Niklas: Hi and huge welcome to you, my lovely listeners. So glad you're here. Today you're joining me for a chat with Martin. Martin is a serial entrepreneur turned pre-seed venture capitalist. He is the founder and managing partner of Incisive Ventures, a pre-seed fund focused exclusively on B2B software and technology companies that reduce friction at scale. Martin, so lovely to have you.

Martin Tobias: Hi, thanks for having me.

Niklas: Looking at your journey, find it really interesting because I think you have seen, I would say, all worlds, like the big corporates, Microsoft, Accenture, ⁓ you've built some startups yourself until IPO, so the full journey, and now you're an investor. Walk us through it a bit and also maybe what did you like at the different stages?

Martin Tobias: Well, you know, I was an early adopter computer computers. I'm pretty old. My first computer programming class, I was actually using a deck, Vax and a card reader. So a bug in a program was literally a bug in this stack of cards. ⁓ But very soon we got a Mac. to and we're programming on Pascal and stuff like that. But my interest in computer science and technology has always been how to solve business problems with computers and software. While my friends in CS in college were writing device drivers for keyboards and know, motherboards and stuff like that, I was trading stocks. So I've always been intrigued with how can this software stuff help people to get stuff done that they want to get done, not the purely technical things. And ⁓ that's why I went to Accenture because they were, you know, applying ⁓ computer programming to solve business problems doing AR, APGL installations, stuff like that. And that's also why I went to Microsoft because I thought they were doing a good job democratizing access to personal computing and everything, know, PC on every desk and all of that. ⁓ But I left ⁓ Microsoft because I had an idea to start a ⁓ software company called

Niklas: you

Martin Tobias: and I had made a bunch of money at Microsoft and frankly I was just getting a little bit frustrated with how big Microsoft had gotten. When I started at Microsoft I was three doors down from Bill Gates and when I left you know he was in an armed ⁓ office with himself and nobody could talk to him. It was just a very different place.

Niklas: you

Martin Tobias: So I decided to start a company and it was really great. mean, it was in three years, we went from starting the company to IPO and we were in fact the last company to go public in the dot com boom. went public March 17th of the, and the NASDAQ peaked on March 21st. So the good thing about bubbles, which we're in a little bit of one now with AI is you want to be inside them rather than outside them if you're ⁓ raising money. kind of got hooked on ⁓ doing a startup. And the reason I did the startup after Microsoft is when my last two years at Microsoft, I was working a lot with startups. And one of the things I recognized is that the people working in startups seem to be having more fun than me at the big company. That's why I did a startup. And I've kind of gotten hooked on that zero to 50 stage. And I started three companies, sold two of them, took one public. But for the last six years, I've been investing. I've always invested in companies.

Niklas: .

Martin Tobias: as an angel, did 250 investments before I started my venture fund. But the part I like is helping founders ⁓ find product market fit, getting, raising capital and ⁓ seeing, you know, their vision get to the next ⁓ stage. So that's why I'm doing pre-seed investing now.

Niklas: you There's probably also something you can learn from your time as a CEO. If you look back and you just had to talk about one thing where you think, okay, I wish I had known that before I started, what would it be? What would you recommend to your younger self?

Martin Tobias: ⁓ Well, there's a lot. The first one is don't believe your own hype. In the dot com boom, there were a lot of people believing their own hype, taking companies public with complete bullshit. You know, I was on the cover of Wired magazine one time. When you get that, it can fill your head and think you're the smartest person in the room. ⁓ You know, and when we went public, it was very successful, but we went public with $10 million of revenue and had a $2 billion market cap. So a P E of 200 and ⁓ I mean, obviously I'm happy with that P.E. as a shareholder, but I also was smart enough to realize that, you know, it was a stupid valuation. So ⁓ all the nice things people say about you, all the things you say about how great you're going to be for your customers. The number one recommendation I would say is don't eat your own dog food. Don't believe your own hype too much because it can get you into a weird ⁓ place.

Niklas: Interesting. Then you invested, I think, into a number of companies later on that become pretty large and that are really different. I looked at Google, DocuSign, a few more that are also in different spaces. What is the common thing that you saw when you met the founders for the first time or you saw the businesses?

Martin Tobias: Well, I didn't invest in Google directly. I was an LP in Ron Conway's Silicon Valley Angels, and he gave Larry and Sergey their first $250,000. And he gave us LPs shares in Google at the IPO. So I own some of those shares today, and my cost basis is $0.50. And the learning on Google was it's amazing to be in some of these transformative companies early, whether you invest yourself or whether you invest as an LP in somebody else's fund. is if you're an LP in somebody else's fund, you're going to get into a lot of companies you would have never seen. I would have never seen Google myself as an angel because I was sitting in Seattle and they were in San Francisco. But if you look at the company, the checks that I have written and the conviction, the reason I did about 250 angel investments before I started my fund and I invested in all sorts of things. And when you're an angel, generally you're like, oh, this guy seems smart. Here's money, 25, $50,000, whatever. And you write off the check and you move on. I went and analyzed those, the ones that did the best, companies like DocuSign, were actually building B2B, solving some business problem, and you know, that I knew something about. I could introduce them to customers, I could help them in product design, and that's why I have decided for the last six years to just focus on B2B software investing. So my insight, where I'm a very helpful and good investor picking and helping companies is in the B2B software. And that's something that as you invest over time that you're probably going to do. is say, for me, I should invest in this kind of thing. I don't invest in crypto. I don't invest in hardware. I don't invest in B2C, not because those might not be good businesses, but just because I'm not the right investor for those companies. So the ⁓ insight I had is when somebody is building something in B2B that I think I understand and can help with, that's a good investment for me.

Niklas: And if you look at first time angels, for example, somebody who gets into investing for the first time, what's typical mistakes? What should people avoid?

Martin Tobias: Well, the typical mistakes are they just write checks to their friends without doing any due diligence. they're not concerned about valuation and they don't think that, but the number one mistake most ⁓ angels make is that they don't think about the size of the round relative to traction and what is going to happen next. There is approximately a 80 % failure rate in angel investments. So if you have a bunch of money and you just want to write checks to your friends and you don't fucking care, fine, write those checks, but you're going to lose 80 % of them. And what, what change, me from an angel to being a real VC is one of the first things I think about before I write the check is how big is the round? Does it give them enough money to achieve the next natural milestones where the next investors are willing to take a meeting and give this company more money? Most angels don't think about the runway that they're giving, who's going to give the next round, make introductions to those next round investors. They're just like, good luck, here's $25,000, $50,000. That is a terrible Angel strategy because you're not thinking about what comes next. You're not giving them enough money to reach a reasonable set of milestones. The worst kind of angel investor and the worst kind of early investment is a company that's raising $25,000, $50,000 every two or three weeks and just paying its bills as it goes along. It never has enough money to actually achieve a set of milestones that would unlock the next set of investors. And I've seen angels do that far too many times. just because they like somebody.

Niklas: Yeah, that's interesting. On the other side, what tells you a founder could be great? Do you look more at ideas these days or more at the people you invest in? What's more important to you?

Martin Tobias: Well, in pre-seed and angel, it's really all about the people. And then the question is, what's your, know, how good are you at picking people? And I get asked this question a lot and it is still a little bit voodoo. I've met... 20,000 CEOs, I can tell you within five minutes if the CEO I'm talking to is in the top 10 % of CEOs I've ever met or not. How do I do it? I don't know. It's pattern matching, right? One of the other problems angels have is that they see too few companies relative to the opportunity set. And so their pattern matcher is generally not as good as a professional VC because they've just met less people. And, ⁓ but When I think about today, ⁓ what makes a great founder, the primary thing, because the cost of creating a software company, building any kind of technology has gone down and down. We can talk later about Claude code and all these things. maybe 10 years ago, having the right technical team, being able to get the right people to build the hard thing could have been a key success factor. That's not as important today. What's really important, what I spend a lot of time on is, ⁓ Does this founder or founding team understand the customer and have a distribution wedge to get their product into the market? Because right now it's so much easier to turn any idea into a product You're going to end up with a lot more competitors in no matter what the fuck your idea is So the real differentiator between a team that's going to succeed in this environment is their distribution What's their go-to-market? I invested in this company called procurable.ai and the CEO had been a VP of procurability for six years for a manufacturing company, the same companies that he was selling his solution to. He tried to buy every software. He couldn't find one that he liked. So he built his own. He was the chairman of a hundred person trade association that had a hundred other VPs of procurement. Those were his first hundred customers. So I got really convinced that he understood the customer problem. He had access to the first hundred customers. As long as he built the product, he could get meetings and get sales. ⁓ So it's the district. distribution wedge.

Niklas: Yeah, as you said, it has changed a lot. was always wondering whether it has changed this much or it has just become more obvious recently. I think software was always relatively easy to build, but you still needed an investment upfront to get to that point. Still, you could talk to the people beforehand and at least get an impression around, ⁓ there a chance that they will buy it? Is this problem real or is it? Did I make it up in my mind? So how much did it really change?

Martin Tobias: Well, what changed was the time to validation, right? ⁓ Some software is complicated to write, might take you a long time. But today, if in... a very short amount of time, can't MVP a demo, a working demo of the major features within a month or so, you're a fucking idiot. And two or three years ago, it might have taken you six months to build that demo, hand coding it. But today you can build working prototypes very quickly, and you should, and show those. ⁓ I think it's just compressed the time. ⁓ to build anything.

Niklas: after you, or maybe you still continue angel investing, but you then you launched your own venture fund. And I think one of the biggest challenges is getting your deal flow. So seeing the deals that are relevant for your specific thesis in your specific field. How did you cope with it? And I think you also built some software recently to help with that.

Martin Tobias: Yeah. So, I mean, I have a pretty big social media presence on Twitter and stuff, even though Twitter just locked my account yesterday and I don't know why. So I might not have as much deal flow from there anymore. But you know, I get about 300 deals a month, which is 10 times more than I was getting as an angel. As an angel, I was getting 20 or 30 deals and now I'm getting 300. It ⁓ turns out when you're a professional VC and you hang out a sign saying free money, everybody wants it. So it's pretty easy to get deal flow ⁓ as ⁓ a VC. ⁓ The harder thing is to get qualified deal flow that fits your thesis for any VC. the challenge in the market is that founders and VCs are both in a low probability matching game. VCs say no 99 % of the time, founders here no 99 % of the time. So I built some software called InvestorMatch.Pro where a founder can upload a deck and it has 10,000 VCs in it and it says here's the 20 best fit VCs for your stage and check size and geography and everything. And so we're trying to shortcut. So rather than cold emailing 200 people, you're emailing 20 highly qualified people and you most founders in this get about a 60 % open rate. So I've written a lot of software to help me find the right VCs or find the right founders and also help any founder find their right investor because when you find that match, it's magical, but it's, hard to do cause it's needle in a haystack type stuff. But the interesting thing is AI ⁓ can, shorten that path. And I'm very excited to do that.

Niklas: really interesting also because you built this yourself with something that you probably would have hired somebody to do like a year ago, one and a half. Now it's possible with Replet. What's your experience with these AI tools? Also, I think you use both Claude Code, as you said earlier, and Replet. How do they compare for your day-to-day basis?

Martin Tobias: Yeah. Well, I started on Replet because while I was a programmer, I haven't programmed in a long time and it seemed like a good integrated everything hosting development environment and, and Claude code two years ago, you had to be a real developer and understand the terminal and stuff like that. And I didn't want to deal with the terminal. So I started on Replet, but I actually just moved a product off of Replet onto Claude code. Claude code's gotten a lot easier to use even for less technical people. But there is actually, so investormash.pro, I did it all on Replet. I'm leaving it on Replet for now. It's 450,000 lines of TypeScript and Node.js code. There's actually a website you can go and connect your GitHub repo and it'll tell you if you had developed this with outsourced developers in India, how much would it cost? And it said it would cost $500,000 to build that software. And I did it for less than $10,000 in Replet credits. So it's still relatively expensive to develop, phenomenally inexpensive relative to just three years ago. So I love it. ⁓ The reason I'm switching at least one of my projects from Replet to Claude code is even at $10,000 there, ⁓ I did another analysis. What if I developed it in Claude code? It would have been like a thousand dollars on Claude code with a max. So, you know, I have the money. I appreciate it, but I'm going to try cloud code for a while and see how it goes.

Niklas: Yes, so for me, I also have a max plan on Cloud Code. You can also cross check what just RP usage would have cost ⁓ with the same usage. it's always you get to, and I think that's fair, it's in the interest of Anthrophic. They need to show some value on the subscription. The subscription model works differently. But it's always, it's a few thousand dollars in. RP credits that you can easily consume on the Cloud Code tooling. What I found was, ⁓ I think with all these tools, I've never looked at Replet, but what's really important is that you have the database schema in some way and the whole source code in one repository. for example, you would use some ORM or something that represents a database structure. And that's really important that all of this is in one one repository for Cloud Code or all these tools to work really well. Once this gets disconnected, it gets really, really difficult. That's something I found, for example.

Martin Tobias: Well, that's true. Another challenge I found with this vibe coding thing is you start with an idea and you start to build and then you add features and you add features and you add features. And so you end up with a bunch of spaghetti and your data model turns to shit. So, on my next project that I'm building now, I'm, I've spent three weeks with Claude, just building a PRD and getting the data model set up front because when you iteratively design these things and add features and don't go back and sort of re-architect the data model, you can end up with something that's really hard to deal with. For anybody that's building anything more complicated than a simple dashboard, ⁓ I would tell them to spend more time doing ⁓ a PRD before you pick up any of these ⁓ things because you'll save yourself in the end.

Niklas: Yeah. Though I have to say, so I experienced something like that, especially earlier when I worked with Cursor. These days, Cloud Code for me at least has gotten very good at that. Once you are careful with the early designs of the data model and how tables are structured, then adding data and creating new tables or other things like that. For example, in PostgreSQL database, it has gotten quite good at.

Martin Tobias: to get it.

Niklas: that and also at structuring ⁓ these days. In earlier days, I, for example, the problem that ⁓ components would get too large in the software. That's really gone. So I'm actually quite happy with the progress.

Martin Tobias: Yeah, yeah, yeah. I had that with Bolt.new. My project got too big in like a week. But I think they've solved that problem, but they haven't solved the iterative design problem and poor data model over time. ⁓ And so I've gone back and asked like, look at this data model and tell me how it should be re-architected after you build 20 different features. ⁓ going back and doing an architectural redesign after you built a bunch of features independently. And these things, haven't tried cloud code yet, but Replet, you know, will take the lazy way out all the time. It did stuff like hard code emails into code rather than making them editable by the admin. ⁓ You know, it just did stupid shit. ⁓ And so you, you do have to know a little bit about architecture ⁓ to make these tools. build something that's good.

Niklas: Yeah, that's also from other guests that I had on the podcast. That's a basic understanding that exists. The value of the really good engineers actually increases because they know what they want to have built and they can, as long as they supervise and take a good look, they actually get just a lot more productive. And that is the main thing. So if you just trust these tools to build things for you, you will not end up with well, these

Martin Tobias: I'm Zach. Yeah. Especially if you weren't a senior or didn't know much about development before you started them, because they will take the lazy shortcut way if you let them. I've seen that in my portfolio. I'm a little concerned about, you know, what's going to happen. You know, how do people learn things like programming? mean, in my first programming job, when I got on my first job, they put me on bug fixes. And that's how you learned the code was you were fixing bugs in some code base you never knew before. ⁓ And a lot of those entry level jobs are now going away because they're super easily done by some of these things. And I had one of my companies had a five person development team had two senior developers and three junior developers. They were all using clogged code, but the reality was the senior developers were producing 10 times the code as a junior developers, even with clogged code. And what they decided to do was to fire the three junior developers and hire one more senior developer. ⁓ and they got a lot more done with senior people. So you're right. These tools can help senior people become absolute superstars, but it doesn't turn junior people into superstars. And it's a little bit of, of an interesting question of, know, are we going to run out of senior people or how do people get to be senior people? If they just start, I don't know how that goes. But that's exactly the same thing for all of these automations. Senior lawyers are replacing analysts with Harvey now. And then how do you become a lawyer if you can't be an analyst anymore? I don't know.

Niklas: Yeah. After my PhD, went into consulting and I started like building PowerPoints and doing these kinds of stuff.

Martin Tobias: Exactly, they start you at bottom to learn the whole thing.

Niklas: Yeah. These days, if you have cloud cowork, it's actually really good at the stuff I did back in the day. So these days, if you have an engagement manager plus a good set of AI tools, you can work with a really slim team and produce results fast.

Martin Tobias: Exactly. You can, which is I think the bigger social question, which is how do we get that next level of experts? ⁓ I don't know how it's going to shake out. It's going to be some interesting disruption, but ⁓ at least for now, ⁓ you know, there's no turning it back. People are not going to have their analysts. I did the same thing when I was at Accenture. I was doing PowerPoint slide decks for my engagement manager and fixing bugs. Those were the two things I was doing.

Niklas: Yeah. And that's how you start and that's how you learn a lot, right? And you have to go through it to get to the other side of the story where you just get the deck at the end of the day and present it to the client and do these kinds of things. But yeah.

Martin Tobias: And maybe they'll still have those people doing it and that person will just use an AI tool to do the deck instead of themselves. So they'll have more time to do more, but they'll be fewer of those entry level people because you just won't need as many of them.

Niklas: I'm also, I have a large question mark on this topic. So I generally believe that like, if you look at the basic things, they remain true. It's we humans, we exchange money, so we don't have to exchange goods and we try to create value. And so we will find work. this I believe in where the same job still exists that existed 10 years ago, I doubt it. So it will be very different ⁓ in a lot of places so that I'm quite sure of.

Martin Tobias: Yeah, I don't know what that's going to happen. And I sometimes get crapped on at conferences or so on because like as a VC, people say, oh, know, AA is eating our job. I'm like, good. I want to fund that company that's eating a job because the reality is SaaS today is a $300 billion business, but labor, the expense for companies to pay people to use SaaS is a $4.5 trillion expense. So the real opportunity is to eat some of that labor expense and turn it into software expense. If you can turn a trillion dollars of labor expense into a hundred or 200 billion of software expense, that's a good trade. Every company will make every fucking day. And ⁓ I think the software market gets bigger and I think the labor market gets a little bit smaller for those particular jobs. And then those people will have to find other things that humans can uniquely do. And I'm very excited about software replacing. It's like, do you really want to go to the bank to deposit a check anymore and talk to a person? No, because you can do it on your phone. Like this software has made, has freed up our time that we would have sat in the car and stood in line to cash a check. Now, what do you do with that time? ⁓ It's freeing up human capacity to do more human things. And I think that's a good thing.

Niklas: Yeah, and I genuinely believe that we overvalue the status quo and undervalue the power of innovation. So I recently told somebody that a few thousand years ago, we lived in caves, right? Today we can go on an airplane and fly around the world and sit in different places and talk to each other. So there's a huge gap in between what we have achieved as a society. And when we look at the overall picture, we live like kings lived a hundred years ago, like, or better.

Martin Tobias: that is better.

Niklas: Yeah. So I think we oftentimes underestimate this value of productivity gains and what the overall status is, which brings us back to the thesis of your fund. I think you look especially for companies that reduce friction at scale. What does that mean?

Martin Tobias: Yeah. So there are lots of people who build software and they're like the software can do this. I'm building it. I'm like, okay, what problem does it solve? What friction in the business does it solve? And the con article example is a docu-sign, right? The old process was you send somebody a PDF and email, they print it out, they sign it, they scan it, they fax it. This was a friction filled process to sign a contract. There was a lot of friction. in the process. Software allowed you to sign that on your phone, reduced all the friction from that particular part of the process. And then it turned out that there millions of signatures a day, got to be a very big market. So that's what I'm looking for is does the software that you're building significantly reduce ⁓ friction? And, but you see friction everywhere. ⁓ You know, in medical records like Did the billing code get set correctly? Did the insurance company reject the thing because of the wrong billing code? Right now, a lot of manual people reviewing that computers can do that review better. mean, so, but it has to, you know, identify a friction point and then basically re-architect that in software. And the software has to deliver the value. You know, for example, I funded this company called Vega Cloud. The guy ran a managed service provider for 10 years. and managed service providers, managed cloud environments for enterprises on behalf of the enterprises. They solved all the problem with people. So they had a whole bunch of systems engineers. They were doing the configuration. They manually had a bunch of scripts they'd written and they had, but it was basically a people job. And he had a hard time scaling that business because it was hard to find systems engineers. There weren't enough of them and they were hard to train and it's hard to scale and blah, blah, blah. He's like, I'm going to solve the same problem, but I'm going to do it in software. I'm going to do a have AI that can reconfigure machines can turn things on and off. And I'm just going to rewrite software rather than a bunch of custom, you know, Pearl scripts. I'm going to write AI software that does the job of systems engineers, but does it at scale. And that company got very successful because their software was fundamentally, you know, read replacing a systems engineer for 80 % of what the main service providers did. ⁓ So the software has to fundamentally reduce friction in some business process that either drops savings to the bottom line or increases revenue. ⁓ One of the perfect ⁓ examples of that is all of these AISDRs that are doing research on LinkedIn and finding people. Salespeople used to do that. They used to open up their own LinkedIn and search for their connections and use Sales Navigator. to find people that were VPs of supply chain at whatever, and then manually do that. Well, that's all friction, all that manual work. Computers can do that at scale much better.

Niklas: I think I, so lately I've seen very strong growth numbers from a number of companies. I always wonder how they get there. For some it's explainable. If I look at lovable, it's the viral loop is quite clear. So I understand how it works. Like you share the website, somebody else sees it, ask you how it was built. And then they also join lovable. That's, it's clear for others. That's not as clear to me. So do you believe that's largely these automations that lead to it? Or is there something else that has changed and go to market in the last year?

Martin Tobias: in go to market? Well, ⁓ No, I mean, I actually think these AISDRs are a net negative to the go-to-market. ⁓ Email was already a touchy medium to try to reach people, but now when you can have 1,000 emails going out a day, ⁓ email, these AISDRs have turned email into an even greater spam engine. ⁓ So I don't encourage people to use things like AI. SDRs in B2B. You mentioned lovable. That's more of a consumer product and I don't invest in consumer things. So consumer products need some sort of viral shareable loop and that's how they grow. B2B needs a quick time to value and a referral system. ⁓ Most of your customers in B2B are going to come from happy customers who tell their friends. ⁓ more than cold outbound or anything like that. So it's trade shows, it's webinars, it's ⁓ things like that and go to market for B2B.

Niklas: So you still think it's similar, it's people to people and exactly these, I would call it network effects.

Martin Tobias: Yeah, maybe a little bit of social media, a little bit of LinkedIn case studies, ⁓ things like that. Yeah.

Niklas: When you look at companies, I think you have said that the first five hires can make or break a startup. ⁓ What do you look for when you look at the first hire? what did you look for in your companies?

Martin Tobias: Well, one of the things, I had an investor ask me yesterday, do you invest in solo founders? And I say, generally no. And he says, why, do you hate solo founders? I'm like, no, I don't hate them. But one of the primary jobs of a founder is to attract other smart people to their quest to join the bandwagon. That includes investors, includes customers, and that includes employees. So what I look for is a founder who can get other really smart people to join their journey. And if you're solo and you've been doing it for two years and you've gotten nobody to join you, that's a pretty negative signal in my mind. You know, maybe you could have a one person billion dollar company, but generally you have to attract other people. And so you look at... the first five hires and you say, where did they come from? Did they work with the guy before? Did they come from Google? Did they come from Microsoft? Is he attracting A players or B players? And you look at all of that. The other thing I look for in a founder that's doing recruiting is do they have a really, do they have a try before you buy policy and are they quick to ⁓ change their mind if the person doesn't work out? I actually like CEOs. I had a CEO hired a CTO. fired him within two months. ⁓ I'd rather he fired him within two months than a year. Right? I want them to make quick decisions. So when I see that, that's actually a positive signal for me.

Niklas: And I think I've read that a lot. that these decisions, when you feel it's not working out, especially very early stage, you just cannot afford to keep it going. You have to make these decisions extremely fast and be very consequent. And because the company just is so small and it's so hard to keep it alive anyway.

Martin Tobias: Everybody has to do multiple things and and probably the biggest hiring mistake that I've seen startup founders do at that first five is Getting somebody from the big company for whom this is their first startup. I made that mistake when I was at Laudai I hired this biz dev guy from Microsoft great resume He's like I want to do the startup thing blah blah blah. I worked for 15 years at Microsoft first day at my company He says where's my assistant? And I'm like, what the fuck do you mean? You don't have an assistant. Get the fuck a work. And the culture clash was just too much. I do not recommend that people hire at least the first couple of employees for their startup directly out of a big company. I'd rather somebody go through a startup, realize that they like or don't like the startup, and then be their second time hire with some experience from a big company.

Niklas: So I also agree. large companies have established processes usually, and I've seen them, I worked in them. So you work usually on very small parts. You never see the whole picture. And even the CEO, they have a very broad, but shallow view. They don't see the details. So that's very different from something early stage where you have to deal with everything. Like if you're a technical or CTO, then you also code at the beginning. That's part of your job.

Martin Tobias: Yeah, exactly. And, and people that were really good at the narrow thing at the big company, they have to really understand that they're going to be thrown 10 other things that might not be in their wheelhouse. I saw that very well. ⁓ when I was doing that bug fixing on my first job, they hired this guy, Stanford PhD in computer science and put him on bug fixing. He quit within a month. Why? Because he's like, I didn't write this code. This code is shit. ⁓ It doesn't meet my standards. I can't work with it. I said, okay, you should go back to Stanford and be a computer science teacher because this is the real world.

Niklas: I think also in a startup, it's a lot about getting things done, right? So that's at the end of the day. I've launched a number of products over my years now. this is always the idea is such a small part. It's also to be understood. then you somehow have to get the product to market. And then you somehow have to find customers and then you have to make them happy and find more and kind of scale this. this is like, and it changes, right? Then you have to hire people on all these different stages. I think it's just, yeah. It's just very different.

Martin Tobias: It is very different and I want people to understand what they're getting into.

Niklas: When startups are raising and they are coming to you, how do you like to be pitched? What are you looking for?

Martin Tobias: ⁓ well, I have a form on my Twitter feed there, ⁓ or on my website. And I like people to use that mostly because it puts, you know, I don't like cold emails because it's not structured and it doesn't go through my AI things that, you know, tell me whether I should look at it or not. So I appreciate it when people use the structured form, even though it's a little harder to do. but what I look for. It goes back to the distribution wedge. Tell me about you, where you came from, why you're solving this problem instead of some other problem you could be solving. Why is this personal to you? ⁓ I want people, I want to understand, ⁓ are they in it for the long term? Are they just sort of, did they read a Gartner report and thought they should build this product? Those CEOs I don't fund. ⁓ So I want to get into their background, their motivation. ⁓ and then their distribution wedge. ⁓ You know, don't show me a giant Tam slide. Show me that you did 200 customer discovery calls and you've got 20 people on a wait list. I'd rather see that than a Tam slide.

Niklas: That's really interesting. Looking at the future, I think you have said that ZAS was bigger than cloud infrastructure. How is it with AI applications and AI infrastructure? Do you believe the same? Is it different?

Martin Tobias: I think it ends up the same, but you got to understand where we are in the game. We're basically three years into this AI ⁓ thing. And I don't know what the numbers are for 26, but in 25, if you looked at all the dollars spent on AI, Nvidia got something like 80 cents of every dollar because people were buying the shit out of Nvidia hardware. That's the same as in the beginning of the cloud. was Cisco and Dell making all the money. but you roll forward five years into the cloud and SaaS started to get bigger because people were building applications on top of that infrastructure. And at the end of the day, the application layer got more valuable than the infrastructure layer just because that's how it has to be, right? You have to make more money on top of the infrastructure than you pay for the infrastructure or it doesn't make any sense. And we're at the very beginning. ⁓ We have... I have some AI application companies that are making a lot of money, but nobody's making billions and billions of dollars. The people making billions of dollars are Nvidia and OpenAI and Anthropic, but those are infrastructure still, right? It's not clear who is going to be the winners in the application space, but my bet is that the application layer on top of the infrastructure is three to five times bigger than the infrastructure layer. And if you look at the 500 billion being spent a year on infrastructure, That means that the application layer has to be $2 trillion to pay for all of that stuff. And it's unclear what all those applications are going to be, but those are the bets that I'm making now is what applications are going to be big five years from now.

Niklas: I found it really interesting. ⁓ read, I think yesterday, that Elon has made a large deal with Anthropic to lease the first center he built, Colossus I, to Anthropic for, 1.5 billion a year, roughly, a month, I think a month.

Martin Tobias: Well, I heard it was that much a month.

Niklas: Yeah, months. Yeah, I was wrong on the numbers. I think it's 45 or something over the course of three years.

Martin Tobias: It's a giant deal. And right now infrastructure is in part the scarce commodity. I mean, he built Colossus in what, like eight months and it would have taken anybody else five years to build that much compute. ⁓ The fact that he can deliver infrastructure way faster, he's going to make a lot of fucking money selling infrastructure. Absolutely.

Niklas: And then on the model side, I think this is also a question that I personally find very interesting. So I believe that Apple still bets on Edge AI largely. So they kind of still believe that we will see on-device models that will be good enough for most things. That's why they don't touch the cloud game. ⁓ They are very good open-weight models, especially in China. What do you think? How will this play out? Will these model providers kind of stay alive or is it really the application layer and the infrastructure layer below that captures who are you?

Martin Tobias: The model providers will still stay alive because you have to use the models to get your application done. So the ⁓ API strategy on all these companies is going to be where they make most of their money. Even though I think chat TPT also have a decent consumer business, people asking the chat bot things and stuff like that to your smartest friend, you know. ⁓ But ⁓ I was just talking about this with some investors yesterday for the first three years of AI, you had giant dropping in token cost like a thousand X dropping in token cost But if you look in the last six months token costs are relatively flat for the for the frontier models And in fact, if OpenAI and Anthropic go public, I think you're going to see them raise prices. I've already heard that, you know, Claude's getting rid of the max plan at the end of June or something like that because they're going to want to just charge you for every token you use. So they're going to have to fix their revenue model. They can't afford to subsidize people on tokens anymore. And so I think the frontier models... prices might actually go up. And what that means, to your point, there are some good open source models and stuff like that. If you're an application developer, what you're going to have to do is write a router and figure out, you know, which calls go to the frontier model and which calls you can do simply in the cheaper models. And you're going to be routing either to internal models or, you know, cheaper Chinese models. And the way you're going to save money is by ⁓ routing your calls ⁓ correctly. And again, that's something that applications have to do. I mean, if your application just runs on Claude or OpenA, and that's your only thing, you're going to have a cost problem.

Niklas: Yeah, this is really interesting for the models themselves. Yeah, I don't know where this plays. think if it really goes away where Claude charges on a token basis, we'll see another set of competitors in the market. That would be my orntropic. That is what I would assume.

Martin Tobias: You will and you already do. mean you have Kimmy and you have ⁓ DeepSeek and all these other things. The problem is they're not smart enough for certain tasks and the frontier models will always be frontier models and they'll always be the better models and people will pay a premium but what will happen I think is that people will just realize that they don't need to pay the premium for everything.

Niklas: That's interesting. A few final questions. You're an avid poker player. ⁓ One poker player lesson that translates to investing.

Martin Tobias: ⁓ It's pot odds ⁓ and ⁓ making the right bet based on the odds of your return. So for example, if I'm in a hand and I have a 30 % chance of winning ⁓ five times my money. I should make that bet even though I'm going to lose it 75 % of the time. ⁓ And that's exactly the same in VC. A VC is always taking the long end of the bet, right? ⁓ You've read the numbers, something like nine out of every 10 VC investments fail. What that means is that that one has to return 10 times more than 10 times its money to make up for all the ones that don't work. So understanding pot odds and the odds of return, that's why You I turned down a lot of companies because I'm like, you're just not venture scale. You're not big enough. Maybe you get to 50 or a hundred million dollars, but that's not enough to return my fund, even if you're successful. So I can't invest in you. You have to be investing in the long shot that can make a lot of money in that. And that's pot odds. So the, what poker teaches you is that you should still make bets that are low probability. ⁓ if the odds of your return are good enough. Here's another example. I was at a poker game. ⁓ During the first election when Trump was running against Clinton, Trump was at 40%. Clinton was at 60%. Somebody at the table said there's a 0 % chance that Trump wins. And I'm like, zero? I think he's at 40%. And I go, what odds will you give me? He said 10 to one. I go booked a thousand dollars. If someone's giving you 10 to one odds on a 40 % chance of something happening, you should take that bet every fucking day and twice on Sunday. Because if you do it 10 times, you're going to win, even though the odds are against you when you make the bet.

Niklas: I think that's, I read it in the Sam Frank, Bankman Friedberg as well, like flipping weighted coins. And I think a lot of decisions are just around that, right? So what is the upside? What's the, ⁓ what's the downside? What's the maximum to lose? I think that's a really interesting theory.

Martin Tobias: Yeah, the nice thing about venture is that you can't lose any more than you invest, but you could gain a huge multiple of what you invest. That's the asymmetry, right? Your downside is limited to your check. Your upside is.

Niklas: ⁓ Do you generally believe that we overestimate risk or that there's something about risk we are not really calibrated about?

Martin Tobias: Well, most people overestimate ⁓ risk and don't take those. It takes a certain person to want to do that game in venture. You have to be comfortable losing 90 % of the time. And that's a hard thing. My wife can't stand it. She's like, you're going to lose 90 % of your bets. I'm like, yeah. She's like, well, how do you make money? I'm like, well, that one has to make a lot of money. She's like, I would die if I made 90 bad bets for every hundred. It takes a certain person.

Niklas: Thank you, Martin. That's very nice final words. And you, my listener, I'll you next time.

Martin Tobias: All right, thank you.