TRANSCRIPTIE 🔗
Yair Nativ: Hey, so welcome to Future Proof, everyone. Today we have with us Guy Gamzu, an entrepreneur and investor invested in companies like Fiverr, Moon Active, and Spike, AI, and some additional ones that he will share. One of the most interesting folks, I have the the privilege of speaking with everyone in a while. We have interesting episode, but before we dive that guy, would like to ⁓ share a little bit yourself?
Guy Gamzu: Hi, thank you for having me and inviting me to your podcast. I really appreciate the opportunity to some ideas together you and your audience. I started my journey as an entrepreneur in an industrial tech and manufacturing background and ⁓ sort discovered the real startups that were focused on technology back in the 90s. And then it became actually my passion to try and help and build technology companies. was fascinated by the talent pool that those ventures actually attracted, as well as the high aspirations to create changes that will touch many people. So I think it's a combination of the scale that drew me into that. And I found myself actually doing only that, my business already more than 15 years ago, ⁓ 20 years ago. Time fast. And then ⁓ since I've been only dedicated my time for startups. 80 % of that is with companies that I'm actually a partner ⁓ in, usually like single number spanning from companies that just ⁓ started one or two years old to all the stuff, probably more than a decade old. Some of them actually public, as you mentioned. that's of my time. The other sort of 20 % is for new ⁓ enter entrepreneurs reaching out to me, whether it's for seeking investments, seeking some of advice or discussion. about what they do and my goal is really to try and be involved and to help and other than that, I also talk to other forums ⁓ in an ⁓ context or other groups of ⁓ that wish to have like a group discussion. So that's me. Maybe just another ⁓ flavor of background. When I started investing was... like I mentioned, end of the nineties. was ⁓ under the impression that I'm sort of late to the game. Internet like early nineties. And ⁓ when made my first investment, there were already large companies, some of them were public. And I you it's just fascinating. So I'm going to ⁓ step in, but big successes seem to be behind us at that time. at that point in time, it sounds ridiculous now, but that was my feeling actually.
Yair Nativ: Yeah.
Guy Gamzu: So would divide it into two. so companies, if if you look at the mass market, companies that started adopting the internet early on were just probably in the hindsight, we're doing the right thing because they used email and they used a website in order to communicate with their market, was just another channel that they opened up. And I don't think that in that sense, there was anything premature, so to speak. So that's one group, but the other group is the group of companies that actually started evolving at that time and were there in order to create a change. So I
Load full transcript
see a difference here between Internet entrepreneurship or the technology startups and ⁓ today native or AI companies that actually start and the mass market actually has some sort of earlier or early adoption.
Nir: With the internet, ⁓ like the conventional wisdom ⁓ ⁓ it took a couple of ⁓ to see sort of wide effects on growth productivity. Are you seeing a similar of trajectory? or predicting a similar kind of trajectory with AI, namely that now, ⁓ are doing it because they have FOMO but the ⁓ actual actual value hasn't played out yet.
Guy Gamzu: Yeah, I mean, there are, it's like history. not exactly the same, but it rhymes. I think that what we then is like a fundamental technology wave that was ⁓ creating what mentioned, like the FOMO. And in many ways, the imagination of founders and also the investors were going much faster than the actual development of the technology side as well as market adoption. thing which is very similar is that if you look at the magnitude investments, so I think that ⁓ by the 90s, we numbers that are even if when you adjust it to inflation, are smaller than the numbers that we see today invested in AI. technologies. So there are a lot of things that are looks that they seem to be quite similar. The most I think the most ⁓ difference is the market cap of the of the big companies. you look at the 10 leading or 10 most valuable companies at that time, ⁓ think they were split between technology There were some technology companies like Cisco, think, and others, there were also like other companies like Exxon or Energy, large companies or pharma companies. And even then the value of the 10 largest companies was around trillion to adjusted ⁓ inflation. we're talking about the top 10 today, think almost most of them are technology related. The is five times higher than that. I mean, when you look at it's inevitable that you'll see some similarities in the behavior and the pattern. And the real question, I think, I mean, is it going to play in the future, is ⁓ obviously harder guess or hard to predict, but we can obviously try to find some similarities and some differences.
Nir: And Guy, if I could follow up on that, ⁓ is it your sense that language model ⁓ AI, which is just slice of that ⁓ technology that can primarily do one thing ⁓ pretty well, language use, is that promising enough to justify these ⁓ exorbitant ⁓ as who's seen the sort of whole arc of this ⁓ play out Are you holding out and more excited for other variations of AI?
Guy Gamzu: I believe that it's going to out very differently to different players. So when look at the market as a whole, and you look very far into the future, then I guess that ⁓ even market is not wrong in promise of large language models. this is like a very, it's a huge shift that can... ⁓ enhance massive changes in how we do how we consume things, how we actually manage our productivity and so forth. The real question is the timing. And that's a huge question. So one thing is actually
to see whether that will be sustainable or it's going to in due course. And the other question is on a micro level, when you are just one player and you don't really care that much about the industry unless you're an investor in the market. But if you are a startup, you may end up in a very good position, maybe even better than the fluctuating bias. on the other hand, you may find yourself getting out of the game because of those fluctuations that can... throw you away, which happens in the past, as we mentioned. So before that, maybe we just need make sure that everybody is with the actual history. So a lot internet companies started with similar promises to become a platform and ⁓ didn't make ⁓ it's not the ones that were like Amazon. mean, I don't remember all the names, but you had companies like that still exists, like travel companies. ⁓ It started very early and even E-Trade or other financial product and so forth. you have and others, Webven. ⁓ those companies were actually platforms. Some of them actually made money. Some of them still exist. Some of them, many of them not, but many them actually came in as a real ⁓ platform. the other hand, I think that when you look at the
Yair Nativ: Mm-hmm. I'm crazy that it's recorded in the middle of the podcast
Guy Gamzu: real economics or the ⁓ ROI of the market as a whole, the entire internet monetization or business success started with a sort of sin, to speak, because the main sectors did work pretty well was actually porn and ⁓ gambling worked better than e-commerce at that time. And when you look
Yair Nativ: over. I'm so confused.
Guy Gamzu: when you look at those industries as a fraction of the offline industry at that time, you'd probably find out that while porn was like, I don't know, 15, 20 % of the market, others that were extremely successful were taking under 1 % of the market. Even advertising, online advertising, when I was involved in that in the early 2000s, after the dot com, even then at the highs,
Yair Nativ: That's
Guy Gamzu: it was less than 10 % of the entire market. Today it's completely different. So yes, a lot of those markets actually grew and obviously actually the markets that were very tiny grew much more because they had more room to grow. But again, it's a question of sustainable, how long can you actually endure in order to get to the market and find out the exact... take the actual cream ⁓ the cake. So that's the history. And if go back to the question of what of things you can actually, should or ⁓ you'd be off doubling down on, that's very, very challenging question as well, because nobody really knows. mean, anyone who's trying to tell you that's correct, that's incorrect.
Yair Nativ: Yeah.
Guy Gamzu: would probably be too vain to try to predict the future. And I don't want to be in that bucket. But I can tell you that there is a lot of things that mislead. especially with, I mean, when you look back on the internet, a lot of things didn't work, technically didn't work. you didn't have broadband initially. was no, until 2007, the iPhone moment, you didn't have too many people actually being able to access this platform. didn't work as smoothly as you'd like. Logistics weren't smooth enough. So there were a lot of things that actually could attribute to productivity or technology. And the same time, you also had issues of market adoption. And so product market fit was almost impossible to get, especially at scale. I think what we are seeing now is actually sort of the opposite way around. do have very impressive technology. You do have a very good distribution or potential for adoption, but there is something very misleading in the sense that the first encounter or what actually most users experience very, very initially is very ⁓ and it actually captures you. So if I expose you to something that... that LLM can do and you never experienced that you're going to be old and excited about it. The problem here, I think, is actually consistency. So because the model by definition is not deterministic, what happens is that oftentimes when you come to rely on that, you get certain results that seem to be reliable, but actually not. And that of... ⁓ and opacity is even more scary for applications that has a lot of dependence on. when you are trying to sell your offering to another business, want the offering to be 99.99 % accurate rather than 95%. Even 95 % is not good enough. The conclusion at the moment, which may change, is that if you really want to be more successful, you need to be more granular or actually break your offering into something more focused or more specific not be sort of ⁓ sold the demo or by the specific use that you are showing because that may cost you a lot once you start deploying.
Yair Nativ: So I have a follow-up question. ⁓ will give an example of a potential client we are ⁓ with, like a law firm. And said that they used a product that summarizes long legal And they said did a good job, but the lawyer still read the entire document regardless. And then there was point of using the product because why use it if I'm reading both the summary and the long document. the product was practically useless because they did not trust it. ⁓ And the level of ⁓ trust have today. And on the other hand, I recently with a senior VP of product at one of the important companies at Valley. And I asked her, should we as a company focus on a narrow topic, that, ⁓ work in a project base until we figure out the best use cases that are really working and then scale the productized app? Again, really long question, what would you recommend? What would be your recommendation to entrepreneurs?
Guy Gamzu: My recommendation is to go bottom up. So to start small, make sure it works, make sure there's a huge product market fit to a very small number of users actually, and then try to scale from there. That's my intuition. That's not to say that other more ambitious projects are not going to work. Probably they
will if they have enough resources and they persevere, but For most of the founders that I come across, you don't have this privilege of endless resources. And given the fact that you do have limited time and this is like a window for you and you have limited resources, I would actually try to focus and to try to something ⁓ to a relatively ⁓ audience. And I can, from my experience, I can share... maybe a couple of examples if that's of interest of projects that worked and others that are more challenging. So let me know if you'd like any of those. So for example, Fiverr did ⁓ project which was focused on what you can call a sort of CRM for each seller. And that kind of ⁓ project
Yair Nativ: Yeah, yeah, yeah, of course.
Guy Gamzu: was extremely successful. I think thanks to the fact that it was very focused on a set of information that was very limited in many ways. So the LLM power was used as a ⁓ to actually be able to communicate in a very free language with the user, but at the same time, the universe of replies and other things that can and cannot be done, including escalation to the actual person, was built into the model. And therefore, I think it actually was ⁓ very, confined in many ways and successful. Another example have is from a founder I met like just a couple of weeks ago. He built on It's a single play. He built ⁓ a startup called which is actually around events. And this, took actually very, very, again, very specific use case, very specific set of capabilities. It's not to say that he only uses ⁓ the or the lingual part. He actually uses a lot of ⁓ image but this is again, in a very, very confined context.
Yair Nativ: Mm-hmm.
Guy Gamzu: that can bring a very quick value to users of that platform. So those are a couple of examples that are pretty straightforward. To give ⁓ a ⁓ example, which is ⁓ yet to seen, I don't know what will be the result, is actually in a company I am involved in. And what we developed there, this company is called Spike, they started off as like, yeah.
Yair Nativ: Mm-hmm.
Guy Gamzu: turning your email into a messaging platform, doesn't matter which email you are using. that works and that's great. And they gained a lot of knowledge in that space. given fact that we want, ⁓ I ⁓ chat our platform. So we said, you turn your email into chat and ⁓ suddenly came in, it's all about chat. So we said, okay, what can we do about it? And we... pivoted over this several times, but initially we started with a product that is designed for consumers, which I actually use even ⁓ prepare for this kind of meeting. I use that, including other meetings that I do. And the reason I do it is because it involves both the context that everyone has on LLMs, but also ⁓ importantly, my own context. So it's actually indexing my own email database. And based on that, it actually gives me specific value at the moment, just ⁓ share with the listeners, ⁓ evolution that. So we started with something consumer. Then you said, no, maybe we
should actually provide enterprises with this solution. So they do it internally in their organizations and actually then came to the conclusion that the best would be to do an open API for that. and the set of SDK, which we are doing now to give anyone who's a developer the capability of adopting it and actually this kind of context or personalized context for their own audience.
Yair Nativ: Okay. That's you want to twist something near
Nir: Yeah. No, no, go ahead because I wanted to follow up with a related but a different question.
Yair Nativ: And it seems that so far, ⁓ hate when people say AI because it's such a big thing, but it that for personal matters, like serving as your own personal context, it's pretty In enterprises so far, it doesn't work. know, Microsoft managed to sell Co-Pilot and there are a lot of GPT subscribers in companies, but effectively the data shows that they're not really using it. and for the most part. Any thoughts on how to create this layer of trust for organizations? ⁓
Guy Gamzu: So yes, think, look, let's look at CRM for example. It's relatively simple, relatively. mean, and there are a lot of them on that. Let's say to replace it, a PBX should be no brainer. so, that, mean, PBX process itself where the answering machine tells you press this or press that in order to go to it through a tree, shouldn't be there.
Yair Nativ: Yeah, yeah.
Guy Gamzu: So if we're talking only about this process, just saying what you really want and asking just a follow-up question can lead you to the right place, even if you leave everything else as is. So I look at that and I ask myself, okay, why, I mean, why isn't it changing for everyone? Because it's a hassle to go through BBXs. Same for CRM. Certain things in CRM can work, certain things cannot. And here, you can actually adopt two approaches. One is to say, again, I'm going for the layer, which is very, very simple and very technical, and I'm going to train only on that data and show a lot of cases where I do a control group alongside the regular group and show the differences in satisfaction and also in follow-up. So it's not just about how happy the customer was in that time, but also follow-up it a few weeks later to see that it actually worked and it wasn't a solution that didn't really solve the problem. The other approach is to actually process the data and offer the person who actually is doing the CRM something that come pre-processed and let them decide whether they want to edit it or use it or reject it and learn based on that. in order to progress towards a more refined offering.
Yair Nativ: CRMs are relatively, you it, like simple. There are big liability issues if
Guy Gamzu: That's true only for some of the cases because if for instance, the CRM has the user, yes, you are right, this is really a problem. We are going to send you a new device tomorrow that costs $3,000, maybe that can become an issue. So the question is really the cost of the mistake for both sides.
So I think that's something that should be evaluated, but please.
Yair Nativ: Yeah. Yeah.
Guy Gamzu: Let's talk about more complex example as you mentioned and see this should be tackled.
Yair Nativ: And.
Guy Gamzu: That's a good example for something that is actually in my mind is confined because again, if you segment the problem and you're focusing on something extremely specific, you can train for that. So what you do is actually, even if you have a bigger problem, I think that the general approach and I'm being very general here and might not apply to everything, but... It's like doing a marathon. mean, if you think you're gonna run it just overnight, it's not gonna work. And the idea is to break it into parts. So I think that the same applies here. You have to break the problem into sub problems and that you are working on a non-deterministic platform, but at least confine it to something that the probability of... ⁓ being wrong is smaller and the cost of it not huge before you actually move to the next step in the process of whatever it is that you are solving.
Yair Nativ: and
Nir: Yeah, and maybe maybe we'll be connected to your point about the human element here. ⁓ curious, ⁓ ⁓ little bit about the sort of if wish, if you will, the difference between your viewpoint as an entrepreneur as an investor and mine as an academic, a founder ⁓ and an investor. ⁓ as you say correctly, have to focus on well described and specific use cases in to convince people their product is useful and scalable. And then people like me who work on of ⁓ AI ethics, we kind of, you have to work on big picture. ⁓ questions. So for example, if you find a and useful product to streamline the work of a law firm, and come up with a document that uses AI to do ⁓ document review in a, you know, helpful way, which has happened. On the one hand, you create sort of ⁓ real value, right? And on the other hand, ⁓ There are questions that arise from that of, you know, de-skilling what happens to an entire generation that now doesn't get that skill of reading documents, finding mistakes themselves. Do you create some kind of effects that by eliminating route work at the low levels, you diminish the pool of people who advance to ⁓ higher roles? know, these kinds of questions in universities. ⁓ There's a analogous sort of example where if you bring in intelligent tutoring systems or technology like Notebook LM, which is amazing, the one hand, it's a successful use case. On the other hand, it means that people read less books. So it seems like there's a bit of a disconnect where founders and investors and entrepreneurs have to focus on the specific use case. Academics focus on sort of... social consequences or figuring out where the technology can have great results and where it's more problematic. ⁓ My sense is that they don't talk to each other that much, this conversation being an exception. So I'm curious about how you those kind of things. it like how do you for example, about
an entrepreneur that does ask these kind of big questions is your sense that essentially this is how capitalism works, that everybody focuses on their own sector ⁓ of ⁓ the economy and then the rest works itself out.
Guy Gamzu: I think that first of all, for a founder, even if you are really focused on something very specific, like sort of... solution, very granular, very small population and so forth. Still, without the big vision, you will find it very, very difficult to communicate and to actually attract investors. So investors are not interested ⁓ so focused because they look for the grand picture. So in that sense, it's very similar to higher causes. so to speak, where it comes to works that you can see in your focus area on ethics and so forth. I don't think it's very, very different because you are building something, you say, I'm building this because this is actually my first step to prove execution and product market fit before I progress. But the problem is that oftentimes because... your way to survive and your way to actually get the grant, so to speak, be able to research is dependent on the big story. So what happens is that you are actually getting convinced by the big story and then you are just all around ⁓ instead of being able to be disciplined enough and to go back to the focus. And I'm not enough with research, but I ⁓ think the same approach could probably apply here as well. So yes, there is the cause of how you go about it, but maybe let's try to look for a very, very specific use case and try to see how we work with that. And that actually requires a lot of creative thinking. And even ⁓ talked before about those, 5 % of wrong decisions by LLM ⁓ that problem with which is that we don't know, it's not that they are saying, okay, I'm wrong. They actually look very correct at the time of delivery. And then you could think about other agents or other mechanisms that are actually designed solely to examine or to question the decision before you actually go forward. And if there's a doubt, maybe there's no doubt and then you actually accelerate it. And maybe that can be applied also to other realms of life where you actually have certain things and then you bring it to sort of human brains that review ⁓ process and review the transparency of the process and take a certain decision and that being fed back to the system in order to make sure that it learns and vice versa. So I mean, and that's one direction of thinking.
Yair Nativ: So ⁓ following up on the human aspect. going to give one last example and let's move to others. But in law firms, they say that the lawyers do not ⁓ want to switch to AI. They're actually forcing the interns to not use AI because they don't ⁓ trust it. In HR teams, they're using manual screening, they are doing it again ⁓ AI and they don't trust the process. ⁓ in marketing teams, the same. ⁓
Guy Gamzu: and we need to create Yes, I think what you're pointing at is the... And that works also for the bubbles. bubbles usually the main is that people actually price the inevitability of the new technology with probability of actually gaining trust and adding value to the users. And I think it's also, essentially it's a matter of probability. So if, for example, I searched certain things on Google, I used to search on Yahoo, by the way, and then Google was born. And in order for me to go only to Google, there was a phase where I used both, of course, and tried to see how it works. Same goes now with LLMs. People, I think, juggle between going to Google, fact checking. going back to the LLM or trying the LLM, maybe sometimes trying more than one LLM to see if there is a consistency, it depends how important it is. And the reason being the probability of the result being accurate. So I think it's, in my mind, it's just a question of quality and consistency. And that's why I'm saying that intuitively, the narrower the scope,
Yair Nativ: Mm-hmm.
Guy Gamzu: probably the trade-off would be that you'd be able to give higher quality or higher confidence faster. That's the main underlying assumption here. And if that doesn't work, maybe you need to go back to the core technology and check what's wrong with it or how to improve it. But actually the layer, if you ask me, the basic layer of the language model is not the issue here. almost, I find it a bit to say that, ⁓ it's also almost commoditized. ⁓ there huge companies making huge investments in it. ⁓ This actually is not the core layer that will bring the value. It's the enabler for this result. And I don't think that actually the quality in the sense of probabilities will come only based on the platform.
Yair Nativ: Mm-hmm. Mm-hmm.
Guy Gamzu: of the core language model. It's a combination of the layers that you actually be able to build on top of it in order to refine the quality and the results. that usually in technology, we tend to think about scale because that's what makes a technology venture different than a traditional one. But I think that it's and it's true. Yes, it is. It's not it's not something wrong.
Yair Nativ: Mm-hmm.
Guy Gamzu: There's nothing wrong with it. But I think that before you solve the core unit, the molecular level unit, you can't really start building a whole scale with it because the scale would come with a lot of flows built into it. And you won't be able to gain the probability game of gaining the trust.
Nir: Yeah, specifically on that guy, was wondering again with your with the historical perspective that have. Are you seeing entrepreneurs in this more able to focus on the molecular unit than in the ⁓ 90s or So in other words, the sort of tendency to go broader than the specific use case, are you seeing more of that tendency now than you did in the 90s?
Guy Gamzu: Yes, I see more tendency because of one main reason. the time, when internet started, the business model was almost non-existent for most of the ventures. So the common was that you need to gain eyeballs or
more participants monetization will come You'll figure it out. And that was really common sense. mean, everybody believed it. And I think that with the AI, even in very early stages, in very well-financed platforms like OpenAI and others, you could actually see from the very early stages that monetization came into play. So I think that today the market is much more mature in a business sense, that the business model and the unit economics should play right. In that sense, I think that there's something more mature about the way startups are actually being formed today as opposed to years ago.
Nir: And what's the sort of parallel or correlate of the argument from the ⁓ Can ⁓ model an existing ⁓ operation? mean, is there even a kind of common sense ⁓ orientation there was?
Guy Gamzu: I believe there is, I mean, there's always a sense of engagement. You to measure first, even before you can make money, you ⁓ to make sure that you have enough engagement and retention. And then you say, okay, I'm confident enough this can be monetized or there's a certain funnel. And also there's more, people more conscious about the cost of distribution. And then they were back in the late nineties at that time, virality was like a buzzword and people tended to believe that, you you build it and they will come because they will hear about it because it's so great and so different. it's the, again, ⁓ inevitability assumption people will have to be there because there's no other option, which again, it's not that wrong, but you had to give it 30 years in order to. ⁓ become a reality, not a...
Nir: Yeah, my sense looking from the outside, by the way, is that the sort of internet bubble caught people by surprise in the 90s because of this sort of inevitability and ⁓ And AI bubble is actually what took you so long, right people, it seems like people are not as caught by surprise by it.
Guy Gamzu: Yes, I think the problem again, even then there were people talking about the bubble ⁓ looked at the valuation and if you were a little bit older, you at it and you said, you know, there's a basic economics and ⁓ I discussions with people that were having regular traditional businesses asking me can, ⁓ I mean, justifies this? And so people posing questions. but FOMO was stronger. people just, they said, yes, I can't understand it, I already, I've been wrong for three consecutive years. Maybe I should do something about it instead of just watching from the sidelines. that's, and the thing is that usually those bubble burst, people by surprise, they don't know what is the trigger. They can't really anticipate the trigger. So they can't, again, like the technology itself, they can't really time it perfectly.
Yair Nativ: Mm-hmm.
Guy Gamzu: While it happens, they just want to make sure that they optimize their participation in the hype. It's very difficult to predict. think that most of us, yeah, you have to be focused, but the same time, you have to be very agile and flexible, which means that you have to have a thesis that you because there wasn't an opportunity or because you just imagined it in your garage, you ⁓ start with something which you go through and you try. ⁓ building something, but you need to have your eyes and ears open for opportunities that comes along because it's not the perfect random walk, so to speak, ⁓ approach. And it's not the one that tells you no, you just need to, decided to go for something double down only on that. I think it's a combination and it's a very tough combination because it's very much like where talk about actual product, planning, again, not product vision, product planning. The hardest task is actually to decide what not to do. And that, applies also to opportunities. When you start working on something, you get a certain traction, but suddenly you see something else which looks maybe much more attractive. And then you need to decide, do I actually go there or do I stick to what I do? And ⁓ is no, I think that there is like a...
Yair Nativ: Yeah. Yeah.
Guy Gamzu: a framework which one can adopt and will guide you exactly to where you should be heading. I think that whatever you do, you need to be very cautious of not swinging the wheel, so to speak, too because may find yourself actually flipping to the side of the road. So I would say that even when you think you divert a little bit, you need to do it ⁓ very
Yair Nativ: here any last questions guy any last insights for us?
Nir: No, I'm fascinated by what I heard.
Guy Gamzu: I think that, again, be participant in this wave of technology change is already a So ⁓ anyone who is pursuing it is going to have ⁓ a very teaching experience, on a very philosophical level. And note, I believe that each of us is actually on a journey to become better and more curious about ⁓ we are doing. And that's why I think that that's actually a great opportunity for everyone to become a participant in this I think that we shouldn't take success or failure that is ⁓ being measured ⁓ economic or capitalistic as the only parameter for what we do. Obviously, that's great and that's a motivation for everyone. We can talk about that for hours, but put it I think that there is a real gain in trying and attempting to build things and to figure out things. And that's why I think this is... ⁓ and something that anyone who capable of joining should try and do that.
Yair Nativ: And so that's a good note to end with. Thank you so much for taking the time to join us