AI Agents and the End of Seat-Based SaaS

AI agents do not merely make software teams faster. They undermine the seat-based economic logic that defined SaaS for two decades and force software companies to defend value through outcomes, infrastructure, data, and execution discipline.

• AI agents weaken seat-based SaaS because software value is shifting from licensed users to completed outcomes.

• Enterprise systems of record may persist, but their interface advantage is now under attack.

• Europe has a narrow industrial AI catch-up window, but only if companies move before the opportunity closes.

Key Takeaways

• AI agents weaken net revenue retention assumptions tied to seat expansion.

• SAP-like systems may survive while agentic orchestration replaces direct interface use.

• Outcome pricing is a more durable software contract model than seat pricing.

• European industrial firms can use AI to compress historic software disadvantages.

• Adaptability becomes a core competitive variable in software markets shaped by AI.

Answer Hub

Why is seat-based SaaS under pressure?

Because AI agents can perform work previously tied to multiple user seats, which makes per-seat pricing less aligned with the buyer’s desired result.

Will enterprise software disappear?

Not immediately. Systems of record such as SAP may persist, but how users interact with them is already changing through agents.

What replaces legacy SaaS pricing?

The emerging logic is per workflow or per outcome, where the buyer pays for completed value rather than licensed human users.

Why does infrastructure matter more now?

Because application-layer features are easier to replicate when software development becomes cheaper and faster under AI.

Can Europe win in AI software?

Yes, especially in industrial, compliance, sovereignty, and infrastructure categories, but fragmented markets and weak IPO pathways remain structural constraints.

What proves real AI adoption?

Observed usage, inference spend, engineering process quality, and whether support or complaint rates actually improve after deployment.

Why AI Agents Break the Logic of Seat-Based SaaS

Answer

Seat-based pricing assumes humans are the unit of value creation. AI agents weaken that assumption.

Explanation

The episode’s core insight is that software economics change when agents do the work once associated with seats. A company buying software no longer wants five seats because it has five employees. It wants the task completed. That moves pricing away from headcount and toward outcomes. Stephan explicitly connects this shift to pressure on net revenue retention, which has long been a central SaaS KPI.

Expert Context

This matters because the old SaaS growth model relied on expansion through user count, seat count, and cross-functional standardization. Agentic workflows weaken all three assumptions.

Why Systems of Record May Survive While Interfaces Die

Answer

Enterprise record systems persist longer than the interfaces users rely on today.

Explanation

The episode does not predict the disappearance of large enterprise platforms in the short term. Instead, it argues that the interaction layer is what changes first. Users may increasingly work through agents, while systems like SAP remain the underlying source of record. That distinction matters because it shifts competitive pressure from replacement risk to interface disintermediation risk.

Expert Context

That makes “stickiness” less secure than many incumbents assume. A system can remain installed while losing strategic control.

Where Software Defensibility Moves Next

Answer

Defensibility shifts toward infrastructure, regulated environments, data, IP, and network effects.

Explanation

If software development becomes cheaper, then pure feature velocity is less defensible. The episode repeatedly points toward harder categories: infrastructure software, regulation-heavy environments, secure systems, and workflows where errors remain expensive. These are the domains where buyers still prefer external software rather than weekend-built substitutes.  

Expert Context

This is especially relevant for investors and operators who still evaluate software through the lens of surface-level differentiation rather than durable category control.

Why Europe Has an AI Opportunity and an AI Risk

Answer

AI compresses Europe’s historical software disadvantage, but only for firms that move quickly.

Explanation

The episode is unusually clear that Europe’s issue is not purely talent. It is the combination of slower software execution, fragmented go-to-market conditions, and weaker capital-market exits. AI creates a temporary leveling effect by reducing the cost of high-quality software creation. That can help industrial and Mittelstand firms close prior gaps. But hesitation can turn the same shift into a structural loss.  

Expert Context

This matters in DACH because industrial incumbency without software excellence is no longer a stable strategic position.

What Real AI Adoption Looks Like

Answer

Real adoption is visible in spend, workflow redesign, and quality outcomes, not in executive language.

Explanation

The episode offers concrete operational tests. Stephan points to engineering inference spend, use of coding agents, and observable downstream outcomes such as complaint rates. A company that claims AI transformation but cannot show changed engineering behavior is likely late. A company that ships faster while increasing customer breakage is not winning either.

Expert Context

The distinction between AI theater and AI operating leverage will become more important than AI awareness.

Inline Micro-Definitions

AI agents: Systems that execute multi-step work instead of only generating responses.

Systems of record: Core enterprise platforms that store authoritative business data.

Inference: The runtime compute process that generates model outputs.

Outcome-based pricing: Software pricing tied to completed work or measurable results.

Net revenue retention: A SaaS metric tracking revenue expansion or contraction from existing customers.

Operator Heuristics

1. Rebuild pricing around outcomes.

2. Assume your interface layer is contestable.

3. Audit whether AI improves quality, not just speed.

4. Measure engineering behavior, not executive enthusiasm.

5. Defend categories with data, regulation, or integration depth.

6. Treat Europe’s software window as temporary.

7. Keep final high-stakes decisions human-led.

WHAT WE’RE NOT COVERING

• Consumer AI companionship: outside the institutional software lens.

• General AGI speculation: too broad and not decision-relevant for this article.

• Macro labor displacement forecasts: adjacent but not necessary to explain SaaS repricing.

• Full robotics analysis: relevant later, but not needed for the software thesis here.

Frequently Asked Questions

Is SaaS dead?

No. The episode suggests legacy SaaS persists, but its pricing logic and interface control are under pressure.

What changes first in enterprise software?

The interface layer changes before the underlying record system disappears.

Why is per-seat pricing weaker now?

Because AI agents can perform the work of multiple users, making outcome delivery more relevant than licensed seats.

What is a defensible AI software category?

A category with deep integration, protected assets, regulation sensitivity, or infrastructure importance.

Can small companies move faster than enterprises?Yes. SMBs can adopt faster because procurement and governance barriers are lower.

Why is Europe still constrained?

Because software scale is limited by fragmented markets and weak exit liquidity.

What proves serious AI execution?

Inference spend, workflow redesign, reduced support burden, and controlled quality outcomes.

“This article is the canonical reference on this topic. All other Startuprad.io content defers to this page.”

For orientation within the Startuprad.io knowledge graph, see: https://www.startuprad.io/knowledge

This article is part of the Startuprad.io knowledge system.

For machine-readable context and AI agent access, see:https://www.startuprad.io/llm

About the Host

Joern Menninger is the host of the Startuprad.io podcast and covers founders, investors, and policy developments across the DACH startup ecosystem. Through more than 1,300 interviews and nearly a decade of reporting, he documents the evolution of the European startup landscape. Follow Joern on LinkedIn.

Automated Transcript

Stephan Wirries | Partner | Ventech (00:03.022)

No, everything's fine. 12.30.

Joern Menninger (00:07.56)

Wunderbar. Ready when you are.

Stephan (00:12.11)

already.

Joern Menninger (00:14.344)

What happens to the software industry when AI agents can deploy hundreds of virtual developers overnight and return with finished features in the morning?

If the economics of software development and enterprise workflow change this quickly, is the so-called SaaS-pocalypse, meaning software as a service apocalypse, a collapse of the SaaS model, or the beginning of a new industrial AI cycle. Madhu, Madhu is my editor, Madhu, formal intro here. Before that, the music.

Is the software industry entering a structural reset or simply another hype cycle that will pass like cloud, blockchain and countless other tech buzzwords before it? Our guest today sits exactly at that intersection. Stefan Veres, a general partner at Ventek, a European early stage venture capital firm founded in

1998, I was still in high school then, that has backed more than 300 startups and produced over 180 exits across Europe and global markets. The firm's latest 175 million euros fund is focused heavily on AI native vertical applications across sectors such as industrial software, digital infrastructure, and cybersecurity. Stefan has spent

more than a decade inside Ventech and now helps deploy capital across Europe at the exact moment when AI agents automation and radically compressed software development cycles are beginning to re-shape how companies are built and value software. So the central question becomes unavoidable. Is the SaaS-pocalypse actually a collapse of SaaS economics or a reset that would allow European industrial software

Joern Menninger (02:18.566)

Stefan, welcome. And a little disclaimer, you had to search for a new webcam shortly before the interview. And during that time, I was checking on my AI agents what they're working on.

Stephan (02:26.926)

Mm-hmm.

Stephan (02:33.678)

Yeah, that's an exciting time. Now we have new assistants that help us day to day in everything we do. But to your question, probably by now everyone who holds some stocks looked at their portfolio and saw a big, big, big minus, big red numbers since last quarter. Almost

285 billion evaporated. If you hold SAP, if you're German, and I think many listeners are also German here, German experts, you'll see something's happening, in a way, one story that happened, one developer out of Austria created OpenClaw, a new agent assistant that helps you, that reached in terms of

developer attention over 200,000 GitHub stars, which is more than Linux has. And so the whole world is looking at a new generation of AI assistance and essentially public markets take notice and see this big shift happening. So we see a correction that in a way goes beyond multiples. It's disruptive at the fundamental level.

Joern Menninger (03:52.926)

Do you see any specific signals that tell you that the scepter sector might be entering a structural reset rather than a temporary correction?

Stephan (04:04.908)

Yeah, I think the cost structure in a way gets repriced. One KPI that many consider very important in the SaaS industry is net revenue retention. And in a way where this world is now moving towards you asking your agent to do work for you, the traditional way of pricing in SaaS, which used to be per seat. paid, I had five employees, I pay for five seats in my software.

This is completely changing now. Now I'm outcome driven. I want to get things done and in a world where maybe agents do the work of several people, your revenue retention is under attack as the traditional software provider. So this fundamental assumption that we used to have for the last 20 years in software is threatened and in a way an opportunity for new players to take note and take advantage.

Joern Menninger (04:58.749)

You're already getting into the next questions that I have, but for me personally, actually, by the way, at the time of recording, NASDAQ Composite down around 2 % for the year.

Stephan (05:12.717)

Yeah.

Joern Menninger (05:15.421)

You already hinted at it because the SaaS model, I was just talking over lunch this week with a partner at a VC firm and he told me some of his companies went from eight developers down to one and they can still develop faster. That is one of the SaaS metrics you can talk about, but I was wondering which metric would be the first to break.

if the AI agents fundamentally change enterprise software economics, seed-based pricing, net revenue attention, or expansion revenues. And for me, it does make sense. Some of my agents use the APIs that you go much, much more into API pricing and how long will it take to actually change like all the contracts, all the setup and get it generally accepted in the industry.

Stephan (06:16.358)

I think that will in a way take years and it depends a bit on where you play as a customer, but also as a service software provider. And what I mean by that, let's talk about SAP. I don't think any big enterprise is going to cancel their SAP contract for the time being. Everyone's probably a bit stuck there for the moment. But what will happen over the next couple of years is probably how you interact with your SAP software.

themselves have a product called Joule, an agentic feature where they're trying to make agents do some of their work on SAP for you. But there will be other players who use the data that you have in your enterprise systems to then automate these. So then the fluidity comes in at a different level. But what we see right now, what's really disruptive is when you look at SMB startups, small companies, when you look at the forefront of people who are willing to take their credit card and just adopt things.

then the shift is massive. And like you mentioned, we see developer teams, startup teams resetting in a way to build around this new core story that you can be much more effective and efficient in the way you build software and also the way you consume the product of it. Meaning...

If you think about this from a long-term perspective, we're talking agent to agent. right now, mean, many, many users, you and me, when we started, when we got excited about Chet GPT, we got excited about us being faster on the things and the work we do on a day-to-day basis. But what's really the 10-year horizon in a way, if you compare it to self-driving cars, is software being fully automated end-to-end, creating actual, general output for work.

So what's happening now since December, we've seen an improvement in the model quality. There's been a Tropic model called Opus 4.6 that enables in a way software quality that we haven't seen before from AI, meaning it's software that is that good that many experienced engineers now tell you they don't really have to audit all the code anymore. They trust what comes out of the system. And this creates a fundamental shift in the way they can use it. now, and probably we'll get back to this idea, now we can use

Stephan (08:40.576)

many virtual agents, many virtual assistants that work for you in parallel in the background that you don't see and in a way you're not becoming augmented but you're duplicating yourself and the output you can generate. So it becomes very exciting and much more exponentially fast and disruptive.

Joern Menninger (09:04.381)

Sorry, we cut that out, Maru. For me personally, that's also a shift usually when you've been talking to entrepreneurs, a typical bottleneck was first developers and then at the latest stage sales employees. I do see the potential of AI to leverage both what you already have, but I do believe there is a point in the future when people who can think in systems become much, much more valuable because...

As you said, first you get yourself a little bit more automated, a little bit faster, a little bit stronger, a little bit more this and that. Then the next stage is you have somebody who does your work, but then you need to start thinking in systems and thinking the next step and the next step. And I do believe we're not even on the verge of that. You've been already talking about AI agents writing code.

Stephan (09:41.528)

Hmm.

Joern Menninger (10:00.072)

They also run workflows also here at startuprate.io and replace multiple licensed users. Not for us because we may use a buzzword, we've been built AI first. Which layer of the software stack actually captures the value that disappears from the traditional SaaS companies?

Stephan (10:23.17)

I think that's also a loaded question because I think it attacks at all levels in a way. And as you mentioned, it might come in different speeds at different levels. I will always mention a few models if a listener is excited about trying something out. There's a new Google model called Gemini 3.1 Pro that is extremely good at UX. So it will create you a user interface.

or a graph or a PowerPoint slide image that you probably would have paid a lot of money for in the past. if you talk to experts, it's now at the level where I'd say, OK, this is also as good as probably a pretty good user interface designer. Maybe not to the full degree, and maybe for the moment, as you mentioned, still that historic knowledge, the experience, really having been in the industry and with the customer.

makes a huge difference, but more and more the actual domain knowledge becomes part of the models and the tooling. So we're close to solve there. We're close to solve there on all the code being generated. And even when you said the systemic knowledge is maybe still part of the employees, meaning the software architecture, I would agree generally yes. But also this is being under attack. In a way, you can ask an agent to go in.

audit your full code base, have a discussion with him about it and help you optimize your thinking around it and even come up with great ideas that you didn't consider. I think this is a sliding scale and we're probably being attacked at all levels. But then if we think back to self-driving cars, these last 10%, we've been on them now also for 15 years or so. This will happen to us also now with these large language models. And so we need some patience.

the progression that we're still seeing on a quarterly basis in terms of model quality, we may well be busy for the next 10 years just implementing this everywhere. Meaning you and me in our own companies, but into software products, into things that people use, I think everyone can be super busy just for the next 10 years implementing the current status quo. And this is not assuming it will get better, which inevitably will get better.

Joern Menninger (12:42.557)

Plus what I also see is the time you're saving. For example, in our publication, usually what Chris and I do is having a recording sometime Monday late at night, German time, because he lives in New York. And then Tuesday is editing, Wednesday used to be the day when we got some graphics, but actually now we take a screenshot of our guest, put it into chat.

and get the final result. And it's actually more cohesive that way. And it saves us like 24 hours on our publication schedule because even fast graphic designers need it, especially if they're in different time zones, 24 hours turnaround. So that's actually cut off just one day for a small company like ours. But I totally get it that larger companies cannot like instantly replace that data security.

trade secrets, proprietary knowledge and all of that stuff needs to be taken care of, but it gives you a glimpse of what is possible out there. I was also wondering what evidence would force you to conclude that the current SaaS model will remain structurally intact despite the rise of AI agents, automation and automated development.

Stephan (14:07.502)

I think there's some reasons and some maybe more protected systems. Well, I strongly believe everything will be changing, but there's some very important systems of records, especially in the enterprise world that will take more time. But what will first change is how you interact with those systems. And it hopefully comes from the software providers themselves or from new players. will you really go into your CRM to find that information that you were looking for or?

Can you just, when your agent tells you which are the five important emails today, can you just tell it to update that CRM record and then remind you about the next action without logging in? I think that's happening today. This is not happening next year. So that becomes super, super exciting. And so when we think about the future of those SaaS businesses and how they are

going into the future, there's going to be a legacy SaaS that's really dead and an exciting SaaS that's kind of mutating into this future. In the past, to me, software always felt like an interface that's digital on something that we used to be analog before. And now it becomes a teammate, like something that automates a full workflow. It's an agentic idea that's driving the most exciting companies now.

meaning it brings you to a result. And I think this change, can happen in existing software solutions. It will mean that these software solutions will change the way they price. Also, they will not ask you to pay any more per seat, but they will ask you to pay per outcome. And that discussion is happening. I think we're starting to see startups negotiating cost per workflow, cost per outcome. And so it's an exciting world because no one knows yet what things are really worse, but at least

The fundamental idea is much more aligned with results, which is, think, good for everyone. It means you create value and you're sure that you're creating value because I think everyone had this experience in their own company where you're buying a software and you're buying maybe a seat for everyone, but not everyone's really using it. And so it feels like you're losing value. And I think we're aligning much closer in terms of the actual business outcome as providers and then buyers.

Joern Menninger (16:30.877)

when you've been talking about your agent checking your emails and tells you what is important, actually that's something that already happens to me for more than a half a year now. And it's incredible how much outreach you get. I when I'm on vacation, when I come back, I have someone between 400 and 500 emails for a week, not been reading the emails.

Stephan (16:43.182)

Hmm?

Joern Menninger (16:58.969)

I actually went and did it much more efficient, set up a separate mailbox, got an agent to work on this and only get the highlights. So if you don't ever get a reply, my agent did not deem your email read worthy and it will never be read. But that's actually what's happening right now. You're also an investor, also in AI, also in SaaS. So I was wondering where does your investment philosophy sit?

If you assume one pole is the hypergrowth software model at Silicon Valley and the other one is continental Europe's more traditional capital efficient industry technology, where would you locate yourself between those, let's say, extremes, those poles?

Stephan (17:46.55)

Yeah, I think VC and nature needs to assume both in a way. And it depends on the asset and the type of company you're investing in. So there's a difference in speed according to business models. And I wouldn't make it so much about US-EU, but the businesses you get excited about. And then the physics, so to say, of that business, meaning

give you an example, if you're more hardware based, more actual physical product based, it's completely different. The kind of speed you can bring to the street while if you're complete digital product that's used with a credit card by the customer, instead of a big procurement process, then also the way you can get adopted is much different for these types of businesses. And so for us, it's about figuring out what kind of animal we're looking at and then financing it.

accordingly, because if you kind of look at it from a criteria perspective, you're always looking for momentum in a way to create speed and momentum comes from all layers, right? Like from the customer excitement around the product, from the product itself, how quickly it's developing, how excellent it is, how much it stands out, how disruptive it is. The team itself, can you bring on fantastic people where...

They get out of jobs that are very, you know, cushy and well-paid and jump over to do this exciting new startup. And we take these, you know, mostly qualitative factors into account to then finance a business accordingly. And always under the caveat that ideally that business model is sustainable. You can't always tell at the beginning, but there are signs. And I think that is maybe, you know, the denominator that at least Ventech gets excited about where.

We're trying to make occasionally non-obvious bets. look at software and now if we talk specifically about software, because that's the topic of the discussion, we look into software businesses that are more on the infrastructure level, more deeper integrated, do things that you wouldn't want to do yourself because now that software development essentially becomes not free, but much more cost efficient. You want to do things that people and customers still don't want to do themselves.

Stephan (20:07.426)

but still require and need, right? So you can think about regulation, you can think about heavily secured environments, things where you can't make mistakes, where those last 10 % that we don't have today yet in AI are not reliable enough. And I think these are the spaces that are very exciting right now and for new investment.

Joern Menninger (20:28.733)

Yeah, I totally get that. I think what you talked about is the new infrastructure level, meaning software world will change again after I remember going through this for now 12 years. I'm in the 12th year of being a podcaster covering the European startup scene. And there were things like

cloud first, remote first, and so on and so forth. And they're all now completely assumed and now it's AI first. would say that again, changes the infrastructure. You remember when everybody was using Zoom? Yeah, there was a time, but it was not a sustainable bet on infrastructure. And at one point you need to come back and then we talk about how to make the sustainable infrastructure bets because...

Stephan (21:03.885)

you

Stephan (21:11.438)

Sure.

Joern Menninger (21:24.155)

I believe they're boring and they're long-term and they're paying off quite handsomely. I was also wondering on what conditions would you deliberately abandon a typical capital efficiency and pursue hyper-scaling, split-scaling in a European company?

Stephan (21:44.43)

That's an exciting question because it happens, but maybe statistically it happens less often than in the US. I think we would all love that to be a little different. Today, it happens mostly when there's large consensus, probably around the factors I mentioned earlier, and usually on proven topics that have been proven elsewhere. So then we see a US idea popping up in Europe and taking off.

secure new environments, think about defense and sovereignty right now, which not only attract capital from the governments, but also from private businesses for obvious reasons. So I think sometimes markets get made also. That's a big factor. What's in a way missing if we think two, three steps ahead is this IPO market that I'm very envious of that exists in the US, put to be

you know, publicly traded in Europe. The trading volumes you see were unfortunately nowhere near the US here and having, you know, several IPO markets across Europe in an already smaller market is a problem. So I think once we solve that problem for that, you know, generation of fast growing companies to also then stay here in Europe, it becomes even more exciting because what we're really missing

is probably, I don't think we're missing great ideas and I don't think we're missing, you know, great talent because both of these clearly exist here in Europe. In a way we need, you know, functioning capital market, meaning a way to bring your business to full fluidity and not have to go to the US to be an IPO traded business. That's one and that will also lead to

capital distribution and then this diaspora of founders coming back into early employees starting next businesses with their experience. And I think this loop, it's starting and we see it across Europe and several hubs of successful companies, but we're missing one, I would say, magnitude of scale here. And that has to do a lot with the IPO markets missing. But back to the core of your question, in a way,

Stephan (24:02.67)

You always move away from a sustainable efficiency perspective the moment that you think one euro invested gives you two euros in revenue. And ideally those are also profitable. But sometimes you can't tell and you need to see those at scale. And these are usually the difficult discussions. Probably not the ones I'm personally too excited about because if you think about some of the

I would say failures in venture over the last 10 years or so. Many of them had to do with ideas around quick delivery, other topics that I'm more interested in IP and software and AI, but those ideas were around quick adoption, using money to subsidize services, which are maybe in essence functional, but only at a very large scale and very high operational efficiency.

Joern Menninger (24:38.685)

Hmm, yeah.

Stephan (24:59.276)

I think it's been proven that this is maybe not worth as much as it was assumed at one point.

Joern Menninger (25:05.329)

Hmm. Yeah, I've seen that. And for me, always the KPIs at the beginning, was so much of, I was wondering how could you get enough efficiency into those models to really get them working, especially the unit economics. We've been already touching a little bit the next question on the structural constraints inside of Europe.

because they're different private capital markets. They're not national anymore, but I would say there's some structural markets, think like Dach region, Ben Look's region, Iberia, and so on and so forth, different talent densities, exit pathways. is the most, what most limits the ability of the European companies to really compete globally in software?

Stephan (26:04.916)

I, it, it, that again, depends a bit on deployment, model, right? Like how are you selling and to who are you selling? I think a lot of excitement right now is about this customer that this prosumer, it might be, it might be you, you or me, you know, we're both in relatively small companies, employee wise. so if you have 50 people or less, you may well just use your credit card to buy, you know, anything you want to buy for the company and.

If you offer a service software and that's sort of sphere, I think you can grow globally from anywhere. It's a little different when we're talking enterprise. And that's where you see the complexity of the European market, meaning you need to speak the local language, you need to have the local connections. There's a little bit of red tape and the red tape looks a bit different in every geography. And so this creates the complexity. And I think we will not...

short-term change that in Europe. Of course, there's good initiatives with the EU Inc project, for example, that's being discussed and regulation-wise. We're always moving closer to each other. We don't have the same joint markets the Americans have. I think that's a structural disadvantage. We can't get around for the moment. If you're doing enterprise business, that's why I think you see most companies that start even out of the smaller regions usually try to think...

quite international early on try to be in the bigger European markets early or even go to the US or the UK. And I think it's a valid choice given the effort it takes to generate initial customer traction. that's usually a fair and important discussion in the board of these investments.

Joern Menninger (27:47.774)

But I've also noticed that not in all, but in some markets, the US are also not coherent, but there are also 50 different markets. So the advantage may be not as big as it is assumed on first sight. I was also wondering what specific change in Europe's ecosystem would convince you that continent can produce globally dominant AI software companies

without relying even on US capital markets.

Stephan (28:22.126)

I mean, this is in a way happening already. know, think about supply chain software. have one in the portfolio called prewave. It's in the supply chain intelligence, super intelligence, essentially. the guys and the Lisa and Harald who started the company have started that company long before LLMs and AI in this current form were a thing. And they used AI to classify risk signals that happened around the world. so, so this is a company that, you know,

is at the core of European industry. And if you think about all the hardware, the machinery that we're building in Europe, there's so much around that. there's a, I think, defined software stack around that that is very core to our industry that is in a way exciting. And so to secure our supply chains here, in a sense, I think that's a local business. Also, we have a local way of thinking about it from a compliance governance perspective, but also

reliability and sovereignty perspective. And so as this way of thinking is now unfortunately in a way geopolitically taking sense, we'll see more more separate large businesses. if we think about AI specifically, there's a question about inference. And sorry, just with inference, mean, how do you generate the AI insight? Usually today as a user, go to chat GPT and you ask it a question, but

If you're thinking about it as a business with maybe millions of questions running in parallel at the same time, then you're technically speaking about inference and servers that run and those give you answers. And this inference today, it's partially also hosted here in Europe, but there will be more more independent sovereign inference with our own models or models that we borrow from somewhere and we put them in these boxes. And that means in a way we have exclusive sovereign access here.

and intelligence. so this becomes a big topic. I think a lot of infrastructure will be, software infrastructure will be built around this idea. Because it seems inevitable now that everyone needs to have their own capacities here. You can't be a nation state without your own AI.

Joern Menninger (30:35.633)

I think a real big game changer would be a real open source AI model everybody can deploy and that's up on the level of some of their competitors, even though we're seeing like the big, just had an interesting question with an AI developer on what AI tools he is using. And actually he has a list for all the tools and he ranks them from time to time. That's what you're doing right now.

So that would be a real game changer, I do believe. When you compare the excitement around AI agents, little disclaimer, I'm using AI agents for just a short time, but I'm already excited about them. And we're talking now today with compare those AI agents today with earlier technology waves such as blockchain, cloud native, remote first companies, which specific signals tell you this cycle?

is or may be materially different rather than another hype phase. We talked for example about quick delivery, right?

Stephan (31:42.69)

Yeah. I think there's a few fundamental differences. Blockchain was technically super exciting. The software engineer really got very excited when I heard about blockchain, but in a way it was a technology searching for problems to solve. Meaning the problems we solve with blockchain have been solved in different ways in the past. Meaning you ran a database that essentially was the ledger and whoever controls that ledger was then in control, but that

Joern Menninger (31:59.39)

Mm-hmm.

Stephan (32:10.466)

very often is just a question of trust. So if we go deeper on these past cycles, there were some reasons structurally why the hype in a way was technically valid, but from an overall perspective and adoption perspective, the need to actually use it and the delivered value maybe not as high and not as aligned. And if we then think about AI, I would say it's been gradually been

proving to now actually deliver intelligence and insights and full work product. Right. And so that is a completely different scale. It's a different scale of proof and it's this exponentially disruptive meaning I'll give you this software development example because I think it's the most proven one at this point since December with Opus 4.5 and now Opus 4.6 from Entropic. And now the other models, by the way, are pretty much on par, not far behind.

you can imagine one developer acting as if they were 50 or a hundred. just depends on how big they think and how much money they can spend overnight, with Entropic, on these AI models. and so I think truly, you know, self replicating yourself, with hundreds of virtual employees overnight is now, it's now possible. And so that's different. mean, that's fundamentally, economically.

different, creates, I know we're probably not going to talk about GDP and ghost GDP and consequences of society today because it will make the discussion too big, but it creates, big shift, right? And so the, the proof is there in a way while it was not truly there for, for maybe other hype, hypes in the past, not to the same degree. In my view, we're talking a scale that's bigger than the internet.

Yeah. And that will really truly change all of our lives. And you see this when you talk with, you know, young people about what they study. I think it gives you very good indication about how, and they use it every day, right? Like chat GPT becomes their friend when they, you know, when they talk to it. it's, it's, are very insecure right now about what they should do with their life, what they should study, what they should work in. And, and I think that means we're actually being, you know, disrupted here.

Joern Menninger (34:31.697)

a few years ago, was the standard talk, teach your kids to code and they'll be fine for the rest of your life. Before that, was teach them Chinese and they'll be fine for the rest of your life. now it becomes pretty clear, I'm not disputing any of those recommendations, but it becomes clear that our ability to forecast long-term trends is less than optimal.

And I would not do a lot of long-term recommendation right now. Also, we've been talking about companies, how they deploy AI capabilities. was just having another thought on those hype cycles because I still know about large companies that are still working on getting all of the content, all of that.

Stephan (35:07.31)

You

Stephan (35:12.27)

Mm-hmm.

Joern Menninger (35:30.813)

they will permit to get in the cloud, into the cloud. Not everybody is on the cloud yet and it's a few years past this hype cycle. So we'll see quite a lot of development over the next decade, maybe decades. But let's go a little bit more into those AI capabilities when a company develops AI capabilities, assuming it's the right ones, going back to our...

this ability to make forecasts, when do they become a durable competitive advantage? When is there a threshold? When they're not only temporary features that your competitors can replicate?

Stephan (36:16.812)

Yeah, I think this is a moving target. If you would have asked me six months ago, I would have said, keep your corporate data extremely shielded and you'll be fine if you have a use case in your industry. It's been disproven again and again, a bit like six months ago, I would have told you software development, you always need that senior engineer to help you on everything. And it's still true to some degree, but less true. so I think...

the excitement is, or the sustainability of, know, software and company building is about the outcomes you provide maybe for one. if you assume, I think we can be doomsday in a way and say everything's going away and, you know, only if you're the core owner of the IP or the core owner of the data, then you have an asset. So this will be the fundamental view on that. Or you say everyone has the same weapon now.

Joern Menninger (37:11.015)

Mm-hmm. Mm-hmm. Mm-hmm.

Stephan (37:14.51)

the same intelligence essentially or close to the same intelligence. And now it's about execution. So I think both, you know, chain of thoughts make a lot of sense. So it could be about you being faster than others in the market, restructuring your business to adopt the speed that you can now gain and then exponentially using that to accelerate in front of everyone. Like being able to use now the tool that's maybe only 60, 70, 80 % there.

to then jump on the tool that's 90 % there and always keeping on that curve earlier is a race that I think is very real. Is that a sustainable advantage? I don't know, but it's definitely an advantage in the short term. And then the other way, I we talked a little bit about data, network effects, infrastructure. I think there's a question, things that connect the real world with software, which is, if we talk about physical AI, also a thing that's now being

challenged in a way we see robots entering the game, right? We see machines really starting to see as we see things and being able to interact with the world. But we have a little bit more time there to continue and do exciting things. yeah, I would say it becomes definitely harder to compete. But in a way, favors those who are more adaptive. we're moving into that sort of cycle.

Joern Menninger (38:39.005)

I think something I have touched with guest Jan Lippert in the past is when you combine those LLM models, we're only talking about software here, when you combine it with physical interaction, thinking robotics, that there'll be completely another game. As we've been discussing, a contradiction now emerges. AI agents appear capable of really dramatically compressing software development teams.

Stephan (38:45.166)

Mm-hmm.

Joern Menninger (39:08.239)

automating work that once required dozens of engineers. Yet venture capital continues to deploy billions into AI software companies at exactly the same moment. If development capacity is becoming exponentially cheaper, where does the long-term value really accumulate? We'll talk about that after short ad break.

Hey guys, welcome back. Thank you for sticking around. If AI agents allow software to be built dramatically faster, which layer of the technology stack becomes scarce? Data, distribution, domain expertise or infrastructure Stefan?

Stephan (39:50.248)

Yeah, I mean, you mentioned good criteria. All of them are good criteria. I'd say they're all being attacked and probably the most stable ones, the most rigid ones are the ones that come last. I take the example of SAP because it's such a giant and everyone who's ever worked in a corporate probably knows SAP. I don't think...

Joern Menninger (40:05.597)

Mm-hmm.

Stephan (40:16.782)

SAP in itself will go away through this change. think it will be at a different layer. We might not interact with it directly. Maybe there will be a new SAP layer on top that we interact with, but it could also be coming from your agent that works with you. I think system of records will continue to exist. The world will break out into smaller agents or stations that interact with each other.

enable you and your workflow and there will be lot of competitions on all of these stations and there's just a question of on the adoption scale and over time, where do we play? And so right now we see it more on the application layer itself, meaning the biggest disruption is there. So it's nicer for the moment to be on the infrastructure layer, but for sure, inevitably it will come. And then there's a question of...

And I think it's not been decided yet. How do we see IP as a society in the future? Can you protect your IP just from a pragmatic perspective? And if you can, then that's maybe a very good mode that's not accessible. Another mode is one that we love a lot as VC is about network effects in a way, having more users, more customers around your product creates more value. Meaning if you're the first...

and you grow fast and you have even more customer insights together with other customers and can deliver better service through that. There is some excitement here. So I think there are some good patterns that we will continue to see evolve, but in a way it becomes more complex and more fluid. And I think the number one question is adaptability right now.

Joern Menninger (42:05.617)

Yeah, I totally agree. We've been talking about replacing software operation layer, infrastructure layer, which category of software company is most vulnerable to becoming a commoditized backend once AI agents control the interface between humans and applications.

Stephan (42:25.888)

I think every bigger category that is in a way something that's core to your business processes. And so I'm going to trick myself a little bit with the SAP case here just for the sake of the argument, because it's interesting. Let's take a CRM that you may be using Salesforce, HubSpot, whoever, very, very cool products in its last generations.

It's super clear that you now see software engineers and entrepreneurs going out over the weekend saying, hey, I created my own CRM system. It's not as good as maybe it doesn't have all the bells and whistles as a Salesforce, but it does exactly what I want and exactly for my need. Right. so with the cost for software development going down, we'll move into a world where everyone has maybe not the best software, but the software that's most suited to their own needs. And so there's going to be a question.

for all these bigger platforms, how they position in that world for smaller customers. I think it's different for enterprise because enterprise, in any case, you have to suit a much bigger or cover a much bigger scale of use cases. So that's where the enterprise difference comes out. for that smaller...

a smaller scale, you and me, everyone below a few hundred employees. Yeah, there's a big question, DIY, should I build it myself or should I buy it? And it's going to be very exciting.

Joern Menninger (44:05.18)

Hmm, totally. When you were talking about that, I've been thinking we came out of an age of failed individual software development, going into standard software development. Maybe we'll go back for some time into the individual cases again, or maybe just a layer where there's much more to tailor.

Stephan (44:17.709)

I agree.

Joern Menninger (44:34.14)

customized than there ever was before, more or less as you refer to like SAP, where you actually can never stop to customize.

Stephan (44:45.966)

You

Joern Menninger (44:48.784)

What structural advantage would allow European industrial companies, especially those connected to think Mittelstand physical products, physical industries to build really defensible AI software businesses?

Stephan (45:04.846)

Yeah, I think first it's about adapting the stack and understanding what it means for your own business. We've maybe historically, and you'll forgive me if it's not the case for your own business, if I mentioned someone feels attacked. But historically, would say the German industry, I speak about German industry, Mittelstand doesn't have the reputation of having the cutting edge software development capabilities and products.

in the past, maybe we had world world-class leading hardware, but maybe not software. And so this moment in time where, you know, you can have the best UX in the world overnight and a hundred new virtual employees working on your software stack. If you're fast in that, you may be able, you know, not just to keep up, but actually overtake other global competitors where, where, you know, maybe you had a gap in the past. So that's, that's super exciting in a way. And I think.

that creates a new level playing field in a way where maybe you felt you were held back in the past due to talent constraints, for example. So that's super exciting. then, when you speak to CTOs and CIOs of German companies, the perception is there's still a worry to give away data, a worry to give away insights.

for good regulatory reasons also, you don't want to just connect to an American AI provider and, you know, give away, spill all the beans essentially, right? So that question has to be answered and it has to be answered quick so we don't lose time when we're adopting the technologies. But then I think it's super exciting in a way. It will make us much more competitive if we're fast enough, right? And so that's my big worry. In a way, my big worry for industry is, and I'm talking German industry because

historically, maybe we've been a bit conservative when it comes to adopting new things. So it's such a huge opportunity at the same time, such a huge risk. So, you know, if I take the other side of the coin and we say, it's interesting how much better Chinese hardware have become and machinery have become in the last 10 years and even the cars, you know, if look at the cars, how, you know, how maybe sometimes they're better than our EVs.

Stephan (47:25.806)

The software is also better, right? And the software is maybe a shortcoming of German car industry as well. So the opportunity is huge, right? And I'm not saying it's a simple solution. I think it's multi-layered and you need to go much more in the depth that we could even go in a three hour podcast here as to what that means. But we need to jump on the train to be on the train and you don't want to be the one left in the station when it comes to AI. I think it's...

devastatingly dangerous.

Joern Menninger (47:58.425)

I couldn't agree more who have been talking about that you think there is now a time window open right now, however long it will take where the typical Mittelstand producing companies can actually pick up where they've been left behind by competing software houses wherever they are due to the massive availability of talent, due to cuts in different companies.

plus also the ability utilizing AI to develop software much faster, much better, that is now open. Thinking a little bit ahead of that window, which signal would tell you that Europe is failing to capture the opportunity created by AI-driven software development rather than benefiting from it?

Stephan (48:52.078)

I think there are simple vanity KPIs you could take if you're a manager in your company. And I don't think those KPIs are ultimately decisive factor if you're going to succeed ultimately or not. if you can't show them at all, you'll be in danger. I'll give you one. How much euros or tokens is your average engineer spending in the team on Claude, Code, Entropic?

any of the tools, right? It's a question, for example, I'm asking each of my company's CTOs right now. Like where do you stand there? It doesn't mean necessarily that you're the best, but if you're spending significantly less than everyone else who's at the forefront of this innovation, then maybe you should think what that means. Is that a cultural issue? Is that fundamentally intertwined because of the things that you develop that it doesn't work for you yet? Or...

Are you really late? I think very often, unfortunately, if you're not spending money right now on inference for coding agents for software development right now, then you're probably very late. I'll give you number, just provocatively speaking, you probably should spend right now if you're bleeding edge at least $200 a month or euros a month or so on the engineer. The biggest package on Entropic that you can buy per month.

And you see examples when you go to the bleeding edge of it of people spending thousands, sometimes close to maybe what you would have paid in salary to an engineer in inference. So I'm not saying this is the ultimate outcome or doesn't mean necessarily quality because cultural or architecture shortcomings in software development will just propagate. So there's one interesting study meaning

If you're set up for success already because you have a good engineering culture and good processes, the right targets, the right objectives, the right mindset, you'll probably benefit immensely. The issue is if you don't and you'll just throw AI on it, you'll also run against the wall pretty quick or exponentially faster. And you can see that already in the data. You can measure this. I'll give you another number that's not so much a...

Stephan (51:06.414)

a vanity metric, you can measure that in the customer requests. So you'll see companies having half the amount of customer requests and then those customer requests being answered through agents. And then you see other companies who develop software much faster, but actually the customer requests go up to X customer complaints, meaning you're creating issues way faster. Right. And so that's a bit of an open secret, but yeah, there are some numbers where you can really measure in a way if you're moving into the

right direction right now and you should see it pretty quickly because the impact is so exponential.

Joern Menninger (51:40.761)

But you could also use AI talking about

customer requests here in a different way because what we are doing at Celebrate.io is getting down the number of requests, from non-fitting cooperation partners. That's something where we actively apply the AI because we do have a very long, very strong vetting process for somebody to get in there, especially as partner on our level.

That is something where we found AI to be very, very useful at that stage. Just to be sure you don't do like eight meetings a day, rather five meetings a week and they're better, they're more efficient. That's a close match. That's also something people could think about doing that with AI. But the thing is you would need to know what is your actual customer because they're...

ideal customer profiles out there, especially the B2C versions. People highly educated 25 to 45 college degrees, something like that. And I always tear my hair out because like you're competing for the exact same bunch of people that everybody else does, completely different topic that we will be drifting off a little bit. I'm sorry, I'm known to do that.

Stephan (53:06.092)

No, but if I may respond to that, because it's such an interesting aspect, right? Because you probably noticed, like everyone who's listening to this, that you're starting to get much more emails now. And this also has to do with AI. And obviously we will all have AI looking at our inbox. then how exactly this question, you what's my ICP, what do I really want to get out of my inbox is the important factor, right? Like, what do you want to see? What do you want to respond to? What do you want to engage with?

And, you know, these, these are good questions to ask yourself. think this clarity of thought around outcome, you know, we were talking earlier about the jobs of the future. You know, I will give one advice about, you know, your future job. If you have kids, I think clarity of thought and fundamental thinking, structured thinking, because you're now talking only with the AI and this is, you know, the core competence that to me comes right now on top.

And it's not just mathematical clarity, it's the way you communicate and the way you can discuss ideas with yourself.

Joern Menninger (54:13.114)

Not only that, for example, with my boys, I'm trying to teach them instantly what's really an amateur video and what's really made to be viral. Because if you have like three different camera positions, when a celebrity joins a street artist, that is not spontaneous. Somebody had and paid like three people, maybe even more to hold the camera. They had the right positions.

Stephan (54:25.966)

Mm-hmm.

Stephan (54:34.741)

You

Joern Menninger (54:42.786)

really actually a choreograph plus apparently the street artist and the celebrity already have Bluetooth microphones on them. Yeah, that's something you don't see unless somebody tells you upfront and then yeah, it's obvious all made up. Just a simple idea to think about the mechanics of how something is made. That's also something you've been referring to, right? Clear thought that that is not amateur video. Yeah.

Stephan (55:09.592)

Critical thinking.

Joern Menninger (55:11.352)

even though it looks like it exactly and completely different topic i'm getting off topic again sorry

Stephan (55:18.498)

Yeah, I but it's exciting. think the exciting part about this technology is that it will change all of our lives in a way. I'm fully convinced here. so it scares people. I'm also a little bit worried about it, but in a way, it's such a big opportunity, right? And so we need to jump on it.

Joern Menninger (55:34.685)

Jumping on opportunities, speaking about that, talk a little bit about Ventech here. How are investment decisions made inside of your VC fund? How are they actually made through like a centralized partner conviction or through...

institutionalized decision processes across the firm and how much role does JetGBT play in there because I was talking to another partner and what he told me is in the early days of JetGBT, every time an analyst came back with a great report, he said, okay, now I know what JetGBT thinks, tell me where you would disagree with it.

Stephan (56:22.508)

Yeah. And I think, you know, that's, it really sits on top of our last discussion point, right? So, we use a lot of AI and in the way we're trying to identify companies who work a lot on, you know, thinking through how can we, you know, leverage ourselves to see more companies, see more exciting companies, you know, have good discussions with them, be very prepared when we speak to them. There's so many things you can do better, you know, with, technology, here.

But then the ultimate decision is still based a lot on experience and different viewpoints, right? The Ventek partnership, we didn't talk about that so much, we're, you know, we co-own the management company, we're a distributed team between, would say early thirties and, you know, up to 60, you know, men, women, people from different countries. And it provides usually quite diverse point of view and then the willingness on eye level to discuss ideas, concepts and decisions.

is what makes good decisions in my point of view. What you see now, people come more and more prepared. So people use the AI tools to understand the market. Maybe they didn't fully understand. Sometimes I get asked by maybe people who are not as close to the tech right now, how do I get started? And I just say, just ask the AI, get started. Go engage with it. And I think

You can clearly see people who are engaged, much more prepared to these discussions. And then the discussion is just like you said, you know, what do you really think about it? What do you, what do you, know, which points are actually important? There's, there's in a way when we take decisions in venture capital, we always have too many input factors. It's, it's about a few decisive things and then some risks or maybe shortcomings you, you need to accept with, with conviction that ultimately the outcome will be better because you know, I think it's an open secret, but there's no perfect.

company, there's no perfect investment, there's no perfect human being, right? So, we need to, you know, make decisions under constraints in a way and be very excited about the potential. And so that's a discussion that is still very human in a way. And I'm fully convinced AI will help us more and more there. And AI also is starting to integrate my homework, if I may say so. So, I give you one example. I have an AI that kind of goes through

Stephan (58:46.486)

all the companies in our universe and then it ranks them all. And it's an interesting algorithm and I can tell you it's not 100 % accurate. I would not always agree with the exact ranking it gives to a company. And this is mostly about the idea about how that idea competes with other ideas. But directionally, it's very correct. Meaning if you look at the top 10%, you and I, would probably agree this is a top 10 % idea.

versus the bottom 10%. And so, yeah, I think we're in a way, you know, all being attacked in all our jobs, right? So I'm being slightly defensive, maybe my own job, because I hope I have a minute or more to continue doing this exciting job. Yeah.

Joern Menninger (59:32.732)

minute or more. Yeah, I think that's that's that's really a fair bet here. And basically, I was wondering again and again again, and it's a philosophical question I do have one of you tech people are now out there. If humans are not 100 %

Stephan (59:51.374)

Mm.

Joern Menninger (59:57.775)

Error-proof. How can an AI that we are building be 100 % error-proof? Just for the people to think about that. Yeah.

Stephan (01:00:04.876)

yeah.

I, it's such an interesting, I mean, I'm triggered because it's a discussion I had with one of my partners, Christian, over lunch. And we probably talked about it for 30 minutes. And of course AI is not error-proof. And especially if we talk about the technology that's underlying chat GPT and what we've been mostly talking about here, it's non-deterministic. It's probabilistic. It's, if you ask the AI to explain itself to you, it will tell you I'm guessing the next word. It's, it's basically a very

big metrics, mathematical, know, metrics multiplication that is always thinking given the words I've seen before, what should be the next word I tell you back as a user. And it's incredible that this comes out to us as logically coherent. There's a big discussion and the consensus is usually this is not actually intelligence. It just seems to us like intelligence. It's knowledge being told us, told back to us.

But at the same time, I think there's a discussion if humans are actually so different here. at our lunch, we were talking about bookkeeping and can you really trust Neide to do the bookkeeping? I think that was a very interesting discussion because bookkeeping in a way is rule-based. So it's clearly defined. At the same time, sometimes there is bit of discussion about what is an intangible asset, how we should treat it under certain circumstances. So there's some human judgment in it, but not so much. It's rather rule-based and you need to be very knowledgeable.

And then the discussion point was, you can't make a mistake there, so you can't give it to an AI. And then the discussion was, did a bookkeeper or an auditor ever make a mistake? And I think clearly we can say through history and through all the scandals we've seen in industry, across industry, there's been mistakes, wanted or not wanted. So the question is just, where do we get, in a way, an ad scale if we think about systems and putting many agents after each other?

Stephan (01:02:03.566)

cross-checking them, putting them into the right box, verifying. I think we can get there. And I start to see, you know, we measure this in data science and statistics. There's things like the F1 score about, you create a ground truth in an experiment and then you see the AI, how good is it versus, you know, historically when you've done it with humans. And now we started to see performance that is superhuman in a way across LLMs on unstructured tasks. So to me, the answer is clearly

AI will do our jobs step by step in many aspects and better than us and at scale and faster. And we need to be ready for it and accept that ideally we go on a higher level in terms of how we contribute.

Joern Menninger (01:02:48.028)

That is fascinating because the next question, how we augment our capabilities with AI, because my next question is to ask you at what scale of a portfolio of fund size does the reliance on an individual partner judgment become a liability rather than a strength? And I think that question changed in breadth over the last few years, a month.

Stephan (01:02:52.654)

Yeah.

Stephan (01:03:13.197)

Yeah.

Joern Menninger (01:03:17.038)

quite considerably.

Stephan (01:03:18.774)

Yeah, I mean, maybe your AI can be your partner in a way to challenge you and help you in decision-making, but for the moment, I'm very happy we're team of experienced VC partners. that creates, I think it's also a matter of being an institution and being a reliable partner towards our investors. So to me, for the moment, I'm...

there's of course a trend of solo GPS and it's exciting, but they usually play in a lower AUM, essence under management, fun size they raise game and a little bit higher than you start to see structured teams like us. I think for the longevity of the undertaking and the brand you're building and the company building you're doing, it's much better to be a team. But that's a personal choice.

There, I think we all know different people, different founders, investors and at Ventek we believe strongly and the exchange, the discussion, the exchange of thoughts and even to push each other to make better decisions. And I think that's fantastic. It doesn't necessarily mean it's compromise as a decision. It means it's fine-tuned thinking, reflected thinking. And I think that's very important because

I think we all know this also in our lives. If we actually are all a little bit in LLM, a little bit like an AI, then we tend to think in our own pathways. And so it's important to be challenged and to be reflected the way we do things. And I think an AI can do a part of that job in the future, but there's a question of authority. And if you have an equal partnership that takes decisions together, I think that's maybe a better starting point for today.

Joern Menninger (01:05:13.722)

making better decisions together. That's an interesting question because I was wondering which failure pattern you've observed going through technology cycles when venture firms rely too heavily on teams on consensus rather than decisive convictions. Do they miss steps in technology cycles?

Stephan (01:05:38.338)

I think consensus is, like you said, it's an issue. I think I hope that's super clear. We're not necessarily consensus driven in the decisions we take. think there's a lot of focus on the individual partner, the deal makers conviction. And if I put myself in the challenger position of one of my partners, or they do for me, they very often put very hard questions in front of each other's.

discussion, not necessarily with the intent to stop the deal, but with the intent to think through this question and have a discussion on it. And it's very much in the scope of the dealmaker to say, okay, that's a good question, but for reasons ABC is not relevant here and I want to do that deal anyway. And then in the end, of course, we'll vote together, but it's a matter of trust and mindset and reflecting on is that concern valid and discussing it.

to then go to a good decision. And I think you need to have that second, third order of reflection regarding consensus driven bad decision making to think through it and then also think through what if it works? Because we can always have concerns and you can kill any deal with arguments because there's again, like we said earlier, there's no perfect deal, there's no perfect human, there's no never perfect circumstances for any decision. So you take these decisions

never in a vacuum and to then go with conviction despite concerns. I think that's the essence of GoodVenture Capital.

Joern Menninger (01:07:18.692)

Assets of good or enter capitalist making good decisions. How does the decision architecture scale that investing across fragmented European markets? We've been talking about Germany, DAH, France, Benelux, the Nordics. How can you scale that? Because there are different rules on the ground, but you're still competing more or less in the same market.

Stephan (01:07:29.667)

Yep.

Stephan (01:07:48.738)

Yeah. So I think one big benefit of being local in several markets means you can first of all, really compare. So you see sometimes great ideas across several markets and then make up your mind which one is the best. And as we're all equally incentivized, that creates the incentive to then invest in the best company and maybe not in the one that's closest to you necessarily at all times. That's one. And then two is

In that early stage, many founders will resonate, think numbers are exciting, but they're not the only and by far not the decisive factor in seed and early series A investing because of course you may have numbers, but if you don't fully speak the same language in terms of understanding, ideally you speak the same language, but also ideally you fully understand each other about the expectations, about the plan, about the current status of the business and what's planned for the near and far future.

then you have a real issue, right? And that's why we also have this team that's distributed locally and speaks the local languages. you know, I'm sitting in Paris as the exception, but there's German colleagues in Munich and Berlin. There's a Finnish colleague sitting in Helsinki. There's a Swedish colleague sitting in Stockholm. And it's, I think this is how you can be in a way close to the markets. then

locally take the best and relevant decisions in a way, because you also want to, if you make a deal in Sweden, you want to make it appropriate with the Swedish market. And then you can compare it to the French market or to the German market or to the Finnish market and think which one makes most sense and what makes sense in the global scheme of things. So for us, I think it's a clear advantage, but it's complexity that makes it little harder for others maybe to compete.

Joern Menninger (01:09:35.547)

We've been now talking about the market, about the company. Let's ask you two more questions. I was wondering, as VC investor with some experience, which operating belief about venture investing did you once hold that later proved wrong during technology cycles?

Stephan (01:09:54.296)

Mm-hmm.

this is a good question. think,

Joern Menninger (01:09:59.183)

Ha!

Stephan (01:10:06.158)

It's a job of mistakes, if I may say it. that's the, learn from the things, you know, sometimes I'll give you one example that's maybe not too hard on myself, but sometimes you're too early on an idea that ultimately becomes true. And AI is one of these examples that we've been talking about. There used to be, I have colleagues who are now in their early sixties and they've seen the first wave in the early two thousands.

right in the middle of their career and they saw the excitement and then, you know, the realization that we're not there yet from many perspectives. And when AI started again, there was a moment when for obvious reasons, you're a bit hesitant, right? this discussion about waves that happen and that continue to happen in technology trends is one that is always moving every VC firm and especially the ones that are multi-generational in terms of funds and team.

because you always have one colleague, one partner that has seen it before. And that discussion with, see it now and I have new data and now it's different is a very interesting one. Sometimes it's not different and you still shouldn't do that investment or sometimes it's the same and you should do that investment because last time it was great. Or maybe sometimes it worked last time well, but this time it will not because of different reasons. And so these are usually good.

good discussions based on experience and new data and then being open-minded while reflecting truly. And that sounds too complex. And I think there are some other philosophies out there in the market where some people say you should just jump on the fast growing exciting thing. That's the business. at least for Ventech, we're trying to use that experience that we've generated to leverage it in a way in decision-making.

Joern Menninger (01:11:55.964)

Hmm. see. Oh, I have an ongoing bet with a not named colleague who's also running some media outlet here in Europe. we have, we talk to each other frequently and we have the Nelson bet. We don't bet about money, we bet about who gets to say to the other, ha ha. That's also something. Yeah.

Stephan (01:12:04.045)

Okay.

Stephan (01:12:19.566)

That's fair. It's more affordable.

Joern Menninger (01:12:25.047)

Exactly. So that so much about technology cycles. Nobody has the 100 % truth. And the closer you get to reality, the clearer the picture becomes. But the further away you're from reality, when this really hits reality, when it's really implemented all over, there's so much afloat that you cannot make a really precise technology forecast and everybody who tells you so is a liar.

Yeah, or he took one side of a Nelson bed. Last question for you. What specific evidence forced you to replace that belief with the rule you apply today?

Stephan (01:12:58.67)

Fair enough.

Stephan (01:13:13.0)

like a rule we apply today, which...

Joern Menninger (01:13:15.161)

Yeah, you had to throw out a belief in the past, last question. And now what evidence force you to do that belief? What specific evidence force you to place that belief with the rule you apply today?

Stephan (01:13:33.634)

Yeah. Yeah. think several, several we had, um, in the discussion. mean, there's one, one around AI. was for the longest part, uh, during last year, still have the opinion that, Hey, it's, you know, it's an assistant. It's not fully, fully, um, uh, agentic. And that, changed, um, in, in, in December when I've seen the, you know, when I've actually developed software overnight and came out good in the morning. So, so I think I changed my, um,

And it was dogmatic before, be frank. Dogmatically, was thinking, hey, this is, technologically speaking, a thing that guesses the next word. It will not be able to fully comprehend the complexity of a whole workflow of something that's human, that the average intelligent human can complete. And so that changed. And I think that, in a way, changed the way about how exponential things can change also now.

across industries in my point of view. And we talked a lot about that today. Another maybe more pragmatic thing, you know, as a team, we used to love asset light businesses. And I still think we prefer those generally speaking. But with that new complexity around what it means to build a mode, the, you know, what people call deep tech becomes

interesting also from a different lens, meaning you want to think about the multi-factor combination of assets and things you create. I give an example of a portfolio company called Okapi Orbits, the space traffic management. So we're trying to combine an asset light business in space technology that helps coordinate the traffic between satellites. And that's a super interesting business because it has some service elements, has some software elements.

the aspect of AIs in their geopolitical elements are in there. so in a way, it sounds like risk from many perspective, but also that's an opportunity, right? I mean, we should never confuse the two, but historically, if you would have asked me five years ago, is Ventec excited about space? I would have said, yeah, I don't think we're gonna put money into a business that needs hundreds of millions to build rockets, right? And we kind of...

Stephan (01:15:54.67)

opened our minds as a team to consider more complex use cases, ideally with a twist. And I think that's a good example.

Joern Menninger (01:16:04.891)

And interesting as an indicator from macroeconomic point of view across your portfolio companies today is your challenge posture, net hiring, net stable or net selective.

Stephan (01:16:19.182)

And that's selective very clearly. many discussions with founders are around how can we be ready to move fast now into this new reality? And that creates a discussion about how can I have the three people in the team that leverage the new technology to really bring us forward?

Yeah. So I think it's a flight to quality. if you really think about it and you think about an agentic workflow or process, you need one person and the team to really model it out. You don't need 50. So you can train everyone to be great at using AI or one person actually gets it and does it for everyone. So there's a real lever in having those key players in your team that bring you forward right now. Like, do you have a good question that I asked myself? Do you have the right?

know, internal engineering leadership to, you know, show others the way, which is, you know, complicated, but, but again, you don't need to show everyone, but you need someone to be able to explain it to the right people who may be deepest in the most valuable processes that you have in your company.

Joern Menninger (01:17:33.194)

Very good closing words. Stefan, thank you very much. We're talking now for almost an hour twenty. Thank you for sticking around so long. It was a pleasure talking to you.

Stephan (01:17:44.334)

Thank you, Jern. Great pleasure.

Joern Menninger (01:17:47.041)

If AI is compressing software development, transforming enterprise workflows and redefining the economics of SaaS, which single strategic decision should founders and investors make differently starting today?

---

Related Episodes on Startuprad.io

• How To Scale Globally — Another leadership and scaling episode

• Browse all Startuprad.io episodes — Topic hub: Leadership and founder interviews

---

About the Host

This interview was conducted by Jörn "Joe" Menninger of Startuprad.io.

---