies, and not so young, building on top of other tools, at the same time pose a very significant threat to Adobe and its capabilities, and raise a question about their right to exist going forward. It's a struggle. That's another type of company we can really talk about, the companies that really run fast with the help of AI, what do you call them? So what I'm saying here is that the struggle here is actually between a classic AI-Embedded company, Adobe and Photoshop and their tools are a classic example of it, and the AI native. It'll be interesting to see who wins here: Adobe's size and power and money and capabilities and base and enormous forces, or the Challenger who says, wait, you don't need all this mess. I think maybe the most classic example, okay, of AI-Embedded is Microsoft, I don't know, with their very, very wide set of products, and the co-pilot brand that's now stuck onto every one of their products and does the hard work of making it smarter, more. Is it successful, yes or no, I think there are opinions both ways here, but there's no doubt massive work is being done there of... several industries and industries, you've probably just been at...

Eyal David: [28:42] they had their conference just now. So in the medical world, like you talked about, basically replacing a doctor, not replacing a doctor, so yes, they're there, Disposable Labs, that's clear, and they're also now killing all the transcription companies, by way of Teams basically.

Yaniv Golan: [28:57] Either yes or no, let's see. Either yes or no, okay, they have a majority of users, so as I said, but... right, although it's not, it's not, you can see that it's, you know, all these assumptions of ours about all these companies that kill companies, yes. Take ElevenLabs, for example, with their models in the voice field. Ultimately that was supposed to be Google's. Google, at the time, when they launched all their telephony, basically the service, and sampled endless calls, it was clear they were doing it to train models, and there was no chance to compete with the amount of data they collected, yes, in those years. But then ElevenLabs came and simply did it better, because that's what they do, there's a group there, they're not a group inside Google struggling with all the corporate stuff. Because it wasn't a significant enough moat, that's what you're saying. No, because I don't think that's the reason. I think the reason is that, no matter how significant your moat is as a big company, a team that came to do the job, that focuses only on it, it's the only thing that interests it, simply moves much faster. If ultimately this claim that you moved much faster holds for penetration, for the first year, two, three. Will ElevenLabs or Google win in the end? I don't know, okay? But ElevenLabs got a nice opportunity, okay? to leave their mark on the world, and we'll see where they take it. The same is true for many other companies. Cursor, I think that's the clearest example. Can we talk about that for a moment? Yes, Cursor in the end is a huge success, but there's nothing in it, okay? It's built on Microsoft's VS Code. It uses models trained on code that came from Microsoft's GitHub. It enjoys a world of all the extensions written for Microsoft's Visual Studio. It's basically a wrapper, right? Around other people's models entirely, not their own models, models from Anthropic, from OpenAI, from countless other companies, sitting on top of all of Microsoft's repositories and Microsoft's tools. So it has no right to exist.

Eyal David: [30:57] But still. So where do you actually place it? I don't think it deserves to be called AI native,

Vertical AI: deep domain knowledge and unique data

Yaniv Golan: [31:04] it's not there yet. I think the next generation of these tools, maybe we'll start calling them, of software development tools. I hope we'll be able to start calling them AI native, because I think and assume they'll be completely different. I don't know, they'll already come from the understanding that says, that's it, the software world there is over, there's not much logic in continuing to work, even on the screen, okay? with the center of your screen being the editor. That's true in a world where you write the code. In a world where you talk to the agent, the center is no longer supposed to be the editor. In terms of real estate, in terms of how much of the screen it takes. I think when we see that, okay? we'll start to understand that a new wave of AI native coding platforms has arrived. Not there yet. I'm curious about the next generations. Another category I think is worth addressing is the topic, or the category, of vertical AI. Now, again, they're not black and white, things can fall either here or there, yes, but what characterizes vertical AI is that these are companies where the team getting the company off the ground is a team with a lot of very, very specific domain knowledge. It knows a certain industry much better than anyone else. Now, companies like these, companies that really come, that are born out of very deep familiarity with a specific industry, many times their moat is also, first of all, yes, from the familiarity with the industry and its processes and with what makes people there move and act and buy, of course. Many times it's also very specific data, data that's specific to this industry, and even if they use off-the-shelf models, you know, they're able, and it doesn't matter if they do fine-tuning or just use the information, they bring information that's unique, many times information only they have, that was very hard to obtain, and they build specific and very significant value. And examples of this, there are quite a few examples, I think these are very interesting companies, these companies. There's of course Aidoc, who do it in medicine, there's Eleos, a portfolio company of ours, that does it for psychological therapy, both for organizations and for the therapists themselves, basically enabling a significant increase in the efficiency of therapy, and quite a few other companies, each company in its own field, and they're fascinating companies. I think the most exciting is to see the companies where people from different disciplines meet, I don't know, inside a problem that's really a problem of a specific industry, each brings their knowledge from their discipline, and AI that lets this whole story move forward fast, and that's usually where the magic is created, that moment of something very significant happening.

Eyal David: [33:43] I want, for a second, for us to go back to embedded for a moment, to say two things. One, there are many companies disguising themselves as embedded, we can talk about that for a moment, because I think it's something really interesting, probably also founders who are inside a company like that, and also product managers who are in a company like that and really aren't sure. There are lots of them, and I'm kind of wondering how you as a product person know how to put your finger on it, if you have some tip or set of tools where you can put your finger on it and say, honestly, like, I'm in this situation. Yes. I think many of the companies from the first AI wave were like that, they were companies,

Companies in disguise, and when not to take VC money

Yaniv Golan: [34:22] all of that, disguising themselves. These are companies where really, in the end, the value they bring isn't theirs, it's mostly not theirs, and the value created isn't enough to justify a new product or a new company, okay? Because you say, okay, fine, there's product X here, okay? with very significant penetration, as required, right, they're not new like you, but they have very significant penetration and significant sales, they have lots of money in the bank, and they have lots of customers, and brand clout, there's a lot of trust in their brand. And you come and present a product that maybe right now is much more significant, and maybe it'll get adoption in the coming months and half-year and year, but long term there's no reason, there's no way you can assume they won't do it. They'll wake up, in their own time they'll wake up, the competing company, the incumbent will wake up, do it slowly in its own way, a year, two years, it'll still have the distribution, it'll still have the customers, and the moment it ships, your story is over. In a second your story is over. Companies of this type, yes, I think it's very dangerous. By the way, what does very dangerous mean? Let me be precise, meaning, I think it's very dangerous for VCs to invest in them. Ask me whether it's right for founders to build companies like that? My answer is completely different. Well, it's a great business, in my opinion. Yes, exactly. As long as it lasts. Right, so I think companies like that, what's right, and these are conversations I have quite a few of with founders, today I told a team like that, guys, I don't understand why you want money from VCs. What's right for you to do is take the small investments from angels, run with it on your own, and come out of this story very, very rich people. You don't need to inflate the company, okay? and start inventing more and more capabilities for it, just so VCs will feel comfortable and write the check that suits them. Yes, maybe over time you'll also become vertical AI, because you'll focus on some area, and then we can talk. Well, if you're in a company like that, then ask the AI. I'll also say about embedded companies that they're also companies that basically benefit from every improvement and every upgrade that happens in the models, which is also... That's an important point. I think it also has the flip side, you know? That when a team builds a product that's AI-Embedded, let's call it, sometimes you're very tempted to say, well, the capabilities, I'm almost getting what I need from the model, not enough, I need to improve its capabilities in area X and Y. And they fight it, sometimes they fight it for a month, sometimes two months, sometimes three. I think countless times, okay? I've seen that at the end of this battle they succeed, and two days later the new version of the model or the framework comes out and simply provides that feature built in, and all that time simply becomes worthless, it was wasted. So many times the right thing to do is simply to say, okay, there's a capability missing here, I'll simply sit and wait for it to arrive, it'll arrive, at an ever-increasing pace.

Eyal David: [37:17] Yes, also something to think about, there's also a pricing issue here, like all the model companies today run on very low pricing, and these products not necessarily. Do you think this pricing of the model companies will continue like this? As a 30-dollar subscription, and I already don't pay like that, they pay much more, but...

Model pricing and AI Resilient

Yaniv Golan: [37:34] Listen, I don't know about the subscription, I can say that the pricing of the service we get from these companies isn't sustainable, it can't continue. Clearly it's subsidized, and subsidized quite a lot. A lot of the costs come of course from the cost of hardware, since there's one hardware vendor that dominates the market, it really enjoys this situation and doesn't have too great an interest in lowering prices, and that won't last. It won't last. The dynamics of these things are known, it's clear where it's going, there are lots of companies making an effort for it to happen. And so I think that if we see, beyond all the software improvements of course, hardware costs drop by an order of magnitude, this whole story will look different. We'll wait and see, it'll be interesting. The next two types of companies, I think the combination between them is what excites me most, but we can get to that later. After that comes AI resilient, right? Yes, so AI resilient is really the companies that, unlike all the companies we've talked about so far, don't wake up in the morning and say: wow, I have to see the latest announcement that came out of Anthropic or any of those companies, but simply don't care what happens in that world, because they have other worries, okay? They're companies that are usually what we called Deep Tech in many cases. These are companies that deal with creating things that are often a scientific breakthrough, or the use of a scientific breakthrough, or a technological one, or even an industrial one, and their challenges are completely different. And their characteristics as companies are completely different too, meaning the time it'll take them to develop their product is much longer, they'll need much more money, you need much more patience as investors, and the outcome can also be accordingly very significant.

Eyal David: [39:21] I'll add to that, it doesn't have to be. I immediately went to okay, this is really, there's the SpaceX logo here, I immediately said to myself, okay, this can be something physical, but it doesn't have to be. It's enough for the thing to be behind regulation or government, and then I already have a moat that makes me like that, but the connection to the physical here is really interesting, I really love it. It's very significant. Yes, although the point you raised is an important point,

Yaniv Golan: [39:42] I think it's important to remember it, that it doesn't have to be a physical product and doesn't have to be a scientific product either, okay? Deep-tech-y. But it has these characteristics, that really it's simply immune to everything happening in the AI market, it doesn't interest it that much, it can improve it a bit, maybe the researchers use AI tools, and the regulatory product people, but it's not fundamental. So SpaceX is really an amazing example of a company of this type, which with great stubbornness did the impossible, and the breakthroughs there are many and significant. And there are lots and lots more companies like that, whether it's Formlabs, which also changed the world for a moment and created a whole industry, or PsiQuantum, who promise to change the world in quantum computing. There are also, as I said, companies we really like, I mentioned two of ours here, NeoLogic, we talked about them earlier, what they basically do in the end, yes, with their different approach to circuit design, is cut the resources required for computing by 30, 40, 50 percent, depending on the configuration, and that's very, very significant, whether it's how much power they consume, okay? how much energy is required to run these systems, or what size they require, and heat. How much space, physically. Chiral Energies, yes, who produce chiral nano-coating. Chirality is a chemical property of a material that, it turns out, has dramatic implications when you know how to use it correctly. For example, one of their products is a coating for conductive materials that simply improves their conductivity by tens of percent. You can understand for yourself what the implications are of improving conductivity by tens of percent, whether in preventing waste, or in lowering the price of hydrogen production, okay? green hydrogen, and a variety of other uses. And one of the wild things about a company of this type is that it turns out we've only started to understand the implications of these materials, and their uses are many more and even more exciting. So we're in a fascinating discovery process there,

Eyal David: [41:57] where every opportunity is even more interesting than the last. I love it, and how curious you are and how you talk about it in the same manner you talked a moment ago about software companies, that's amazing to me. It is indeed amazing and fascinating.

Yaniv Golan: [42:11] Sometimes it's scary too, because suddenly you realize you invested in a company where an incident can happen where the container full of acid breaks, spills on the floor, the lab worker walks in, slips on the acid, and you realize different things happen here,

Eyal David: [42:33] it's not that production went down on Thursday. This is an investment you make, like, do you have more investments like this in fields like these?

Yaniv Golan: [42:41] Yes. It's a first investment, right? Yes. Okay. So you know what you're getting into. Yes, yes, we know the implications of it, and still it's fascinating and it's different, it's still, it's not most of our investments, it's a certain part of the investments. And this year did you feel you're making more resilient-type investments? No, we try to keep the balance between these fields. Look, it's also, it's not just this year, right? It also depends on the fund's life cycles, there's some significance to that too. Yes, I don't know how many years you want, it's seven, right? Yes, but every fund has its life cycle, and it's a not-very-secret secret, right? that our thinking as investors changes depending on whether we're at the beginning of the fund's life, and we have another X years, right? until the company matures, or toward the last part of the fund's life, when you're already investing less for ten years ahead, and you want to see companies that will mature in less than ten years, right? So investments of this type, let's call it Deep Tech, okay? usually happen more in the first part of the fund.

Eyal David: [43:39] So I, okay, I'm running through the deck here, these are actually signs that are maybe more relevant to founders, which is great, when not to go to VC and run, as you said, actually already,

Yaniv Golan: [43:51] run with the product? Yes, I think we'll see more and more companies like that, that really don't need VC funding. That may also lead us to a different conversation about the future of VC in general, maybe come back here for it, it's a super interesting conversation.

Eyal David: [44:04] What do you think are the most important product skills today for a product manager on an AI product? On an AI product. It's interesting to say on an AI product, because it's a very broad category. We touched here on the different segments, you know what? A product manager in general, because they can easily find themselves in a company like that too.

The product manager's skill in the AI era

Yaniv Golan: [44:23] Yes, this whole definition of a product manager is going to change, and it'll be very interesting to see where it ends up. I think the skills of a product manager in the end, a product manager, let's call it a real one, okay? in the end it's a kind of, let's call it risk management in some way. On one hand there's the desire for innovation and features and moving forward, and the whole company pushing about what the next release will be, and on the other hand there's the negotiation against the organization's capabilities and resources, and some kind of risk management, yes, in this story. That's what seems to me, yes, the thing to bring in right now, and that's what will create the most significant impact, in my view, having checked. All the other related skills are more soft skills, but I think that's ultimately the real goal, somewhat, of a product manager, beyond the more project-like definitions, yes, which I strongly believe in. Yes, I think in the end the product manager is in some way the true soul of the company, okay? I can even take it to the extreme and say the whole company, as far as I'm concerned, exists to realize the product manager's vision, okay? to serve the ability to really turn it into a real product. We're in some kind of revolution here, because this basic skill, of I understand the need on one hand, I understand my resource constraints on the other, I also understand the political pressures, and from that I create a product definition that moves forward on the right vector, was always constrained by the ability to execute. A product manager can do, male or female, they can, there's only so much they can do, okay? In the end lots of other bodies in the company are responsible for execution and everything related to it, and suddenly the product manager discovers they no longer have to wait. And I wonder, and I'm curious, okay? to see what the next generation of this profession will be. I think one of the possibilities is that we'll discover that it's, it's, it's,

Eyal David: [46:27] it's the only real function that remains. I signed up earlier for AI Work and I saw that really one, when I signed up, one of the segments is product, so afterwards I'm curious, I'll peek at what it is, but I think I understand where they're taking it more. Actually I think there'll be a rise in the number of product managers, like you say, once the how becomes very simple, then basically what's left? What's left is being a founder. Right. Now the company, I slightly disagree, I think you'll agree with me too. The company is there to make money, and in the end we need more founders like that inside the company who will break through and create more revenue streams like that. Really, when the how goes down, what you do becomes much more interesting. So I really think there'll be all kinds of generations and all kinds of types of product managers here. I, I want to believe there'll basically be more mini-founders inside companies, I want to believe that's what will happen, and you're saying the number of developers will probably go down too. So maybe we'll see more of that kind. Listen, I don't know either, I see at least in the trends that the number of product managers is actually rising, and I don't know if it's related to AI or it's simply a motion of people wanting to enter the field that already happened before, but that's what I know is happening now.

Yaniv Golan: [47:33] I think first of all I'm interested in looking at the skill before we talk about roles in these companies. The skill of the product manager, I think we'll agree it's going to be very important and significant, and also produce lots of founders inside companies and outside companies. Like, their own companies, because in the end, when you remove everything else, that's what really matters. Okay? That's what really matters. Now, once I have this skill I can either hire developers or hire AI developers or both of them, yes. But let's talk about founders then, there'll be many more solopreneurs, that's clear.

Eyal David: [48:06] On one hand. On the other hand, really, there'll be the whole world of disposable apps and so on, the way I see it, and then everyone can make themselves the next TripAdvisor for their trip to Rome. So how do you see the... I'm jumping in for a second because these are also questions people asked, and I'm already trying to wrap up toward the end of the episode. A fascinating conversation. We still haven't talked about the last segment, which I also want to bring in. Yes, so I just want to hear the... I dove in there, I want to hear your opinion on B2C, where it's going in terms of opportunities that can be seized today.

Yaniv Golan: [48:39] A. I don't know. I'm curious about that too, it's happening while we're talking about these things. And I think yes, there are going to be many more solopreneurs who will create many more solutions tailored to people as much as possible, and that's going to happen in the next year or two, that's the time window. That's the time window this world has. What will happen after that I don't know. Okay? I can say that even if I can create my own TripAdvisor for my trip to whatever, not everyone wants to. Not everyone, even though it's maybe simple and easy and possible, not everyone wants to. There are lots and lots and lots of people who want it packaged and tailored specifically for them, with as little mental effort as possible, because they want to enjoy the trip, they don't want anything else. You know? They have other things they're busy with. So where will this balance end up, between the ability, which is clearly going to be insane, to create solutions built for ourselves. You know, you talked about disposable apps. There are more fascinating questions here, okay? After all, in the world of enterprise systems, okay? in the end all these enterprise systems are dashboards and reports, okay? and forms and that's it, more or less, okay? So, and a database behind it, of course. So in a world where that's all there is, do I really have to create a new enterprise application each time, or can I simply use, yes, that really brings us back to good old code, linear, as you called it actually. Not that point specifically, but rather the question of UI, I think. When I can generate the UI on the fly, okay? whether it's done by an AI agent or in Python code, okay? when I can generate the UI on the fly, my question is whether that's right at the basic human psychological level, okay? that the UI gets created on the fly, or whether I as a person need to know that when I move my hand left and up, I always reach the File menu. Okay? No matter which method I work with. We'll find out. Here too, I think the world ultimately goes toward patterns,

Eyal David: [50:43] and in the end these things that get created on the fly, which is probably also what will happen, it's what we also see in the conversational interface that's happening, in the end there'll be a very clear pattern that this is the most normal UI. I tend to agree with you. But it'll take time and it'll take iterations. But it seems that's the direction, as you said, it's dashboards and not overly complex problems, I think, at the end of the day. Okay, let's touch on the roll-up, because I really want to bring it in.

AI Rollup: buying a business and upgrading it with AI

Yaniv Golan: [51:08] Yes, so look, the roll-up is something that was also born recently, a bit later, and basically it was born out of an industry insight that says, wait, if we basically, if it's so easy to use AI to improve businesses and make them much more profitable, then now I'll take some domain, doesn't matter which domain I chose, and I'll improve, I'll take the software and use AI, and improve the business's performance by 90 percent, for the sake of argument, because it's possible. And now I'm going to sell it to these businesses, how much can I really charge them? After all, I'll never get the 90 percent I saved, right? So I'll get 10 percent, I'll get a subscription of a thousand dollars, of ten thousand dollars, of whatever. Wait, but why actually? I have another alternative, I can simply sometimes, yes, buy this business, and pocket the whole 90 percent I saved, okay? as the operator of the business. And that's basically a thesis that says, let's use these capabilities to roll up, to buy a series of businesses in a certain field, improve them operationally in a significant way using the new tools, and enjoy that performance directly in our pockets. It's very reminiscent of what private equity firms once wanted to do, and so it's basically a kind of private equity play that's technology-enabled, and still more and more VCs are trying this model. I think the whole industry is curious to see whether it'll end as an expensive and unprofitable experiment by VCs who have too much money,

Eyal David: [52:44] or as a new model for building companies. We'll wait and see. Cool. I had such a great time, we also ran over time, I just had a really great time. Yaniv, thank you very much. Come on, thank you very much. It was very interesting for me. Come on, cool for now. Bye bye. Hey everyone, thanks for listening. If you found this podcast valuable, you can subscribe, follow us, of course, for more episodes on Spotify, Apple Podcasts or any other app. And of course if you didn't find us on some app, I'd be happy if you wrote to us. We'd really appreciate five stars on every platform and for you to follow us, so more listeners can discover us and find the podcast. You can also find previous episodes on any app or on the YouTube channel, we have links in the description. So until next time, come on, be efficient, and bye bye.