Latest / State of the AI Union / The Future of AI: Insights from Recursive Ventures
Transcript
- 0:10Hello, everybody, and welcome to this week's episode of the State
- 0:13of the EI Union. I am your host, Laura.
- 0:16And join with me today is Itamar Novik of Recursive Ventures.
- 0:21Itamar, welcome. Would you like to introduce
- 0:23yourself? Yes, and I'm so, so excited to
- 0:25be here because and thanks everybody for listening because
- 0:29probably most of your listeners don't know that you and I used
- 0:31to work together for an extended period of time at Gigya, which
- 0:35was a very successful and fun ride.
- 0:37And you were doing magnificent job there running customer
- 0:41success and other parts of the business including go to market.
- 0:44And since you've become such an amazing go to market
- 0:47professional, like Laura is the first person I call and I'm
- 0:50like, oh, I have a company that's right to figure out how
- 0:52to do like sales strategy and sales go to the market and all
- 0:55those things. You've become such an amazing
- 0:57force in the industry. So, so excited to be here today.
- 1:00Wow, I wasn't expecting that, but thank you so much for the
- 1:03shout out, Edamar. Hey.
- 1:04You're also. You're also mentor and advisor
- 1:08to our companies that recursive ventures, so it's going to be a
- 1:11good reason why I hired you to get that job done.
- 1:13Right. All right, Well, tell the
- 1:14audience about yourself tomorrow.
- 1:15What are you up to today? Yeah, sure.
- 1:17So, okay, so really quickly about me, I've been in tech for
- 1:20my entire career, which is 25 years.
- 1:21I'm a repeat entrepreneur and startup executive and I am also
- 1:26an investor. So my last company like 360 I, I
- 1:29was there for like 13 years, sort of from the early stages
- 1:33all the way to the IPO and beyond.
- 1:34Now it's a publicly traded company, actually crossed a $7
- 1:37billion market cap. So that that's, that's kind of
- 1:41my last company. Before that, I was a gig
- 1:43together with you and eventually the company got sold to SAP for
- 1:46$350 million. And I've done a bunch of
- 1:49different roles in startups and bigger companies across, you
- 1:52know, technical engineering, products, business development,
- 1:55finance and so on and so forth. I'm also an investor, so I've
- 1:58been investing in startups since 2010.
- 2:00I've got over 150 companies in My Portfolio including two DECA
- 2:05corns, another 5 unicorns and sort of another 15 successful
- 2:08exits M and as kind of sub billion dollar outcomes.
- 2:11And these days I run a fund called Recursive Ventures
- 2:14Cursive, which is a pre seed seed fund, $30 million is our
- 2:20last fund and we invest between half a million, $1,000,000 in
- 2:23sort of, you know, emerging technologies.
- 2:25For the last five years actually more, I've been really focused
- 2:28on data and AI. So for me, this kind of amazing
- 2:32AI wave, which we've been seeing for the last three years since
- 2:35the introduction of ChatGPT is sort of, you know, an amazing
- 2:38opportunity to spread my wings and support more founders as
- 2:42they build AI first companies. Awesome.
- 2:45Thank you. And I, I love you know what you
- 2:46said about like you're, you're focusing right now on data and
- 2:51AI and I think those two things come together.
- 2:54Tell me more about how you decided to narrow in your focus
- 2:57on data and AI together. Yeah, absolutely.
- 3:00So first of all, I've got an extensive background in this
- 3:02space. I actually started my career as
- 3:03a data engineer writing stuff like PL. sequel queries and,
- 3:08and, and all sorts of fun stuff back in the day.
- 3:11Then I built $100 million data business at Live 360 and I kind
- 3:15of got to know the data mafia, if you will, in the US.
- 3:18And even in the pre generative AI sort of era when we had a lot
- 3:23of like natural language processing, computer vision.
- 3:26And now obviously those things have been massively accelerated
- 3:29with generative AI. It was very clear that kind of
- 3:33your AI is only as good as the underlying data, right?
- 3:36That's kind of the key thing. So as I started investing in
- 3:40more and more DAS data as a service companies, Placer and
- 3:44Safegraph and and others who became very successful.
- 3:48I started seeing back in 2014, 2015 timeframe that there's
- 3:52really like AI companies starting to get getting built on
- 3:55top of this data, which is unique data, very high quality
- 3:58data and so on and so forth of a role data played in those AI
- 4:02companies. And through that, I started
- 4:04making more and more AI investments as well as I
- 4:07understood. So the relationship between the
- 4:08two. And it was actually already,
- 4:11I've already launched Recursive Ventures 3, which is my latest
- 4:14fund with a data and AI focus and then ChatGPT habit.
- 4:18So perfect timer. Yeah, yeah, yeah.
- 4:20That's awesome. And as a disclaim, I'm part of
- 4:23Recursive 3, so like that's helpful for me.
- 4:25But I'm sure that you get pitched like 100 times a day
- 4:29from AI startups that are like popping up everywhere nowadays,
- 4:33right? So for you when you're looking
- 4:36to evaluate these AI companies, what are some of the top signals
- 4:41you you are looking for that determine whether or not a
- 4:44company as like real staying power?
- 4:46Yeah, that's a great question. So let me start by covering the
- 4:49basics, you know, of what we're looking for in pre suit and seed
- 4:52stage companies in general. And then we can dive more
- 4:55specifically into the AI sort of defendability mode side because
- 4:58it's that is actually a core part of our thesis in ventures.
- 5:03So first of all, obviously there's the team.
- 5:05The team has to be complete in the sense that there's all the
- 5:09right DNA, all the right pillars to drive this company to
- 5:12success. Since you and I have spent a lot
- 5:14of time in sort of enterprise settings, enterprise sales
- 5:17cycles, we know that we got to have somebody who really knows
- 5:19how to sell, right? Selling more technical products,
- 5:22less technical products, it doesn't matter.
- 5:24Somebody who knows how to run an enterprise sales cycle, which is
- 5:27obviously can be lengthy and very time consumed and very
- 5:30resource consumed. So that's one piece.
- 5:32The second piece is obviously the technology.
- 5:34We are, you know, investing and supporting tech companies.
- 5:37So it's technology and product are key.
- 5:39And I think 1 interesting sort of point here that I like making
- 5:43is that what was is not necessarily what will be.
- 5:46So for example, these days when I see a very experienced CTO
- 5:50founder who was built big systems in web or mobile, that
- 5:55doesn't mean that they're necessarily going to translate
- 5:57to be a great AICDO, right? We're kind of looking for people
- 6:01that really are like up to speed and have been up to speed with
- 6:04the latest sort of thinking around generative AI.
- 6:07Because I think one thing that generative AI does other than
- 6:10vibe coding, which is obviously a paradigm shift in like
- 6:13productivity and how quickly you can deploy software and, and QA
- 6:18it and all that, that's kind of people get the memo on on vibe
- 6:21coding, but I think they don't necessarily get the memo on how
- 6:24the entire architecture of software is being rewritten with
- 6:28AI now we've moved from like thinking about APIs to thinking
- 6:32about MCP and agents talking to agents.
- 6:34And that might be the future of software architecture.
- 6:38So if you're coming in as a CTO with like some of those, like
- 6:41all the paradigms, I don't know if you'll be able to build the
- 6:45right technology for the next generation of products.
- 6:47So, so looking at the technical side where we're looking for
- 6:50people that have stayed up to speed and are up to date on
- 6:53everything that's happening now as it's shifting so quickly.
- 6:56So that's on the team side. Then we have Tam, which is
- 6:59totally addressable market. That's always been a pillar of
- 7:02VC investing and that hasn't changed in the AI era, right.
- 7:05If you, you want to build a venture backed company, you want
- 7:09to shoot for the stars, you want to shoot for a company that can
- 7:12have a billion or more dollars in revenue.
- 7:15So it can be a $10 billion company, because that's really
- 7:19what you need to eventually justify continuous VC
- 7:23investments up until an IPO or a very big M&A.
- 7:26So that Tam piece is something that we've always checked for
- 7:29and we'll keep checking for even in the AI area.
- 7:31Now getting to the last piece, which is actually what you asked
- 7:33about, but I wanted to cover the that was important is the Moat.
- 7:38So we have now gone through this wave of companies and some of
- 7:43them are in vibe coding, some of them are in customer support and
- 7:46customer success and sales tools and so on and so forth, where
- 7:49we're talking about fin wrappers on top of Open AI and Tropic.
- 7:54And that's obviously a concern because what happens with
- 7:58startups is, is, is you kind of you got to make sure that you
- 8:01can go all the way. And if you don't have a mode, if
- 8:04you don't defend ability and other entrance, whether they're
- 8:07incumbents or smaller start up companies can enter your space
- 8:11and replicate the copy of product in a couple of months.
- 8:14So it's not a defendable product or technology that's an issue
- 8:18because what happens when you have all those entrance is
- 8:20basically it becomes becomes a race to the bottom right.
- 8:24In venture capital and startups in general, we like companies
- 8:27that can command prices, that can command margins, right?
- 8:30Like high operating margins. This whole thing about like, you
- 8:34know, very, very valuable technology companies is building
- 8:38the fact that the marginal cost of servicing one more user
- 8:43trends to zero and then you can command high prices.
- 8:47So you can ink very profitable companies, right, that pay
- 8:51dividends when they're big in public.
- 8:52So if there is a lot of competition and it's easy to
- 8:55copy you because you're a fin rapper, that's an issue in
- 8:58startups. Use equal stop at some point.
- 9:00It can stop at your seed, it can stop at your Series B, maybe it
- 9:04stops after the IBM, but we need to make sure as VCs that it can
- 9:08go. All right, right.
- 9:09And I, I think I remember here like reading one of your
- 9:12LinkedIn polls where you said, you know, the companies that you
- 9:15invest in are not those that are the best at doing something.
- 9:19They're the only ones that do it and the only ones that can do
- 9:22it, right? So that's what you're talking
- 9:24about here? Yeah.
- 9:25So that's that's kind of our preference.
- 9:26And then peeling the onion one more level on that is what does
- 9:30that mean? What does it mean to have a Moat
- 9:33right in, in, in AI? And a lot of people are kind of
- 9:36like, we don't know there's no Moat.
- 9:39I think there is. And I think there's going to be
- 9:41multiple, there are multiple angles that my companies are
- 9:45deploying today to create that Moat, to create that
- 9:48differentiation and it's working.
- 9:50So we can dive more into that if you want.
- 9:52Tell us, what are some of the moats that you're seeing right
- 9:55now that are really defensible? Yeah.
- 9:56Oh, absolutely. So the obvious one is obviously
- 9:59having proprietary data. When you have data that other
- 10:02people don't have, and that data is highly relevant for the kind
- 10:05of tasks that your AI or your agents are trying to accomplish
- 10:08at the enterprise, that makes a world of difference because you
- 10:11have a much higher accuracy, much higher quality, right in
- 10:15your ability to execute those tasks.
- 10:16Proprietor data actually happens through many different ways.
- 10:19Sometimes it's a data set that only you, your company has.
- 10:22Sometimes you strike amazing business development deals with
- 10:26other folks to buy their data with exclusivity.
- 10:28And that's also that exclusivity gives you that edge, right?
- 10:31That proprietor that other people don't have at other
- 10:34times. And a lot of my companies are
- 10:36doing that. They're actually really, really
- 10:38good at picking up and even more so getting reinforced
- 10:43reinforcement learning data from their customers, right?
- 10:47So one thing is tapping into data that your customers have
- 10:52and that obviously is really important in AI so you can
- 10:55provide them better services and things that are really like
- 10:57agents or models or SAS software built with agents and models
- 11:02that really fits what they need, their playbook, their workflows,
- 11:06right? So that's kind of obvious, but
- 11:08it needs to be done well. The other thing that people
- 11:10don't think about as much is if you create a virtuous experience
- 11:15with your enterprise customers where they're feeding you with
- 11:19this is good, this is wrong. This is where your agent should
- 11:22get better. This is where your agent should
- 11:24have stopped and asked me the question of whether they should
- 11:26proceed with task ABC or not. If you do a really good job of
- 11:30picking that, you're basically getting your users to label your
- 11:35data for you. And then with forced learning
- 11:38techniques, you can improve. You can create this virtuous
- 11:42cycle where your software is getting better and better and
- 11:46better, and that's very powerful.
- 11:47And I'll explain why if you deploy, let's for simplicity,
- 11:50let's not focus on agents because that's a more
- 11:52complicated thing. But let's just think about a
- 11:54customer, an enterprise customer using a a piece of software that
- 11:59uses an underlying model to deliver a response, right?
- 12:02To do whatever analytics inside that company, right?
- 12:05If you have reinforced learning, if you are iterating and
- 12:08improving the underlying model, for fine tuning, for better rag,
- 12:12for better knowledge graphs based on the actual experience,
- 12:16based on the actual work the customer is doing, you're
- 12:18creating a very defendable piece of software because the next
- 12:21vendor that would come in would have to start from version one
- 12:23point O you're already in version 3 because you have two
- 12:27years of data of your customer using the product, making your
- 12:30model bend for them. So I'm hearing in this that like
- 12:35the customers are the ones that actually help the model become
- 12:39better. That's one.
- 12:40And then the second thing that is basically like the first two
- 12:44market advantage where somebody else could replicate it, but
- 12:48because they don't have the same customers or you know, it's
- 12:51going to take a longer time, then you know, that's how it it
- 12:54makes them defensible. And it's very interesting,
- 12:57especially for somebody like you, Laura, who's a go to market
- 13:00experts because actually a very successful go to market can
- 13:03essentially lock up the customers, which locks up the
- 13:07data crowning a winner. Yeah, and makes the makes the
- 13:10customer sticky too. Right.
- 13:12It makes the customer sticky because you have a better model,
- 13:15because you have more data, because you acquired customers
- 13:17faster than anybody else. Just think about it.
- 13:19So now it's not just a race to scale, it's a race to who gets
- 13:25the most customer data in the vertical the fastest.
- 13:28First, Yeah, yeah, yeah. Which is very interesting.
- 13:32It's an interesting twist on SaaS, right?
- 13:35We did, we didn't race. We'd go to market to get as many
- 13:38customers in the SAS era because we want to have case studies and
- 13:42success stories or we have this customer.
- 13:43So it's going to get incrementally easier to get the
- 13:46next customer. Once you have customers right
- 13:48now that kind of dynamics is reinforced with with customer
- 13:52data. So it's very interesting.
- 13:55There's more modes hang, hang in there.
- 13:57Another very important one is how good is the team in actually
- 14:01leveraging the data? And this is not, this is more
- 14:05technical, right? And it also depends on your
- 14:07ability to acquire different data points, understand the
- 14:11relationship between the data that you acquire and the
- 14:14problems that you're solving, right?
- 14:17And then leverage cutting edge techniques to do a better job at
- 14:21fine tuning models. A lot of companies out there are
- 14:25just glorified rags, right? They just know how to pull some
- 14:27customer data and then call open AI, entropic, Gemini, whatever
- 14:32it is and in a response back. I think that could work in some
- 14:36use cases, but in others it only gets you so far right.
- 14:39And my, I guess more technology savvy companies, they actually
- 14:44do much more than that. They go and do fine tuning,
- 14:47which means a couple of different things.
- 14:48You know, it means that they use open source models also open
- 14:52source data and they're taking data out in post training,
- 14:56right, in fine tuning and they're putting new data in.
- 14:58So that kind of overtime really changes that base open source
- 15:03model that they used into what they need to win the vertical
- 15:06that they're in and the quality and your ability to do that
- 15:09well. So this is data science would
- 15:11absolutely impact the quality of your product.
- 15:13So that could be a Moat. If you have a great technical
- 15:15team that knows how to fine tune well, knows how to build those
- 15:18models, you could be winning over the competition.
- 15:21So the adult technical skills. The last thing I would say, and
- 15:25you know, I've got examples in My Portfolio is I think there
- 15:27are a lot of different interesting models that are out
- 15:30there around data consortiums and open source, right.
- 15:33So for example, I have a company called Clarity that does you
- 15:36have a community of they're in the deep fake business.
- 15:38They're trying to take deep, deep fake videos, which is
- 15:40obviously a big growing global problem.
- 15:43And they massed community hundreds of times, scientists
- 15:47globally from all sorts of organizations, from universities
- 15:50to research organizations and so on and so forth, who really are
- 15:53passionate about solving this Deepik problem.
- 15:55And they're all building their own models and they're
- 15:57collecting their own data. So one, one thing here that's
- 16:00interesting is how they're creating that community of
- 16:03scientists who are creating data and models and then building a
- 16:08commercial service on top of that.
- 16:09It's like the next generation of open source, if you will, right?
- 16:13And again, that creates a mode because the next company that
- 16:16comes over doesn't have 100 data scientists working for them for
- 16:19free as part of this open source movement to solve this problem.
- 16:23So it is, it is significant defendability they're bringing
- 16:26through that. So just giving that as an
- 16:27example. And there's also data consortium
- 16:29examples, which is an area that I know, you know well at that
- 16:33could, you know, make a, make a dent in creating that mode.
- 16:36There's more around that, but those are the main things that
- 16:39I'm seeing these days, OK? I wanted to actually go back to
- 16:42what you said about like the race to the race to getting more
- 16:46customers. What and I, I think, I think
- 16:51that, you know, there's a long tail of customers that we can go
- 16:54after just like any kind of SAS product, right.
- 16:56But when we are, you know, thinking about training models
- 16:59and, and stuff like that, we want the enterprise customers
- 17:01they have like the, the big scalable data that we can use
- 17:05and that we we can, we can find you.
- 17:07What are some challenges you are seeing with your founders when
- 17:11they're trying to go after their first enterprise customers?
- 17:14Well, I mean, I'm going to skip the obvious challenges of seed
- 17:18stage companies going after enterprise customers because we
- 17:21know, we all know how that works.
- 17:22You know, between that can I trust a small startup to sock 2
- 17:26and compliance and complexity and professional services which.
- 17:30All of those are there, and let's say all of those are
- 17:32there. You, you know all that.
- 17:34So your audience knows all that as well because they're go to
- 17:36market experts. So I think with AI specifically,
- 17:39there's a lot of mistrust out there, right?
- 17:42Like on one hand, people get it, There's a lot of decision
- 17:45makers. They're just budget for AI.
- 17:47People get that AI is a must now generally and externally for
- 17:51organizations. People managers, CEOs understand
- 17:55that if they don't jump on the AI wagon, their company is at
- 17:58risk. And it's a very acute risk.
- 18:01And they understand that on one hand.
- 18:03On the other hand, the technology in many areas is just
- 18:06unmature enough. Too many issues, quality issues,
- 18:10hallucinations, overfit, you know, compliance issues,
- 18:15concerns would trace like how how did this data get here?
- 18:18Or especially if it's a B the B the C like something that is
- 18:22being sold to an enterprise customer that then is being used
- 18:25with a consumer. Is this thing going to hurt our
- 18:27companies reputation? Is it going to be able to, you
- 18:30know, meet and exceed our brand guidelines, right, in terms of
- 18:35engagement with customers? So there's all those quality
- 18:38related questions. I kind of broadly called him
- 18:40quality, but obviously there's a lot of no one's here and it
- 18:43comes down to can we trust this thing?
- 18:46And I think the answer in quite a few solution verticals today
- 18:50and for quite a few buyers out there in the market is that
- 18:54often the answer is no, not yet. We need humans overseeing it.
- 18:58We need humans in the loop. We need, we need to have KPIs
- 19:02around quality and we need to see improvement in those so we
- 19:05can trust the AI. So I think for a lot of my
- 19:08enterprise companies, and again, there was also very early stage
- 19:11companies, you have to be very smart about how you have these
- 19:14conversations with prospective customers, right?
- 19:17It's like what how can you drive trust?
- 19:21How do you show to the customer that this thing is going to
- 19:24improve over time? Because some of the improvement
- 19:27is driven by the underlying technology, not necessarily by
- 19:29the layers that this company builds, right?
- 19:32Some of the technology improvement needs to happen with
- 19:34open air, right? My startups can't necessarily
- 19:36solve them, but they'll get there, right?
- 19:38So you almost need to be able to show chart like a road map with
- 19:43your customers that builds trust and confidence with this new
- 19:46technology. And do you think that means like
- 19:49that, like, I mean, a lot of buyers that are tasked with, you
- 19:54know, adopting AI initiatives, They, they are, they're business
- 19:56people like me and I'm not technical.
- 19:58Does that mean that like you are also recommending, Hey, try to
- 20:02get somebody technical in the deal to evaluate the software, I
- 20:06mean, or the the AI or, you know, evaluate your, your
- 20:10different shaders upfront. Is it?
- 20:13Do more Pocs? What does it mean sort of
- 20:15practically? Yeah.
- 20:16So I don't think it necessarily means that you need to have more
- 20:18technical people in the sort of procurement discussion.
- 20:21No, not necessarily, because the goals you're actually actually
- 20:24trying to hit with a lot of those systems are business
- 20:25goals. It's like your KPIs.
- 20:28You need to have this response rate, you need to have this
- 20:30quality rate, you need to have this recall, whatever it is, you
- 20:33as a business person, you know, what's the quality bar you're
- 20:35trying to meet? Just a random example, I have a
- 20:38company in My Portfolio that does that.
- 20:40Does it sell? They sell hot like big ticket
- 20:43items through texting and WhatsApp and messaging, right?
- 20:47And it's in lay, it's replacing salespeople who are texting
- 20:50customers. They're asking questions about
- 20:52this mattress or about this insurance plan, right?
- 20:54So you know that today you have KPI's around that it's like,
- 20:59what's the time to respond? What's the NPS score, whatever
- 21:02it is, right? Like what's the conversion?
- 21:04You know what the metrics that drive success are.
- 21:06You just need to make sure now that the AI systems replacing or
- 21:10augmenting humans doing that work meet or exceed the same
- 21:14case, the same goals. So I don't think any technical
- 21:17people necessarily, you absolutely need POC's and you
- 21:21also need POC's because of the second thing that I wanted to
- 21:23bring up. And that is a major, major
- 21:26threat in the eye, which I think people are not talking about as
- 21:29much. We can do as much as we want
- 21:31with like advancing the technology itself, better
- 21:34models, better agent infrastructure and frameworks.
- 21:38Great. That will get us only so far
- 21:40because the other piece absolutely has to happen and
- 21:43that is human adoption of AI. And humans move at the speed and
- 21:48pace of humans move a billion miles an hour.
- 21:51I don't care. I still need to embrace it.
- 21:54And that is going to be a process that is going to take
- 21:56multiple years. And that process, the process of
- 21:59building the right human machine interaction, what is the
- 22:03experience? When does the agent stop and ask
- 22:06a question? What does the agent show when it
- 22:09like shows you its plan of operation?
- 22:11What does it show? How much detail when do we need
- 22:15a human in the loop? When do we don't need the human
- 22:17in the loop? And this is not a generic
- 22:19question in the sense that you can apply it once.
- 22:22You have to apply it to every single use case in workflow that
- 22:25AI is going to solve, which is a massive problem.
- 22:28And it's also not something that you can accelerate with AI.
- 22:31You have to have product managers and designers and
- 22:35engineers solve this problem of like, OK, so my customer wants
- 22:40an AISDR. How does that look? What is the dashboard?
- 22:45What is the interface? How is it going to work with
- 22:48account executives? How is it going to work with
- 22:50existing SDRS? We don't have good answers for
- 22:54all those questions yet. So this is this is a race to
- 22:57figure out two things. One is the technology, add
- 22:59technology, we're going to figure it out.
- 23:00But the second thing, which will go in its own pace is that human
- 23:03machine interaction. And that is a very critical
- 23:06piece. I want to tie back to your
- 23:08question around startups and companies in general selling AI
- 23:11solutions to customers. They need to think about the UX.
- 23:14They need to deeply think about the UX to really Dr. adoption
- 23:18and success with customers much more than in the SAS era because
- 23:21this thing works completely differently.
- 23:23And you know what? Like when you were talking about
- 23:25that, I was just thinking that sometimes when buyers come,
- 23:30sorry, when, when vendors come and they go, we're going to
- 23:32solve over your problems for you.
- 23:34And because our AI can do this and they're not being thoughtful
- 23:37about this whole change management process that, you
- 23:41know, you're, you're describing, it can feel very insensitive.
- 23:44But more than that, I think it puts a lot of burden on the
- 23:47buyer to like have to think of those things and design all
- 23:50those things and their buying cycle is around actually the
- 23:53design of what it's going to look like.
- 23:56So I would say, I would say just to echo your point, like good,
- 24:00good AI sellers are going to be able to address the change
- 24:03management in a way that's designing like what you said,
- 24:06like the user experience and the change adoption, right.
- 24:10So maybe that means ramp up times are going to be slower,
- 24:14you know, because customers will will pay as they go or something
- 24:17like that. Yeah, and a lot of it is helping
- 24:19customers get up to speed and educating them on how to use
- 24:21those AI tools, right. So part of it is UX, part of it
- 24:24is education, part of it is natural transition.
- 24:26Part of it is going to be rescaling, upscaling, and people
- 24:29getting kicked out of the workforce while younger, more AI
- 24:31savvy people come in. It's a process.
- 24:33So like whoever is telling you that AGI is going to take over
- 24:36the world tomorrow, I think it's a little bit delusional, at
- 24:39least on the end of by software side of things.
- 24:41I think it's going to be a 5 to 10 year cycle.
- 24:43Not because the technology can build it fast enough, because
- 24:45again, as as humans we need to adjust to this thing and it
- 24:50will. It will.
- 24:50Take that awesome. Let's let's talk about a little
- 24:53bit about sort of the market saturation under over saturation
- 24:57and some valuations. How do you think about those two
- 25:00things? Are we, are we over saturated
- 25:02right now with AI startups? Are we overvalued?
- 25:05No, I well, two different questions, so let me address the
- 25:08first one. So I am in the camp of So tech
- 25:12today is 15% of global GDP 15. I think tech plus AI could be
- 25:1750% of global. GP Oh my God, OK 50%.
- 25:21Well, think about it, Drivers. Analysts.
- 25:24Sure. Engineers, lawyers, accountants.
- 25:28Should they keep going? No, I can.
- 25:29Do every one of those verticals is going to go for a
- 25:32transformation that has some element of AI replacing humans?
- 25:37I'm going to go back to being a chef.
- 25:38I think that's sort of AI proof. And there's no AI is going to
- 25:41replace your amazing restaurant with you being a chef.
- 25:45I think that's actually a good plan because there's going to be
- 25:47an AI agent, go to market executive that's going to
- 25:49replace you, Laura. Yeah.
- 25:52No, but here's like every, almost everything that we humans
- 25:56touch is like work or productive stuff that we do is going to be
- 25:59like most of the work or all of the work is going to be replaced
- 26:02by AI or we're going to be able to do 2 to 5X more work because
- 26:05this AI is going to make us so much more productive.
- 26:08But it's one or the other. So if you're in my camp and you
- 26:13believe AI can be 50% of global GDP, then I could put a stake in
- 26:18the ground and say, no, we don't have enough companies doing AI
- 26:20because it's such a huge opportunity.
- 26:22I mean, Mark Andreessen coined the $10 trillion opportunity.
- 26:24So like if you have just trying to scope this, if you have the
- 26:28Magic 7 now and they're like, you know, trillion plus dollar
- 26:30companies and they're all tech companies, most of them in
- 26:32Silicon Valley, right? What I'm saying is we might have
- 26:34thirty of these and they're going to be worth $10 trillion
- 26:37each, not one trillion. And by the way, it's all
- 26:39happening in Silicon Valley, but that's a different conversation.
- 26:41So you and I are in the right place at the right time
- 26:44potentially. But but if you believe that,
- 26:46then yeah, maybe there aren't enough companies, maybe we
- 26:49should have more. And I think so.
- 26:51I have this data somewhere. The number of startups,
- 26:54companies being created on an annual basis has been growing,
- 26:58pretty growing consistently, but not at a fast-paced for the last
- 27:0315 years and that hasn't changed a lot in the last two years.
- 27:06So I don't think we have necessarily too many companies.
- 27:10It's just everybody's doing AI now.
- 27:11So that's why it feels like maybe there's a saturation, but
- 27:14I don't think there is in terms of valuations.
- 27:17We're in the land of the have and have nots.
- 27:20So we have the LMS, right, The open a eyes of the world
- 27:24entropic who are getting this stratospheric set valuations,
- 27:27which are maybe justifiable, maybe not.
- 27:29I don't even want to comment on that because it's such AI don't
- 27:32I don't deal with that day-to-day, right.
- 27:34I'm an early stage investor. Those are very late stage
- 27:36companies. But, but, but I think you know,
- 27:38at the application layer, which is what I focus on enterprise AI
- 27:41first applications, I think they're the valuations are
- 27:45pretty much what they were in 2020-2021 stage.
- 27:48So like era, so like really good founders.
- 27:52Good valuations, I mean those, those are good valuations
- 27:5520/20/2020. One, yeah, they're pretty high.
- 27:57They're very founder friendly, but they still work for VCs.
- 28:01Like I know my model and I'm like, OK, if it's a steel
- 28:04entrepreneur and they're proven, I would pay that 20 to $30
- 28:07million post on the 1st institutional round.
- 28:10If it's a less experienced the preneur, I'll pay the 8 to $15
- 28:13million post in sort of the first round that I invested and
- 28:17my model works that way. It doesn't break if the
- 28:21valuations double, yes, it will break and I will have to change
- 28:24strategy, but the current valuations are high yet
- 28:27sustainable. That's kind of my conclusion.
- 28:29You know, I don't know if you heard this that, but like in
- 28:332025, you know, there were like some billion dollars that were
- 28:37going into AIAI funded funded companies, but really most of it
- 28:44right was going into like what you're saying, like Anthropic,
- 28:46the LLS. By far.
- 28:48I think there's a lot of confusion with financing.
- 28:50We're talking, let's talk apples and apples.
- 28:52There is, you know, 10, the magnificent 710-2050 companies
- 28:57globally who are investing massively in AI infrastructure
- 29:01and compute a trillion dollars. I don't even know how much is it
- 29:06over investment on that? I don't know.
- 29:08And I actually don't care because it it's not what I'm
- 29:10talking about. Same goes to the sort of the
- 29:14infrastructure, the stat that you need to build AI
- 29:16applications, whether it's LMS, whether it's data
- 29:18infrastructure, whether it's tooling, all that.
- 29:21I think there's been a lot of money that goes into that and
- 29:24it's very, very high valuations like like you mentioned.
- 29:26And I think that's sort of been played out as we speak.
- 29:29What I'm talking about is this, this layer of like the last mile
- 29:33that the application layer, which is actually where most of
- 29:36the value is going to be created.
- 29:38I think they're the amount of capital being invested and the
- 29:41valuations are same. They're not crazy like what you
- 29:45see in those two other areas that I mentioned.
- 29:47And I think they will actually be accelerating over the next
- 29:50few years because now that we have the infrastructure now
- 29:52again, it's going to take us 5 to 10 years to figure out the
- 29:54applications. Yes.
- 29:56And at the application layer, as you're talking to these
- 29:59founders, has it been like more difficult given today's market
- 30:04conditions to like raise money or is or is the money coming in
- 30:08for them because it's AI and people interested in?
- 30:11Yeah, I think it's a little bit again the have and have nots.
- 30:15So if you have some significant track record in AI, you're
- 30:21spinning out of open AI, you know, you work to Google
- 30:23DeepMind, whatever it is, then obviously you're, you're, you're
- 30:28a hot founder and you'll get funded at very high valuations.
- 30:31And then there's a lot of people who don't necessarily, you know,
- 30:34fit into that bucket and could struggle raising capital.
- 30:39But in general, this is one of the best times that I've seen in
- 30:42my 15 years of investing to start companies.
- 30:45The capital is there if you're focused on the rights of the
- 30:47problems, which is obviously for many folks, AI and AI
- 30:50applications. And I think it is easier to get
- 30:54funding now than was through the majority of the 20 tens, right?
- 30:59It's probably very similar to what we saw in 2020-2021.
- 31:02So you can call it maybe a little bit of a bubble, but if I
- 31:06personally think this time around the bubble is justifiable
- 31:10because this is so transformational across every
- 31:13vertical of business that we know.
- 31:15Awesome. So, Itamar, thanks so much for
- 31:17being here. I have a final question for you.
- 31:19It's a little bit of a curveball here, but if we were doing this
- 31:24podcast again in 2030, which sounds like a long time ago
- 31:28away, but it's really only five years away, what do you think
- 31:30would be saying about it about today's podcast?
- 31:34And what do you think would surprise us the most?
- 31:36I think we would be very surprised by the fact that some
- 31:39of these companies that we we keep talking about today, like
- 31:42open the eye in a traffic and and so on and so forth are going
- 31:45to be commodities and we're not going to care about them.
- 31:48And we're going to talk about a completely different set of
- 31:50companies. We don't even know their names
- 31:53yet. And it's a little bit like what
- 31:56happened back in 1999, like the bubble days where everybody was
- 32:00talking about like Webvan. Webvan is going to come and
- 32:03they're going to deliver, you know, groceries to your house.
- 32:06And it took a while, but at the end it, it happened with
- 32:09Instacart and we didn't think about Instacart.
- 32:11We didn't think about Instacart that way.
- 32:13We didn't understand that we had that mobile and location and
- 32:17dynamic workforce and then this thing would work.
- 32:19So I think you're going to see something similar.
- 32:21I think we might see some very transformational things like who
- 32:26knows, maybe Salesforce won't be the leader in CRM anymore
- 32:29because somebody will figure out how the leader in CRM actually
- 32:32looks and it's going to be in the AI first CRM, right, that
- 32:35knows how to connect all those dots.
- 32:36And so I think, I think that's what we we'll be talking about
- 32:39in 20-30. We want like open the eye would
- 32:40be like an after father would be like, oh wow, here's like 1020
- 32:44amazing companies that didn't even exist in 2025 that are now
- 32:47have multi, multi, multi billion dollar market caps.
- 32:50And I've completely changed how we do business in the verticals
- 32:53that they operate in. And I think it's going to happen
- 32:55in the next five years. Yeah, I believe you.
- 32:58I do, actually. Thank you so much for joining
- 33:00me. Any last words for our audience
- 33:02before we close out here? Yeah, I guess.
- 33:04You know, I always love giving this advice just because I have
- 33:07received the same advice and was the best advice I ever got.
- 33:10So I was working on the Internet and a person that I really
- 33:14respect back in the day you and I were working on Internet stuff
- 33:16together told me in the mark, whatever you do, you have to
- 33:19jump in the next wave. And back then it was mobile.
- 33:21So I was lucky enough to take that advice and I learned a lot
- 33:25from it in tech. The most important thing that
- 33:27you can do professionally to grow as you know, as a
- 33:30professional, as a founder, as an executive, as an IC, it
- 33:33doesn't matter is when you see a huge new technology wave that's
- 33:37going to disrupt and completely change the world, you got to
- 33:39jump on it. You got to jump on it early.
- 33:42And guess what? If you, if you, if you get a pay
- 33:44cut, if you get slightly demoted, that might actually be
- 33:47a very good thing for you in the long run because if you're
- 33:49jumping the next wave of technology, you'll be more
- 33:51relevant than ever and your ability to create value could
- 33:56significantly increase. So that's kind of the last word
- 33:59of advice and I'm very happy for you though transitioning into
- 34:02being an expert with AI, go to market, you're jumping on that
- 34:06wagon as early as possible and you're doing a great job at
- 34:09that. So congratulations.
- 34:10Thank you so much. All right, thank you, Itamar.
- 34:13Thanks for being here. To our audience.
- 34:16If you enjoyed this content, please leave a comment, give a
- 34:18like so that we can give you more of it.
- 34:20All right, see you next time. Thank you so much.
- 34:22Bye, Laura. Thank you.
- 34:23Thanks for everybody listening. Join us again.