The SaaS playbook is dead. What replaces it? | Andreas Goeldi, b2venture

I hear from a lot of CTOs, a half of my developers are fully on AIB's coding. They don't write the signal line of code anymore. It's just all agents all day long. And the other half refuses to use this new fancled stuff, right? So guess who's probably going to win? This is Melanie. And this is Christian. You're listening to Follow the Gradient, the weekly podcast with actionable insights about how to build a business while staying sane. We're living in the midst of an AI revolution. We've moved past the chat GPT moment of a few years ago, and now we are in the Claude code moment and a realization trick it through. You cannot simply sprinkle AI pixie dust over the old SAS Playbook and then expect 80% margins to last forever. As our guest today argues, we are witnessing a fundamental renewal of how the internet works. We are moving from the era of selling the tools and dashboards to the final automated outcomes that customers actually care about. In this new landscape, the real winners are those who can provide a full service with AI driven efficiency. Over the next 45 minutes, we sit down with Andreas Goldie, a partner at B2Venscher, who has seen the internet from every angle, from coding industrial systems in the '80s to building video advertising giants in Boston. Andreas has a rare time traveler perspective on tech cycles. We dive deep into the concept of service as a software, into the era of wipe coding. And we talk about new defensive modes in this new world and helpful mental models like AI as the new Excel. Honestly, this episode is a must listen for everyone remotely navigating the taxi. And before we jump in, a big thank you to every one of you guys helping us bring new inspiring guests to the show. The best way to continue to do that is to subscribe and follow to the show on your favorite platform. For more of our tangible advice and exclusive data charts of our intelligence unit, visit followthegradient.io. And now, enjoy the conversation. [MUSIC PLAYING] Andreas, you've seen the internet from every angle, so to say, you were coding industrial systems in Fortran and Kabul in the '80s. And you built website when the web was just a few pages in German speaking Europe. You often say that the illusional thinking kills more startup than bad ones. Looking at the current AI landscape, what is the biggest illusion founders in BC's client to right now? First of all, I think you have to be a little bit delusional as a founder, obviously, right? You have to be almost obsessively optimistic. But I don't think that's the same thing as believing in something that is quite obviously wrong, because that can lead you down the wrong path. And the delusion that I currently see most frequently in the market is that people believe that the good old software, the service playbook still applies to the world of AI. And you just bring a little bit of AI pixie dust over it. And suddenly you have these crazy growth rates, and all the incredible flexibility. And I don't think that really goes together. I think we are definitely seeing that the SaaS playbook is getting faced out increasingly. And the newest coming, we all don't really know what that new thing is going to look like. It's a process of discovery. And actually, as you already mentioned, I'm quite old, right? I've seen a lot of the IT industry. And I've been around who knows, seen the early days of software service in the 2000s. And people don't remember that, of course. But it was also an entirely new idea that you could charge for a software product on a monthly basis. And it just could buy software. If you're a credit card and immediately start using it, you're used to be much more complicated. And with AI, I think we're seeing a similar renewal now of how all of the stuff works. But again, thinking that the old rule still apply, and you can still have 80% growth margins, but also magic AI is not going to work. Now AI, we've moved past initial shock, so to say, or excitement, whatever you want to call it, of ChetGbT in 2022 when it was first released. Now founders are trying to figure out where the durable value actually lives. As you also were hinting to with like playbooks that know a longer work, you've interestingly categorized this into this four distinct pockets. So one is enhanced as us, democratic creation, service as software and autonomous agents. Can you give an overview on how you arrived at this four buckets and why you think also in this mental model? Why does it make sense? Yes, I did what everybody who has ever seen a business school from the inside does. You do a two by two matrix, right? Always works. It's a simple concept, but it works. And I basically looked at which businesses are really more product oriented versus service oriented on one hand and which ones are really AI first, critically versus more traditional software plus a little bit of AI. The results are basically these four categories. And I've also, of course, looked at what we are really seeing in the market and there are very clearly these categories in many ways. I think the most frequent one is, of course, traditional sales with a little bit of AI stuff on top of it, right? That's not terribly original, but it works in many cases. And then you have these other categories that are now emerging and quite new in comparison. And if we now dive a little bit deeper into service as a software, you've mentioned we are moving from selling tools to selling all outcomes. And if I'm a founder, like, why should I be building a service replacement instead of just the better dashboard to say? The answer is because your customers actually want to buy that, right? They want to buy results and not the software. If you want to be a bit cynical, you could say, software is just a crutch bad solution for somebody who actually wants to get an outcome. You don't really want to buy a tool. You don't want to buy the hammer. You want to buy the nail that is in the wall that you can hang in your picture on, right? And I think the software industry has a bit forgotten about that because we're already worked for a long time and everybody just want to have the super scalable solution. And I think with AI now, we are going back to a world where it is more possible to provide a solution and outcome a full service, but at efficiencies that are reminiscent of what you can do with a software business. And interestingly, I experienced this in my own career. My last startup, he based in Boston, which was in the field of video advertising. We managed very large-scale advertising campaigns for large brands. And we used AI very early form of it quite heavily. And we were able to more than double our revenue in one year without hiring another person. It's simply because it was so highly leveraged behind the scenes. And our customers actually bought that as a service. We tried to sell the software, but they were like, no, we don't want to deal with this complicated stuff. To sell us the result, please. And that actually how it worked out. And we are now starting to see this in many categories where a customer's again, want to just buy an outcome in some degree of handholding on top of that. And thanks to AI, it suddenly becomes possible to provide that. And when we now then also think about the pricing part, because like we read your essay, the fourth wave, where you mentioned that the pricing will move beyond $20 to see it, how do founders in the AIH actually price AI value are outcome effectively? Yeah, pricing is always one of the hardest things like an entrepreneurship. We've been very complicated, very iterative. And I think it's becoming even more complicated because you now suddenly can sell these outcomes on one hand, which are hard to price because you have to understand, what is it really that the customer wants to buy? How do they value it? How do we see the ROI on this? And that's a whole separate discussion. You also have these traditional software price points 20 bucks per month that traditional says, right? And people just use it as a default kind of. Now you have more expensive packages. You can also buy a 200 a month package from OpenAI, et cetera. You can buy it on top token plan. So I think we are entering now a time of where suddenly pricing is very creative again. And that's a great opportunity for founders to really figure out how can I differentiate my product on a pricing basis and not just on a feature or service basis. And we see quite a few creative startups doing that, really doing very unusual stuff. And I think that's very encouraging. But it can also be confusing, right? Because you have some of these so many degrees of freedom that you can deal with. And coming back also to the point that you said before, and also these four buckets, because you've said the democratic creation allows experts to build niche sales without huge development teams. And you were making the example from flexibility in Boston. And we see that more and more like small teams who are really making a huge boss. Does this mean the era of general, generalist Suss is over in favor of like 10,000 hyper vertical micro startups? Or on the flip side, maybe ask differently, does this rather mean that Suss is that? And companies will just build the stuff which Suss used to do for themselves. How do you see that? I don't think that you are going to vibe code the equivalent of SAP anytime soon. First of all, it's packages are, of course, hugely complex, not just in a technical sense, but also in the sense of the underlying business logic. And I don't think that's realistic. And it doesn't have to be, right? I think what we are going to see is a combination of different approaches where you still have established systems of record that are built and sold fairly traditionally and have to be the solid underpinning of everything. And in addition, you can much more easily now build companion products that solve a very specific problem. To give you an example from our own firm, B2 Venture, we have a portfolio management system that is traditional SaaS product. We also have a CRM view Salesforce. And we have always struggled to create, to find a portfolio management and not management and a portfolio modeling tool that allows us to project what returns might be and where to reinvest in which startups and so on, the very specific use case. And there was simply no product on the market that fit our needs with all these integrations have been needed. Now recently, Karim, one of the very talented people in our ops department just vibecoated the solution with replets in two weeks or less. And very specific to our needs, it taps into all these established systems, but it's a solution on the complexity of a normal SaaS product, right? And that's now possible. And you need deep understanding of the specific problem to do that. But now it's possible without writing a single line of code manually. So I think the next phase is going to be dominated by this mix. And in many niches, we're probably going to see a much increased competitive intensity because it's becoming much easier to add features. One interesting factor that we also see in currently is that established SaaS vendors, now thanks to AI, can much more easily create more functionality. They can go much broader and suddenly start competing with other people that were in adjacent sectors. And these other people in the adjacent sectors do the exact same in the other direction, right? So suddenly you have these much broader products that compete in all kinds of sectors against each other. And I think that's going to have very interesting dynamics in the next couple of years or so. Because suddenly you can buy a product that has a much broader scope than it was possible before. And I think that's going to change the dynamics of competition itself. Also, what becomes easier for every B2B SaaS business is building integrations with their client systems, right? Because when you think about the gold market, that used to be a huge pain and a huge kind of slowing everything down and that now also becomes easier. Absolutely. And in surprising ways, especially with the latest generation of coding agents. So I silly example, but I have these smart light bulbs at home from Philips. And it's a pain to control this with an app. I wanted to control it from my computer. And that's a total cloud code. Figure out how to do that. And it came back after two minutes and told me, press the blue button on the connector box. It was like, well, OK, sure. Then it came back and it said, OK, I can control your lights now. So it literally hacked my lighting system. It read the API documentation scanned my network for the connector box. And then told me the human which button I have to press to solve the problem. And that's a small example. But we are seeing similar things in the enterprise. These agents are now able to understand full API documentation's specific code for a specific problem. You don't even need MCP or any of these standards. They tend to be fairly generic. I think this is going to speed up massively how we can integrate data and pull it from A to B between different systems. It opens up a lot of new use cases. And if the company-- I mean, if this is not for you, then easier for a bigger startup or company, but really for anyone. So the question arises about the defensibility, about the modes. You said recently that the defensibility comes from data quality and integration into regulated industries. But if software is becoming easier to copy, is context the only remaining mode? Or what is it? Good question. I think a lot of the traditional modes are obviously going away at the time that you need to implement new features and such. There is certainly an argument for proprietary data that is still clearly a mode. But I think that's where or then most people probably think, especially proprietary data that is also useful for a large group of customers. I think dealing with complicated regulated environments is definitely still differentiator, because that's painful and slow, not because of technological reasons, because of the whole administrative stuff around it. So understanding that deeply can be very compelling. And then I think still having a complete user experience is very compelling without even going into the technology that is underlying. And once we have a portfolio company called Vestegas, they make software to digitize large construction sites. And the total number of times they mention AI on their website, the customer face website is zero, because their customers don't care. Highly modern AI stack in the background, all the fancy agent stuff, et cetera, that you can think of. But the customers don't care. They care about this stuff working for them. And you will probably scare them away if you were told them that the agent is now running your construction site. So I think that's an example for how traditional product management craft, really shaping a solution with the underlying technology, of course, that is now possible, can still be a really defensible mode. And like this example actually connects with the other question I have. And it is linked to this whole theme of theater. Because on LinkedIn, you have seen that. We see that on a daily basis, this demo theater that takes over LinkedIn, very impressive videos that might very much like substance. But it's impressive, anything. Oh my god, I absolutely need to implement it. But then you try it and it doesn't really work. And because you realize it's probably more complicated or it's really just scrapping at the surface. So how does a founder prove they have real product market fit when an LLM can make your prototype look like magic? It's still at the end of the day in the hands of the customer, right? You have to get-- yes, you probably have to have a fancy marketing video nowadays to actually find customers and get them to use your product. But at the end of the day, users will still decide if this thing actually works or not. I don't think you can win paying customers in the long run with just a very superficial demo prototype. You occasionally see companies that have this kind of superficial product, wins on initially are very quickly and then they flame out there equally quickly simply because the product doesn't really do what it promised. And I think that's increasingly a problem in AI industry because it's becoming so noisy. I think the products that really work might grow a little bit more slowly initially, because it takes a little bit more time to really get used to it and understand what it does. But that's going to be much more sustainable in the long run. So my recommendation would still be build great products with the latest technology that is efficient as you can really talk to customers, understand customers. I think these old rules don't change. The speed increases certainly, but I don't think the basics are that much different in that aspect. And when we now also take maybe a closer look at the rapid capability gains of general autonomous agents, while their progress is most visible in coding, tools like Cloud Code, OpenAI Codex or OpenClaw, they are already super strong in non-coding workflows and are also replacing some of the custom-built agents. You said that these journal agents are becoming the AI era equivalent of Excel. Good enough for a wide range of smaller use cases with that specialized software. But what does that actually mean for founders beyond personal productivity, especially as some sort of packaging agent skills as products? I think for founders, there are really two very bad ideas that you can currently pursue. One is that you go after something that is very obviously on the roadmap of the large AI labs, right? So especially coding, that's currently the most valuable field, obviously, right? Economically, and OpenAI and Frappi Google, et cetera, are all pursuing it. They are much richer than you are as a founder, obviously, so don't do that, right? That's probably going to be very difficult. The second thing that is a bad idea is to produce a product that can presumably in relatively on relatively short term be solved by generalist agents. And I think the border shifts almost every day there. And I'm experiencing this myself. For instance, one and a half years ago, I coded a specific agent that was used for market analysis, something that you do as a VC all the time. You have to understand markets and competitors and so on. And back then, it was necessary to hand code this agent and call LLM, and so on. Now I don't care. I just tell codex, hey, figure it out. And it just goes off and does it. Sometimes it writes code for it. And that's just basically done. And we are seeing similar things now in productivity, the plugins, for instance, from my infropic with Cloud for Excel and PowerPoint are all ready amazing. Then you can do similar things with other agent platforms as well. So there is a whole bunch of startups that wanted to be AI for finance. And unfortunately, that's probably done, right? You just have your friendly Cloud instance that you use anyway in your Excel. And they can do almost all of that stuff. Having said that, there's still, I think, the high-end range of very specific use case where that doesn't apply, that are too specific for the big labs to pursue. And that there's still some room, I think, for startups at the high end of the market where also customers might have a much higher willingness to pay compared to the more general stuff. Do you have examples that come to your mind like very specifically? For instance, in finance, what I mentioned earlier, a complex portfolio management system is so specific and needs so much underlying expertise. That's not something where you just write a prompt Cloud does it in Excel, that's not realistic. So I think in these kinds of use cases that need deep, deep expertise that there's still a lot of room for startups. So actually, then the main knowledge is becoming more important, right? Exactly, yes. And I think, again, there is also a distinction between very deep domain knowledge and fairly generalistic stuff. So we recently saw a very bad day on the stock market and then for a big release, this new plug-ins, but I think including the legal industry and finance industry. And of course, used it and a trivial NDA review, right? It does easily as well as a lawyer, just in two minutes instead of two days. And then for basically three. And I think this kind of very generic knowledge work expertise is rapidly going away because you can now pre-package it as a skill file for an agent and probably will have more sophisticated methods going forward, but it has become very easy to emulate this. But I would not do, let's say, an M&D transaction based on Cloud Code. That's so specific. It needs so much general knowledge and pattern recognition and understanding of deep context that's still for the foreseeable future, I think, the domain of human experts or in some ways, in some context, of course, the support of very highly specialized software. Now, I want to dive deeper a little bit on the situation many founders find themselves in right now and also then the VCs who have them in their portfolios. And if I'm now a founder who has started his or her company 46 years ago, I'm sitting on a traditional SaaS business with anything between 30 and 200 employees. What does my course of action look like right now? What can I do? What can I think about what mental models to apply to really get ready for the future? I think it's not very original what I'm going to answer, but you have to get very aggressive about AI or else it will be left behind. And this applies on several levels. You have to think very hard about how you can use AI to make your product much more useful for your customers. And there was a lot of hope for a while that you can suddenly charge your customers much more money because it's now has a fancy chat bot in it and so on. I think that's not really what's happening because that's almost now just the base case for the people expect. But there are ways how you can use AI to make it a lot more useful, a lot more efficient for your customers. And this increases your lock-in effect that you're going to have but it also might increase willingness to pay. The second aspect is also you need to use this on the product side as a defensive measure. As I mentioned earlier, your competitors are probably using AI to build new features that are going to compete with your stuff and that you have to have an answer to that, right? Make your own features better and add new features that make it more useful for your customers. And then the third area is really internally usage of AI very heavily. And we're seeing a lot of interesting discussions in our portfolio companies I hear from a lot of CTOs. Oh, half of my developers are fully on AI based coding. They don't write a single line of code anymore. And the other half refuses to use this newfangled stuff, right? And then yes, there's still a case for handcrafted code for particularly difficult use cases. But being aggressive about is efficient as you can is very important. And as an example, we have several portfolios that are using AI, of course, the further customer service. I'm hearing regularly that they've already replaced 40% to 70% of user requests by AI. So the agent handles it end to end. And in some cases, they've even granted the AI agent direct access to the underlying code base of the software product. So if somebody complains about the bug, the agent goes into the code base, looks like, oh, yeah, that's right. It's actually a bug. And it files a pull request automatically. I have to fixing it to the developers. So the human developer already looks at the agent already fixed somebody, something that somebody complained about an hour ago. That kind of efficiency is obviously quite dramatic. And I think if you're not trying to embrace that, you will be very quickly left behind. And I think it's one thing as a founder or as a leadership team to adopt this mindset is another thing to really push this mindset and enforce it in the organization. And I think you've also alluded to that in the past. You said that AI adoption might fail in some organizations due to change management. So can you dive deeper on these points and what you have seen so far on the ground? Yes. First of all, on the level of individual contributors, I think work will change very rapidly now. And we again see this initially now in coding, but suddenly the developer, you're not just writing code. And that's not the main part of your business anymore of your daily work. But it's really coordinating agents. It almost feels like a management role. Interestingly, Peter Steinberg, the creator of OpenClaw, said also in a tweet that he thinks he is so effective in a chain-decoating because he used to run the developer team. So he knows how to manage people. Now he manages agents, right? And this is a fundamental mind shift. If you are a developer who has always just written code and somebody product management told you about to implement, and suddenly you have to control all these agents and give them guide in the very different mode of working. So I think educating people about that, giving the time to learn this, train the minute, is going to be very important. Then a second level is middle management, right? So in any sizable company, you have people who's main job is coordination, right? They don't necessarily take very important decisions, most of the time, but they really shuffle information around and keep the trains running on time. And there's an argument now that this might increasingly be taken over by AI as well, right? Just simple information, logistics, and such. So suddenly you probably are going to get rid of middle management layers and that's going to change very much how companies operate. And then I think on the top management level, you're going to see very similar things. You have to really decide how aggressive you want to be about AI, how quickly you want to get this into your company, how you're going to deal with people who refuse that or don't live up to that. And we have already heard from including larger companies where they basically said very openly, hey, if you're not using AI, your career here is done, right? Please leave the company. So I think this is going to get quite radical and not always pleasant, to be honest. I would, on that point, with the how an organization actually looks like and works, because I agree with you, if you look at an organization, so many frameworks we have or processes or ways of working are made or have been invented because you need to coordinate humans. And you need to coordinate a typical sender recipient problem that I have something in my head and I need to coordinate with other humans with AI that's obviously gone because everything is centered around the agent. So how would a let's apply to a text that how would this look like? Do I only have a couple of really senior people? With a lot of, for instance, domain knowledge or knowledge on scaling in a particular sector or country and then using these agents to supercharge them or how do you think this is going to devise? It will take time to go through this transition. I think the vision of having the one person, $1 billion company is still quite far away, presumably we'll see. And not easily surprised anymore by AI, it goes much quicker than we probably expected. But that's still going to be quite complicated. I think what we're probably going to see is, again, the thinning out of the middle layers of companies, so there's going to be more direct impact that you can have as a founder because you can control these agents directly, but you can also coordinate your human employees more directly through these means. And then on the other end, you're going to have more individual contributors who suddenly have much, much higher output. And that means, of course, that you suddenly have very different frameworks and processes for how you organize a company. So there's going to be a need for a lot of new inventions in how you structure an organization, how you build the necessary tools around that. I think this is still wide open. And by the way, an exciting opportunity, of course, for startups, right? Figuring out how you do that and maybe building a product around a service, whatever is definitely one of the most exciting things that is currently going on. I think the way the economy works will look quite different in a few years from now, especially on a company level. On that last point, what is your prediction in particular for engineers? So will we have mostly only senior engineers working in tech companies? Or will there also be junior engineers? And if only senior, how do you train new seniors? Because you do need juniors at some point. That is a big question currently in many professions, not just engineering. I'm hearing the same from law firms, for instance, right? Where you can say, all the new, the not super demanding stuff that junior lawyers do normally, right? The reading long contracts. And so on can be done by the AI. So why do we need so many junior lawyers? And the question is, of course, OK, who trains the next generation of senior lawyers, same, of course, exactly with engineering? I don't think we have a good answer to that right now as an industry, but also as a society. I'm only hearing, for instance, from students that I work with quite a bit here in the context of Start Global at University of St. Gallen, that has become already noticeably harder to find a job after you graduate. So once it is that I recently read is that the number of job openings for graduates that are handled by the career service offices of Swiss University has declined by over a third in the last two years. Right? And that's, I think, not just AI, but much of that is AI considerations. So it's going to be quite tough. And how do we deal with this? I think on the same level, there's an opportunity for young people because they don't have these preconceived notions about how work actually works. Right? You can give them an AI tool and just tell them to run wild and solve problems and they will probably find a very creative solution. But this needs a high degree of agency. So I think if you're a young person and approach the current situation with the mindset of, oh, I have to be mentored and people have to explain stuff to me and I first do the simple test before I get some more responsibility. That's going to be tough, right? So you have now the opportunity to really be very leading and show the old people how it's actually done. And again, it reminds me of when I started my career in the early days of the internet, right? I was 25 and we had an advantage because there was literally nobody got any experience with this stuff. So as a 25 year old, you were on the same level as much more experienced people. And we are seeing very similar things now with AI as well. It's not a coincidence that you are already hearing about a wave of founders in their early 20s, sometimes dropping out of university, right? That happens more and more simply because you have this advantage. To go back to your question, no, I don't think we only have senior engineers, I still think we will have this mix of different experience levels. But people will probably go up the experience curve in much different and probably much quicker ways. Now, I think most of the things we have talked about also apply to venture capital firms and the way venture capital firms run and organize their processes. And you've said earlier that you also run your own experiments, you said in an earlier conversation with us that you use OpenClaw to pull some market intelligence reports. So can you give us an overview what the impact of AI has been on things like your deal flow? Yes, sourcing is by now something that is very strongly AI enabled for pretty much any VC for a month. So the days of, we're just going to a conference and all the wait for warm interest, forget it. Everybody now uses AI tools. And this has led to much higher transparency in the market. For instance, I was just at a pitching competition at ETH in Zurich this week. And out of eight companies, I already knew six simply because we had seen them in various AI scanners or sources. And it has become incredibly transparent. So I think the idea of having this proprietary deal flow, et cetera, that VCs used to be proud of and differentiate themselves with that's largely gone. And it's actually the interest of founders that it's largely gone, right? Because you can then probably pick the investors that are the best fit for you and not just the ones that you happen to be able to get a warm intro to. So that's just a more efficient market. But that's kind of table stakes, right? You just have to do this and everybody is doing it. The next step that is really understanding markets and market dynamics deeply. And this is also something that has been heavily automated already. As an example, in the good old days before when I had an initial call with a startup, I probably flipped through their pitch deck, right? That maybe did the Google search or tool and then had an idea of what they were doing. Maybe I knew the market, maybe I didn't. And then you had this conversation, got pitch, right? I don't do that anymore. I throw it into an AI tool. And when I show up for the first call, I already know a lot of detail about this market. The AI has already analyzed the positioning of the company compared to its competitors. It has already pointed out what has to be true, what has to be an outlier success. It has already created questions for me that I could ask, that are very specific, and often find stuff in the pitch that would have probably overlooked. It is suddenly a very different quality of conversation that you can have, right? So it's not necessarily much of a time saving into some extent. It's even harder because you have to digest all this information first. But I think it's a massive step forward in terms of quality. And then, of course, due to the illogical processes and stuff like writing investment memos and so on, we use AI for all of these steps, right? And instead of summarizing content yet again for an investment memo, the agent can do this much better than most humans, right? Easy enough. But, of course, we don't delegate the investment decisions to AI quite yet. I'm experimenting with it. It's actually interesting. Sometimes they give you an interesting perspective, but that's actually where you still feel the limitations of current AI tools. And it's not about the reasoning. I would say the top models now can reason as well as the most intelligent humans, presumably. But it is really about the broader context, right? About broader pattern matching. And yes, I experienced a similar situation 10 years ago and years how it played out and by I think it applies to this situation. So this broader context is not something that an AI model can have, at least at this point, but the purely analytical work can be very heavily automated. And to some extent, that's positive and useful because that gives us a lot more time than to really work with the founders, right? And talk about the really difficult topics. Also, process more deals flow in a deeper way to find stuff that might not be super obvious, right? So you can drill much more deeply thanks to this additional automation. And you already hinted that if you're doing everything with AI, like it sounds as if the question is here maybe also a little bit provocative, but what is the actual value out of a human VC? You say that so far, like still the senior ones who can go back 10 years ago and see patterns, there, yes, there seems to be still a value. But what about the rest? Like, where also when I think about two years, like where do they fit in and where is the added value to do have a human VC in 2026? The standard cynical answer, somebody still has to go to all the parties. But of course, I don't mean that. Even though it's true. I mean, it more serious about relationship building is very essential, right? Yeah, I can't really do that. It's also a lot of service work for existing portfolio companies, where you still need humans to understand what they really need and to make introductions to go into market sense on. I think the analytical part, we are really at the point where yes, you can automate a lot, but you still have to apply healthy judgment. And I see this with our juniors that they have started to heavily use AI tools, of course, all the boring stuff. But of course, they still have to apply their own judgment, their own creativity on how you actually analyze a potential investment case, right? And how you get interesting reference and things like that. That is not something that you can automate where you need human creativity, the judgment and understanding. So I think to some extent, it's very similar to coding, right? You don't have to manually type out code anymore. You can direct various AI tools to do all the dirty work for you and you can go to a higher level of applying this judgment to it. And again, I'm learning a lot from our young or colleagues who are very aggressive about using this stuff and that's exactly the dynamics that we like to have. - And now from the other side, I mean, we mentioned that earlier in our conversation that we're seeing four person teams hitting millions in revenue. So do these companies even need VC or is the venture model itself being disrupted by capital efficiency? - That's a great question. And I think everybody in the VC industry is quite concerned about that because the times where you needed to raise millions to even build a decent first version of a software product are definitely over, right? That's very clear. You can now wipe code in a couple of weeks, something that is decent enough to find paying customers and don't need money for that. However, there are fields where that's not possible. Anything with a physical component is of course very different. That's a reason among other things why we have started investing much more in fields like robotics, for instance, right? You can't wipe code a robot. It's still something that has to be assembled by hand, a digital face at least. And there are of course other aspects of building a startup that are still very capital intensive. You could argue that if it is so easy to build products, selling the product is not easy, right? It has become actually much harder and suddenly you probably need new approaches. You need more capital to actually bring your product to the market. And I think right now we are seeing this development already. I keep hearing from people that traditional sales approaches, outbound sales, et cetera, have stopped working because everybody's now getting spam, if all is AI generated, outreach, males, right? It's basically gone. And you have to get much more creative, right? And this can mean that you have to spend significantly more money on these aspects of building a business. Now one aspect that we haven't really touched upon and I know time is running out, but I still want to ask you that, especially because you have built a company also in the US and you are living in Europe and you are from here. We have a strong focus on how to scale a business from Europe as Europeans. Now you've said that Europe has a lot of catching up to do regarding ambition. Many say that. I think we also got much, much better. But does the fact that AI lowers the cost of entry help Europe close that gap or does it let the US move even faster according to you? I think it can help Europe quite significantly because it's still comparatively harder to get financing here, like cliche, but it's also there's a lot of truth for it that European investors are probably a bit more conservative than their American counterparts. So I think being able to prove something out in Europe very quickly now with little capital could definitely help. It is also an opportunity to get better at go to market, which is still a weakness that we have in Europe compared to our American counterparts, of course. And maybe AI suddenly allows us to automate a lot of stuff that always helps back European companies. For instance, cutting and selling across countries with different regulations and all that terrible stuff, maybe that's suddenly much easier to automate. That's still early days hard to judge yet, but eventually I think we could see that. I don't think the impact of AI on mentality and risk-taking willingness will be significant, but it could be very hard to try to still need to work on that. No doubt. But I think it will open now new opportunities, especially for young founders, just to try something, throw something against the wall and see what sticks, because suddenly you can do this if extremely efficiently. And I think that might be something that could benefit us in Europe quite significantly. Now we're moving to our rapid fire section before we're closing down. So we're just going to ask you four or five very short questions and you can answer whatever comes to your mind first. First question is a thin wrapper startup, a legitimate business model, or just a temporary arbitrage of the model makers? Currently just arbitrage, unless you go very deep in a specific industry that you really understand fully. If you were starting a company tomorrow, would you hire a human CTO first, or would you hire a world-class agent or administrator and write the first 50,000 lines of code with an LLM? I would probably write the first code myself with an LLM coding tool and then figure out what kind of CTO I really need for this particular project. What is the one specific European cultural trade that is actually competitive advantage in the world of AI where usefulness trumps raw intelligence? Europe builds technology that actually works and we have a long history of doing that. And I think with AI, a lot of people are doubting, is this stuff reliable enough? And so on, that's good justification for these doubts. And I think European companies have the sense of quality that could really be a differentiator for that. And in addition to that, we also have a long industrial history, of course, and tradition, so all this new robotics wave that we are currently seeing, I think we are very well positioned for that. Which will be more valuable in 10 years, the company building the most human-like AI or the company that quietly automates 90% the foreboring back office process like insurance claims. Probably both doing well, but I would say probably the one that is doing the boring stuff predominantly. I'm not entirely sure that in a few years, these are necessarily going to be very different companies. To be honest, we'll see. And finally, if you could give one reality check to a founder listening right now, who is feeling overwhelmed by the pace of change and all the things going on, what would it be? - I would ask myself, are you trying all this stuff deeply enough by actually playing with it? I have a personal rule now that I don't do any work anymore before I have not tried to automate it with AI. And this really teaches you what the possibilities are, but also the limitations in a very tangible way. And I think you can only apply AI effectively on your own company, your own organization when you really have this deep intuitive understanding from actually doing it. - That is very wise closing words. The very last question. Now, if a founder is listening or a founding team, and they want to learn more about you and about your work, you do at B2Vensha, how can they get in touch with you? - Easy to find, Google me. Difficult to find. We have a website on B2Vensha.vc, where we also, by the way, have a sub-website called the B2Vensha startup resources, which is a collection of various handcraft and the hand-selected sources about all aspects of startup management. I often hear a lot of good feedback about that because it takes away the doubt about what should I actually believe. That's why we put it together. And you can also easily find me on LinkedIn. I also write frequently about AI topics. That's my Saturday morning ritual that my agents such as 10 topics in a pick one and then write about it mostly by hand, by the way, still. But yeah, I'm easy to find and I almost look for what to interactions with people about all these topics because we're in the middle of a very rapid development, obviously, and nobody knows the answers, but we can have smart conversations about it, hopefully. - And we will put everything of that in the show loads. And obviously, if anyone listening right now, Mrs. and SA of you, I think we also referenced that a couple of times now on newsletter, so you can pick it up there. But Andreas, thank you very much for the very interesting conversation. - Thank you very much for having me. (upbeat music) - And this was another episode of our Follow the Gradient Podcast. - And now we would love to hear from you. How did you like it? What should we improve? Do you have any suggestions, ideas, a dream podcast guest? Please let us know. - Thank you and see you again next week.