For anyone building in AI, the Merantix AI Campus is a one-stop-shop for idea sharing, networking, and collaboration. So when I was introduced to Sean Escola, an American biotech founder and computational neuroscientist who recently moved to Berlin, I said he should come hang out for a few weeks here as he pondered his next venture in the “Neuro AI” space.

Sean joined us recently at the AI Campus to bounce ideas off of our Merantix Capital team and connect with the various parts of our Merantix Group — from AI solutions business Merantix Momentum to our Merantix Capital portfolio companies, like fellow biotech founder Greg Vladimer, as well as other residents here on campus.

This is a no-strings-attached offer we extend to seasoned serial entrepreneurs looking to build an AI venture in their domain. We are happy to let you tap into our ecosystem to help you on your way. You never know, maybe we’ll end up working on something new together via our venture studio at Merantix Capital, where we build and invest in category-defining application layer AI companies.

I recently sat down with Sean about his background and to ask him about his time working from the AI Campus. Our edited conversation is below.

Can you tell me a little bit about your professional journey?

I have somewhat of an unconventional background. I’m a clinical psychiatrist, and I led a computational neuroscience lab as an assistant professor at Columbia. I founded a biotech-meets-machine-learning company in San Francisco in 2017 called Herophilus, and we ultimately hired an external management team, although I stayed involved and was on the board. We had a business relationship with a company called Genentech for co-development for some drugs, but they actually got excited about our technology platform and decided in 2023 that they would acquire it.

The sale of the company was a good transition moment for my family. My wife is German and we wanted to be back in Germany. I moved to Berlin and tapped into my investor network, and I heard really good things about Merantix and the AI Campus. Then you and I had lunch and you suggested I come hang out on campus.

What are you looking to build next?

My last company was squarely in biotech, using a lot of machine learning applications to try to do data analysis of these large scale data sets that we were collecting for the discovery of new drugs. It’s modern techniques, but still a standard playbook.

A direction I’m now exploring, although it has an overlapping set of skills, is how we can use ideas about how the brain works as well as datasets that are collected from human or animal brains to train AI models in order to endow them with more human-like cognitive capabilities.

In other words, rather than using AI for neuroscience, it’s using neuroscience to build better AI.

What did interacting with the Merantix Group do for you?

I got a sense that you and Rasmus are really excited and helpful to founders who are familiar with one system and now thinking about doing something in a different system. For me, I understand the startup scene in the US but needed to understand more about what company formation looks like here.

There’s a utility to partnering with a team like Merantix who are very intimately familiar with the intricacies of building in Europe, and also at the same time recognize how to maintain efficiencies from the American system.

A major thing on my mind as a potential founder is access to good talent, and that’s really something that Merantix offers. You can really see it on campus, whether that’s through events or for instance the clear technical engineering competence at Momentum — and how they work with your portfolio companies.

You recently had a second child (congratulations!), but where are you at currently when it comes to building your next company?

We’ll have to see what happens in the next couple of months once I’m able to put more effort into trying to build something, but I would be very excited about potential partnering with the Merantix Capital team. From our time together, I could see firsthand how your relationship with your portfolio founders is supportive and quite collaborative, when it can sometimes be somewhat adversarial with investors, because your goals are often different.

Thanks for the chat Sean! You are certainly welcome on the AI Campus anytime, and we’re excited to see — and maybe be involved in — whatever you build next!

Are you a serial founder looking to build a category-defining venture on the application layer of AI? Get in touch, and come immerse yourself on our AI Campus like Sean did.

If you're an ambitious AI founder in 2025, the options have never been more varied or more confusing.

As someone who has been building and investing in AI since 2016 (and an entrepreneur myself), I spend a lot of time speaking with founders who are looking for their next move. They’re typically repeat founders with industry domain expertise interested in founding an AI-native venture and looking for the right partner to help get them there.

Sometimes we discuss Merantix Capital’s “founder-in-residence” program, where a founder explores multiple AI business cases over four to eight weeks with us, spends the next 2-4 months on validation ( building commercial traction), and if we both build conviction, we invest €1M for 15% preferred shares and 10% common shares (institutional co-founder equity).

To be quite frank, with multi-stage VCs pouring capital into pre-seed stage and “pre-idea” companies, there is mounting pressure on the studio model that makes it seem outdated. The reality is more nuanced. For the right type of founder, the venture studio model is not only viable, but the smartest path.

Caveat: I am biased. At Merantix Capital, our venture studio is core to our identity, through which we’ve helped create more than 10 amazing companies like Cambrium, Deltia, Ovom Care, and Libra. We also make investments in companies that were founded outside of our studio approach (about 50% of our latest fund).

Take it from me, over 20 years as a founder and eight years as an investor, I’ve seen trends in the early-stage ecosystem come and go, but the really successful founders have always recognized that how you start matters just as much as what you start.

Current market dynamics are dizzying

The early-stage AI market today is flooded with capital. Major VCs (especially those with large growth or crossover funds) have aggressively moved downstream, writing large checks at the earliest stages of company formation. In many cases, they're backing operators or repeat founders without a concrete idea, betting that these individuals will figure it out.

Why? In part, it’s a hedge. Late-stage investing has become more challenging, but the excitement around AI has created a sense of urgency to secure early access to promising talent. The big funds can easily afford to write off early bets as the cost of staying in the game.

This capital abundance brings its own issues. With more money chasing a relatively static pool of talent, founders are under pressure to scale fast and meet unrealistic expectations, often before they’ve validated a problem, let alone a tangible product or GTM.

What do venture studios do differently?

This is where the venture studio model (and I would argue our venture studio!) stands apart. Instead of betting on potential, we are a partner in the core foundation of a successful venture: early validation, commercial traction, capital efficiency, and team quality.

It’s true, we write smaller checks than a mega-fund, and our founders may take slightly more dilution upfront. We won’t argue otherwise. But we offer something large funds can’t: deep support at the earliest stages from an AI ecosystem that includes corporate design partners, domain experts, and an in-house talent pipeline centered at our AI Campus in Berlin and the AI Hub in London.

Many of our portfolio companies have paying clients at or soon after incorporation—often secured through our ecosystem. Four of our last five studio startups, for example, reached €500K in revenue within the first six months. Three also went from pre-seed to seed in under 10 months.

That level of early traction simply isn’t possible without trust and access, two things the Merantix platform offers in spades.

Efficiency is the edge

Capital efficiency is more important than ever these days. Startups without early validation often overbuild. That means bigger teams and more burn just to get in front of customers. Our approach flips that script. With early access to design partners and strategic customers, our startups can iterate faster, avoid premature scaling, and stretch €1M in ways others need €3-5M.

Other investors might want a fully-formed team, but we’re comfortable starting with a strong founder and co-building the right team from there (including co-founder matching through our network). That flexibility means we can bring in the right people at the right time without overstaffing.

When we put out a call for specific profiles on these teams, we receive hundreds of applications within days. That talent density is a force multiplier, and one of the key reasons our companies punch above their weight.

Who is the venture studio model right for?

The plain truth is that the studio model isn’t for everyone. It’s best suited to founders who:

• Are repeat entrepreneurs and understand the complexity of building from scratch

• Have deep operational experience and want to lead but not go it alone

• Bring technical expertise but need commercial or regulatory co-founders in complex sectors

The venture studio model is, in short, best for founders who are long-term thinkers. Those are usually the best founders! They understand that speed, talent, and traction matter more than retaining an extra few points of equity early on. Not to mention that higher traction in the beginning typically means less dilution in later rounds for founders.

Moving beyond the buzz

Why did I feel the need to write this? I don’t want to sound defensive or over-advertise our venture model. It isn’t for everyone and, like I said, half of our fund is geared towards “non-incubations” too.

But it also has bugged me that “venture studio” has become a dirty word in VC, because I think it’s a great model that produces huge results.

There’s a lot of noise in the AI world right now. Every startup is "agentic” and every founder is a visionary. But building enduring companies still requires the hard stuff: talking to users, recruiting great people, and making products people actually want.

That’s what we help founders do. Not through hand-holding, but through actual partnerships and a trusted brand. Our value is acceleration through talent and traction.

Last month, Merantix Capital had the pleasure of hosting a great AI pitch competition at the annual START Summit in St. Gallen, Switzerland.

We heard from six great early-stage AI ventures, but one really stood out to our panel of judges: LogicStar AI, which is building autonomous AI agents that solve one of the biggest bottlenecks in software development (the manual and slow task of maintaining and debugging production code). The company recently announced its €2.9 million in pre-seed funding round, led by Northzone.

I had the opportunity to sit down with LogicStar AI CEO and co-founder Boris Paskalev after his successful pitch to talk about how he’s creating “Silicon developers” and the future of agentic AI. A seasoned founder, Paskalev’s last venture DeepCode was acquired by Snyk.

Adrian: Congrats on winning our pitch competition. Agentic AI is obviously a hot area right now. Tell us a bit about your core vision and how you define an agent.

Boris: Our internal definition of an agent is an entity, ultimately a “Silicon developer,” that fully autonomously produces tasks and calls large language models. We distinguish between agents and tools. Agents can call other agents or tools and work together to produce what we want. Ultimately, it’s about freeing engineers from the laborious tasks of maintaining code—investigating, reproducing, and fixing issues with high-fidelity, production-ready code.

So it’s about allowing engineers to focus on creativity?

Exactly. A lot of agentic AI tools talk about freeing up engineers for more meaningful work, and I agree. It’s already quite autonomous. The question is: what percentage of the backlog can be fixed autonomously?

That kind of automation frees up a significant number of engineers in larger organizations to focus on building new things. As our “junior Silicon developer” gains experience, it’ll take on more things like fixing documentation, refactoring code, and validating human work. But the biggest value today is automating a part of the backlog without engineers lifting a finger and doing it much faster.

How much faster?

From what we’ve heard consistently, it takes engineers between one hour and a day to fix a bug, on average maybe four hours. We can fix it in seconds. That’s the real value. It’s about building self-healing applications and fixing issues before customers even notice.

Do you see this ever replacing developers?

No, we’re not claiming we’ll replace developers. Complicated stuff will still need a human. But our goal is to take away a large portion of the work they don’t want to do. One of our key differentiators is that we’re not just building a co-development tool.

Today, most tools that are called “agents” expect constant human input. You prompt, you review, you discard or tweak 80–90% of the output. That’s just not scalable. The return on investment isn’t there. You spend more time reviewing than just fixing it yourself.

Are there particular industries where LogicStar AI works best?

Honestly, it’s across the board. We’ve seen it work from big banks to small startups, even open-source libraries. The main limitation right now is language. We only support Python. JavaScript and TypeScript are next.

The real challenge with adding a new language isn’t technical. It’s building the benchmarks to test the agents properly and ensure they improve over time. So when talking to folks on our waitlist or potential design partners, the main question is: “Do you have a Python app?

You’ve founded before. What's your approach to company building, based on your past experience?

Talk to customers as early as possible. Right now, we’re onboarding design partners and building out the initial integration. We want them to shape the product. The focus is building a strong product, early go-to-market and visuals that show the value clearly before we scale up sales and support teams.

What’s the benefit of building in Europe?

Honestly, it’s the deep tech community. ETH Zurich and INSAIT have incredible researchers. My co-founders are top experts in program analysis and LLMs. We’re not lacking in talent at all.

We’re also building out frontend teams in Bulgaria, where we have strong ties and can recruit great engineers. But customer support and sales will likely be U.S.-based, since that’s where a lot of our early adopters are.

What’s your LLM strategy?

We’re model agnostic. We use Claude, GPT-4 Mini, Gemini 2.5. Whichever works best for a given task. Different agents perform better with different models.

We’ve already got 12 agents working across different steps, and each uses a different model depending on what’s needed. That flexibility is key. As models improve, our product improves. We also care a lot about cost-efficiency. We want to build a sustainable business, not one that loses money hoping to scale later.

How did winning the pitch competition at START help you?

It gave us visibility. But really, it reinforced our mission: to build the first silicon maintenance engineer out there that frees up engineers.

Thanks very much, Boris! Good luck to you and your team, and congratulations again!

In recent conversations with founders and investors alike, one theme keeps resurfacing: in today's AI landscape, switching costs are dangerously low. As someone who spends every day thinking about AI innovation and its long-term trajectories, I believe this is one of the most underappreciated risks (and biggest opportunities) in building enduring companies.

In venture investing, you're betting on companies over 5-10+ years. Traction today is meaningless if users can easily leave tomorrow. If a startup loses momentum because customers switch, regaining it is extremely difficult. Commoditization pressure is real, especially as many foundation models and AI tools increasingly resemble commodities (I highly recommend this article by my good friend Rob Toews.)

In this substack I wanted to outline Merantix Capital’s approach to dealing with switching costs, why they matter more than ever, how we believe startups can build defensibility, and how emerging technologies like continual learning might change the game.

In short, the goal of an AI founder should be to create switching costs so high that replacing their product feels like replacing a well-trained, high-performing employee -in other words, very difficult.

Why Switching Costs Are So Low in AI

• APIs and models are interchangeable. Companies can swap model providers almost overnight. In many cases, swapping from one LLM provider to another is as simple as updating an API endpoint. I see this in my own day to day: I switch tools as soon as a slightly better, cheaper, or faster option appears.

• The speed of innovation is relentless. New models and techniques are appearing every few months. Even companies with strong technical teams can be outpaced if they rely purely on product speed without building underlying defensibility.

• B2B switching happens faster than consumer switching. Business users are rational, technical, and highly cost-driven. Unlike consumer behavior, which is often sticky due to habits, brand perception, or inertia, business decisions are made on ROI, performance, and cost…

• Commoditization is accelerating. Models today largely deliver similar outputs, and continual learning is not yet widely adopted. This makes it very easy for users to move from one provider to another without any real switching pain.

• Low integration depth. Many AI applications layer thinly on top of models without deep integrations into broader systems (e.g., ERP, EHR), making it easier for users to switch when something marginally better arrives.

How to build and evaluate high switching costs

To build truly durable companies in AI, founders must think deliberately about switching costs from day one.

We find it useful to think of switching as compounding layers.

The deeper you move a customer down these layer, the harder it becomes for users to switch away:

Each layer deeper on this chart compounds the stickiness of your product.

The Emerging Frontier: Memory, Continual Learning, and Durable Moats

Today’s LLMs are powerful but fundamentally static: they learn during training and then stop adapting. True human-level AI requires models that can learn continuously throughout their deployment lifecycle.

Several workaround techniques exist today:

• Fine-tuning: Helps update models post-deployment, but is batch-based, expensive, and limited.

• Retrieval-augmented generation (RAG): External memory approaches are useful but face scaling and latency challenges.

• In-context learning: Allows models to "simulate" adaptation during inference but without true weight updates.

However, none of these approaches solve the deeper problem of building durable, evolving AI models.

Continual learning, where models update their own weights on the fly without retraining from scratch, represents a profound shift. In his article, Toews argues that continual learning will, “unleash AI’s full potential to power hyperpersonalized and hypersticky AI products.”

In other words, imagine an AI legal assistant that becomes better at drafting documents in your firm's style over time. Or a recruitment AI that learns your exact criteria for screening candidates without explicit reprogramming. Models would develop deep, personalized knowledge that is hard to replicate elsewhere, creating switching costs akin to replacing a well-trained, high-performing employee.

Companies like Writer and Sakana AI are already experimenting with architectures that enable this self-evolving behavior. These breakthroughs hint at the future direction of defensibility in AI products.

How Founders Should Think About Switching Costs from Day One

Switching costs aren’t a "Series B problem." They must be part of the early design and go-to-market strategy:

• Move fast but design for depth: Speed to market is crucial, but not at the expense of defensibility. Focus on integrations, memory, and workflow entrenchment.

• Be intentional about data: Find ways to own, structure, or personalize user data.

• Audit your defensibility constantly: Revisit the checklist regularly as the product evolves.

• Think ahead to continual learning: Even if not feasible today, position your architecture for future personalization and adaptation.

Conclusion

In AI, defensibility is no longer just about building technically superior products. It’s about durability: how resilient your customer base is in the face of rapid technological change.

That's why I prioritize investments in startups that integrate deeply into critical workflows, prioritize user trust, create strong switching costs, and build toward the future of dynamically learning AI systems.

If you're building a company that understands this challenge and is actively working on durable moats in AI, I'd love to hear from you.

In the fast-paced world of AI, staying ahead is tough. We see new models and companies emerging every day, making it harder to maintain a lasting advantage. But there’s one strategy I’ve seen gaining traction lately that I think is underutilized—and it’s incredibly effective: the Trojan Horse business model. I presented a longer version of this at this year’s PakCon.

Here’s how it works.

The Challenge: Staying Ahead in AI

When you’re building an AI company, there’s a ton of hype around data and open-source AI. Some of the traditional advantages still apply—having a great dataset gives you a head start, and proprietary AI models can offer a temporary edge. But the reality is, open source is democratizing the space fast, and your edge today can disappear overnight when someone else comes along with a bigger budget or better AI. The “data is the new oil” mentality? That’s from 10 years ago, and it’s lost its shine.

The Trojan Horse Strategy: Start Small, Disrupt Big

Here’s the real opportunity: instead of trying to build something massive from day one, like a new ERP, EHR, or CRM system, use the Trojan Horse strategy. Build an AI-powered productivity tool that solves a specific problem for companies. Think of something that integrates with the systems they already use—optimize their workflows, streamline processes, or make their data work smarter for them.

Now, here’s where it gets interesting: as your tool becomes more ingrained in their day-to-day operations, it starts to collect data, add value, and ultimately become indispensable. From there, you can expand its capabilities, make it the go-to interface for their systems, and over time, you’re positioned to replace the legacy system entirely.

Case Study: Ficus Health

One of the best examples of the Trojan Horse strategy in action is Ficus Health, a startup in the healthcare space. They realized that up to 40% of doctors’ time is wasted on documentation—time that could be spent treating patients. Rather than trying to disrupt the entire Electronic Health Record (EHR) system overnight, Ficus Health built an AI tool that listens in on doctor-patient conversations, automatically generating medical reports and sending them to patients if desired.

This simple but powerful tool solves an immediate pain point—documentation. By integrating seamlessly with existing systems like the EHR and Radiology databases, Ficus Health became an essential part of a hospital’s workflow. Over time, they’ve expanded into new areas, like diagnostics, treatment recommendations, and even clinical trial recruitment, making their tool even more valuable.

Here’s the real kicker: as Ficus Health collects more data and integrates deeper into these hospital systems, they are now in a position to eventually replace the legacy EHR systems that have dominated the industry for years. It’s the perfect example of how you can start small, solve a real problem, and ultimately grow into something much bigger.

Why It Works

Some of the most valuable companies today, like SAP, Salesforce, and Epic, didn’t get there overnight. They built systems-of-records that are now irreplaceable. All company data flows through them, and every business interaction relies on these systems. It’s hard to sell a brand-new ERP system to a company that’s been using the same one for 10 years—but a productivity tool that sits on top of the system and makes life easier? That’s a much easier sell.

This strategy allows you to start small, build trust, and gradually expand until your AI solution is the core of their operations.

The Big Opportunity

For many industries, the Trojan Horse strategy is a huge untapped opportunity. And let me tell you—this isn’t a secret. Companies like SAP and Salesforce are already investing heavily in AI because they see the writing on the wall. They know there’s a risk that startups using this approach will eventually disrupt them.

So, whether you’re a founder or an investor, think about this: the real value in AI isn’t just in building bigger models or having more data. It’s about embedding AI in ways that solve real business problems and scaling from there.

There’s no better time to build. And if you’re thinking about launching a venture using this strategy, reach out. I’d love to hear what you’re working on.

Richin Kabra is one of the newest founders to join the Merantix Capital portfolio, having recently teamed up with our founder-in-residence Anouk Moll on a new London-based venture at the intersection of AI and capital markets. 

Their fintech company is currently in stealth mode — but we couldn’t be more excited about what they’re working on. And we couldn’t be more excited about Richin, who joined as co-founder and CTO this summer after working as a senior data scientist at Meta for advertising auction science and previously as a senior quantitative trader at Citadel Securities. 

I recently sat down with Richin for a chat about his learnings from his time on Wall Street, what AI brings to modern finance, and the challenges of building in fintech right now. Our edited conversation is below. 

You began your career on Wall Street, eventually working on the ETF business at Citadel Securities, the American market making giant. What did you see there and did it make you think about becoming a fintech entrepreneur yourself?

Almost 20% of all U.S. equity trades are executed through Citadel Securities, yet they only had a small presence in ETFs.

It was almost like starting a new venture within a well-established giant. While I was a quant trader running the emerging market ETF desk, I also took on roles as a business development agent and an exchange connectivity agent. I was in discussions with the Johannesburg Stock Exchange to build connectivity there, and I was also reaching out to the Indonesian Stock Exchange, where connectivity hadn’t yet been established. We had to consider the amount of risk capital we wanted to deploy in each of these exchanges, essentially thinking about product-market fit.

This approach translates well to entrepreneurship: you have a roadmap, a sense of product fit, and you’re continuously testing adjustments in capital deployment to maximize returns and gains. That made me think - if I can do this for a behemoth like Citadel Securities, I could do it for something smaller as well, something I could call my own and engage with more deeply as a product.

What opportunities did you see at the intersection of AI and finance? 

There is a backlog of data processes and harvesting that is done very manually, including data entry and numerous mid-office risk compliance factors. All of these have a structural element to them. Anytime there is an input with structured data fields, I believe AI can handle that much faster than human intelligence, simply because it’s a repetitive activity involving structured data. Improved operational processes are definitely one pillar.

The second pillar, I would say, is risk modeling. We've repeatedly seen instances where correlations in financial assets have been mis-modeled. This was a major issue in the 2008 crisis, where the U.S. mortgage crisis stemmed from a miscalculation in the correlation across mortgages.

Do you think AI could have helped avoid that financial crisis?

I think improved models can surpass human intelligence. In other words, AI can bring an incremental improvement to risk modeling, which currently relies on two-dimensional models with standard deviation and time as the two axes, merely predicting changes in standard deviation.

We’re not just talking about changes in risk. We’re asking: if this were the scenario, what would the other influences be? We’re considering factors like housing credit, employment credit, and other opportunities in the landscape. How would that impact employment opportunities? How would it affect inflation? How would the Fed adjust interest rates in response?

Think of it as a game of chess. Traditional models take a single step - if a player makes a move, what would you do next? AI, however, can go much further, predicting up to 13 steps ahead to anticipate moves and counter-moves in response to an opponent.

After Wall Street you moved to Meta working on advertising. That seems like a pretty big jump.

If you think of ads as an asset class, they’re quite similar to traditional asset classes. You have a buyer, a seller, and an exchange. In traditional asset classes, the buyer might be someone in the retail world who wants to buy Apple stock. The seller is typically a market maker who’s willing to sell that Apple stock, and the exchange is usually the London Stock Exchange or the New York Stock Exchange.

The ads market works in much the same way. The buyer is the advertiser, and the seller is your screen time. The exchange is typically run by ad networks like Google or Meta.

What’s particularly interesting about Meta is that when you control the pricing for the auction system that connects buyers and sellers in an ad auction, you're heavily involved in regulation to ensure pricing is clear and justifiable to users, explaining why they’re seeing certain ads. Additionally, you aim to make the system as efficient as possible in terms of targeting.

We used user data to predict the best interactions for users — what they wanted to see. And for advertisers — what they wanted to show users. It’s a double optimization process. We would predict the best return for advertisers and the best return for users, then combine these predictions in a weighted mechanism to show users the most relevant ads. This way, users don’t dislike the platform, and advertisers get the best value for their investment.

What are the biggest technical challenges right now within building in fintech and AI? 

I can probably answer that in three pillars.

The first would be data protection and privacy. If you’re looking into consumer credit lending, you’ll want to consider data privacy, especially when it comes to protected classes—housing, employment, and credit—and avoid using those protected classes. I also believe gender should be considered a protected class. It's crucial not to use that data to build proprietary models for individual clients.

The second pillar would be explaining models. When you run AI in the background, everything often becomes a black-box structure. The outcomes are very different from the inputs because there are significant operations happening in between.

The factors that contribute to these operations aren’t always explainable because there aren’t always human-understandable factors. For example, when assigning a loan or credit to someone for purchasing a house, in traditional models, you’d look at employment, wages, and any other mortgages held by the person. But in an AI model, these factors are weighted together in a complex way.

This leads to a convoluted mesh where it becomes difficult to describe what’s actually being used to an operator, regulator, or compliance officer.

The third piece I’d touch on is initiative change. FinTech has been around for a much shorter time than traditional finance or tech industries. When it comes to adjusting to new models and transitioning to different infrastructure, you need to upskill people. You also need to convince regulators that these new models are solid. When risks and balances aren’t in place, failures can occur, and regulators often focus on these single failure events as points to revisit, which can slow progress. 

How have you found working with Merantix Capital and being part of the larger Merantix community?

I just came out of a coffee chat on model implementation, and everyone was super friendly and welcoming to a new person. It’s been incredible. Having a campus environment feels very wholesome. It brings you back to a time when you were in college with everything around you. Accessibility really boosts productivity.

Dr. Lynae Brayboy has one of the most interesting stories of all the great founders in the Merantix Capital portfolio. A US-born physician and scientist and lifelong tech enthusiast, Lynae was formerly the chief medical officer of Clue, the Berlin-based period-tracking app.

Last year, she joined our venture Ovom Care, which combines modern in-person fertility care with advanced AI-based technology, as co-founder and chief medical officer.

The company has been on a roll lately, raising its €4.8 million seed round in the spring to further its mission to make reproductive care more affordable and successful. The company has a clinic in London and other location coming soon to Portugal, already providing tailored, personal, and tech-enabled care to more than 100 patients.

Lynae and I recently sat down at the AI Campus in Berlin for a chat about Ovom, how technology can enhance the patient experience, and the challenges and opportunities of the European healthcare landscape. Our edited conversation is below.

Adrian: Tell me a bit about Ovom Care and how you came to join.

Lynae: Ovom Care is a hybrid IVF clinic, and we leverage AI tools to improve IVF outcomes. We hope that our AI tools improve not only the patient experience, but accessibility and success rates, meaning live birth rates. 

At Clue, which is a period tracker, it’s amazing that women can actually keep track of what's happening in their cycles. This is usually the first indication that something has gone awry with your reproductive capacity and success. And so I was at Clue for two and a half years, and, you know, things happened in that time. The economy changed. Capital became very difficult to acquire. It was a very tight time, and unfortunately my team was let go. When one of my team members was looking for a job, I wanted to help her, so I was on LinkedIn more. Otherwise, I'm never on LinkedIn. I came across Felicia, who was a founder-in-residence here at Merantix, and she said that she was starting a stealth fertility tech startup.

She was looking for a co-founder and chief medical officer. So I reached out to her, and the rest is history.

Was it a topic that you had thought about before at Clue? The idea that there might be a way to kind of bring these AI tools into the fertility space.

What was missing from a lot of period tracking apps is, “What do you do now?” You found that you have this pattern. You found that you have this data. But what do you do with it? Usually people bring it to their physician, and their physician goes, “OK, nice, great.” They may be able to act on the data. They may not. The whole point of Ovom is that we want to leverage the data to understand not only about you sitting in front of us, but other women like you. When you think about women's health care, it's lacking data. It's lacking research. It's lacking innovation.

Having the ability to use things like the Oura Ring, like trackers, like LH strips and put them in a cloud and understand what they actually provide. How does that track reproductive outcomes when they actually do get care? No one's done that before, and that's what we're seeking to do at Ovom.

What do you see as the big advantage that AI and machine learning bring to the real-life clinic experience? 

When we think about how we manage people in our clinical guidelines, that comes from studies that might have — woo-hoo! — 1,000 people in them.

What if those 1,000 people had 3 million data points, which you can get from cycle tracking. You can get from someone going through an IVF cycle. This is what we're missing in medicine. We're kind of blind to it because we haven't had the tools. What we want to do at Ovom Care is really integrate that, not only into visualizing your data, but how that changes how we manage the patient. 

Can you walk us through the patient experience at Ovom and how technology comes into play? 

It's first important to say what's the traditional experience. In a traditional IVF clinic, maybe you see an ad in the subway or online or in the newspaper, you hear it on Spotify, and then you give the office a call and you give them all your information. You book your appointment, and then you drive in or you take the subway in, and then you sit there and someone talks to you and you pay them. You decide whether you want to go on with care.

It's a lot of email and a lot of telephone calls. Maybe the nurse calls and you can't pick up, and the nurse is saying, “Make sure you do this and do that.”

We wanted to turn the experience on its head, because that's not very patient-centered. If I have to chase you to get information about my care, that's difficult. Instead, we wanted to allow patients to not have their lives disrupted so that they could actually start their care experience in the comfort of their home on their couch and only come in for what is necessary.

Today at Ovom, people will go on our website or they'll see an ad, they'll click a link, and then they can automatically have a free introductory educational call with me, a clinician, not with a salesperson, Afterwards, we get them onto our app, which is free, no obligation.

They can upload all their records. You have no idea how it works in a traditional clinic where you have to come in with your stack of papers and show people, “This was my lab from last week, and this was my lab from two years ago.” It's a mess.

Why not take that data into consideration and be able to visualize it and say, “Oh wow, your periods actually have been shortening. Oh, wow, your flow has been getting lighter. Oh, wow, this is a new symptom for you that you might be experiencing.”

This is how we now can intervene earlier as opposed to later, when you've already now perhaps gone into menopause prematurely or maybe when you already have a more advanced disease than when you potentially could have come in earlier. 

How do you marry being in charge of a high-growth, ambitious company with delivering the level of care that I know that you care so deeply about? 

It's not that hard to teach a clinician who's interested how to use tech. People who've been using maybe antiquated ways in their other jobs have now come onto the team, and they're really excited to use tech to reach the patients and to have a better experience.

Being a tech founder, it's what I've always done. I've always been on my phone. I've always been on my computer. But now it's for a higher goal, a higher mission, and also to drive R&D so that we can offer better quality care. Not tomorrow, but maybe five years from now, based on the AI tools and also the innovations that Ovom will bring forth.

How do you view the European healthcare landscape versus the US?

I think when you say European versus American, you're saying huge versus huge, right? Even in different states, healthcare is administered in different ways in the United States. Some things that are legal across the border are legal in another state.

It's the same in the EU, unfortunately. When we talk about reproductive healthcare, it's a minefield for patients because what is allowed in, let's say, the Czech Republic is not allowed in Sweden, which is not allowed in Spain. 

What we want to do at Ovom is allow people to have the families they want on the terms they want, sidestepping all of the things that the government, the patriarchy, and society has imposed on the laws of the land. I mean, we're having a battle right now. The right to IVF in the United States is not being protected, despite whatever a politician might say before a debate about protecting IVF.

We've already eroded our basic reproductive rights. I would say in the EU, you can still have access to contraception so that you can have a child when you want. In the EU, some people have access to egg freezing.

That's not universal in the United States. In the EU, there is some access to IVF, although it's very restrictive and very limited. In the U.S., it's only about half the states that have any type of mandate.

There are some differences, but the underlying thing is that the government is not necessarily thinking about women's rights. They're not thinking about reproductive justice. They're not thinking of LGBTQIA communities when these laws are written. It makes it very exclusionary, very discriminatory for the majority of people who need IVF. 

How do you see AI and technology can lead to more democratization of fertility care? 

One, we can help people have earlier education and access. We should all know some fundamental education about how reproduction works so that if you are now 30 years old seeing this information for free, you don't have to wait until you're 40 to go, “Ohh wow, I didn't know. No one told me.”  And AI can help with that because I can't say it in every language. I can't be everywhere. I can't be at every conference.

The other thing is that we can look at pricing models that help people in their locale. Right now we have a clinic that's already open and functional in London, and that has one price point in London. But in Portugal, where we're opening soon, we have a very different price point. People can contrast and compare online without leaving, without having to take off work, you know, really with what they want to do. And so that's accessibility, which is a big, big issue.

Having different clinics in different places, we can offer things that are maybe not offered in one clinic. And having that information readily available is also helpful.

The place I want to get to the future is that AI tools can really revolutionize the way that we look at human gametes. We're already using the tools that are cutting edge that are on the market today, but our R&D is really interested in looking at even a deeper layer. AI is the only way to do that. How do you process a million data points from one cell and have that reflect what's happening in the clinical outcomes? 

It gives transparency also to the patient so that, say, you go through a cycle and you don't have the outcome that you want. You're not able to get pregnant. But why? The “why” is always difficult for patients, because a lot of IVF clinics just say, “Oh, just cycle again and again.”

People don't have unlimited resources and unlimited time. Now if you know that your eggs are poor quality, maybe you decide you want to do egg donation. This is what you have money left over to do. Or maybe you decide I want to adopt. Having that empowerment allows you to have more control over something that's been wrested from you, because people do feel like they're out of control when now they can't have a child when they want to have a child. 

At the risk of being self-promotional, how has working with Merantix helped you and your company, beyond the financial investment. 

I would never have met Felicia, my co-founder. Felicia would have probably never met Cristina, our third co-founder, had it not been for Merantix.

We didn't run in the same circles, right? Felicia was here had the ability to be in Berlin because of the Founder-in-Residence program. We had the ability to talk to see if it would work out as a founder-co-founder relationship. Those early days are really, really important. Now I've experienced what it's like to incubate here and to be sort of like a baby company, where you get help and where you get resources.

We had an event in the Merantix AI Campus town hall, which really kicked off our momentum in July of 2023, and now we're planning another event.

We've grown so much. I just talked to a contact I had when I was at Clue, and they were like, “So what are you doing now?” I told them, and they said, “You already have 22 employees?” We wouldn't have had that momentum if we hadn't started at Merantix.

Thanks very much for the chat!

In this article, I will talk about the manufacturing industry, which constitutes a keystone sector of the global economy. Its contributions span direct economic impact, indirect benefits through supply chains, and infrastructure development. Improvements in manufacturing processes have far-reaching ripple effects that permeate the entire economic landscape, driving enhanced productivity, increased competitiveness, environmental sustainability, and workforce transformation. Hence, we can secure long-term sustainable economic growth through innovations within the manufacturing sector. To write this article, I talked to Dr Silviu Homoceanu, an expert for the intersection of AI and manufacturing. He recently co-founded Deltia, a venture that is using Computer Vision and Edge AI to optimize assembly line processes. The article is based on this interview.

Opportunities for AI

Assembly line processes constitute a prime use case for AI. As Homoceanu pointed out, a significant portion (about 90%) of assembly processes in industries, such as automotive, is still manual. Companies typically optimize their assembly lines yearly by employing external consultants, who manually observe and note workers' activities, which is an expensive, biased and error-prone process. Deltia aims to automate and optimize assembly line processes using computer vision and AI to address this issue. They install vision-based sensors, such as cameras, at production line stations and use neural networks deployed on powerful, on-site devices to extract metadata about workers' activities in a privacy-preserving manner. This metadata provides information on production cycle durations, anomalies, activities, logistics, and supply chain issues. Deltia collects and presents this shop floor analytics to line and production managers, enabling them to optimize assembly lines more effectively.

Besides assembly line optimization, there are many more AI use cases within the manufacturing industry. Other applications include predictive maintenance, quality control, supply chain optimization, and energy efficiency. Machine learning algorithms can analyze vast amounts of data from sensors, cameras, and other sources to predict equipment failures, detect product defects, optimize supply chains, and identify opportunities for improved energy usage.

Furthermore, AI-driven robotics and cobots (collaborative robots) have the potential to revolutionize the manufacturing industry by working more efficiently and safely alongside human operators. Machine learning can enable these robots to learn from their environment and adapt to changes, enhancing their performance and reducing the need for manual interventions.

In conclusion, the application of AI and machine learning in the manufacturing industry is poised to transform various aspects of production, resulting in increased efficiency and enhanced product quality. As AI continues to advance and integrate into manufacturing processes, the industry will undoubtedly reap the benefits, driving long-term economic growth and strengthening its position as a keystone sector in the global economy.

The current state of manufacturing

The manufacturing industry is at varying levels of digitization, with some sectors, such as the chip-producing industry, being highly digitized, automated, and data-driven. A high level of digitization enables proactive action based on data analytics. On the other hand, the automotive industry has made significant investments in digitization. However, as mentioned above, the final assembly still involves a considerable human element, with manual and offline processes.

If we look towards small and medium-sized enterprises (SMEs), the level of digitization decreases further. Some SMEs have machines capable of producing data but may lack the knowledge to extract and utilize the data effectively. Many companies still rely on manual processes, clipboards, and paper-driven workflows. Some SMEs are attempting to introduce digital solutions, such as iPad-based systems, to collect cycle statistics and process performance data. However, more than these methods are needed to ensure high process quality and tap into the potential of an entirely digitized factory.

Market pressure is pushing companies towards digitization, especially in regions like Europe and Germany, where labour costs are higher. However, SMEs are increasingly aware that they must digitize, automate, and optimize their processes or risk losing competitiveness. This loss of competitiveness could result in workforce reductions and shifts in production sites to more cost-effective locations. This pressure drives increased interest in closing the digitization gap between SMEs and larger companies already implementing extensive digitization in their processes.

Innovating in a traditional industry

Often entrepreneurs experience significant pushback if they try to innovate in a conservative industry like manufacturing. But, according to Homoceanu, gaining acceptance among established players is possible if you have a deep understanding of the industry and empathy for the people who work there.

Despite manufacturing being a traditional industry, most people within the sector welcome AI innovations. The growing popularity and awareness of AI technologies, such as Chat-GPT or autonomous systems, have made the integration of AI more acceptable in the manufacturing space. However, there is some fear among workers, who may worry that AI technologies could replace them or invade their privacy. To address these concerns, companies must be transparent about their intentions and communicate that AI will be a copilot or assistant for workers rather than a replacement. In order to increase transparency and alleviate fears, one approach that we at Deltia are taking is to provide workers access to our system.

There is ample room for more entrepreneurs to enter the AI-manufacturing scene, as manufacturing significantly contributes to the global GDP. SMEs, especially, could benefit from increased support to democratize access to technology. By helping SMEs to become more efficient, entrepreneurs can contribute to making products more affordable. As the foundations of AI evolve, more innovations can be expected in the manufacturing space. However, entrepreneurs should demonstrate benefits for SMEs from day one without expecting them to have highly skilled personnel in data analytics or AI.

For AI entrepreneurs to succeed in the manufacturing industry, they must be customer-centric and adapt their offerings to the sector's realities. In addition, they need to understand that SMEs may not have the resources or expertise to handle complex data sets or advanced IT infrastructure. By bridging the gap between the two worlds of AI and traditional manufacturing, entrepreneurs can unlock significant value and drive innovation in this crucial industry.

1 in 8 women globally is diagnosed with breast cancer at least once in their lifetime. 685,000 of them will lose their lives to the disease each year.

Too many women are enduring invasive treatments, from chemotherapies to mastectomies. They face physical and mental pain on the path to surviving the disease, and their lives, post-diagnosis, are irrevocably and assuredly changed forever.

Yet it doesn’t need to be this way.

Research has consistently shown that the earlier a breast cancer diagnosis is made, the higher a woman’s chances of survival. Early-stage cancers detected in screening often have over a 90% survival rate, versus the 30% survival rate seen by advanced cancers. The gap between these percentages rises further in low and middle-income countries. Early-stage cancers can easily, affordably, and effectively be treated. Advanced cancers cannot.

The solution is therefore clear: to significantly reduce the number of breast cancer fatalities worldwide, we need more screening to catch these cancers before they even get a chance to take hold. So what’s stopping us?

Artificial intelligence (AI) has been viewed as one of the leading solutions to catching these cancers early. It’s almost the ideal use case for increasing the efficiency and accuracy of mammography screening, because of the way screening is typically managed. Women at risk, generally those around 50+, are advised to have their breasts examined regularly through mammograms, even if they don’t have symptoms. Specialised screening physicians then search these scans for subtle, early signs of developing cancer. Screening is a very repetitive task. Only 0.7% of all women in an asymptomatic screening population have breast cancer, but every image needs to be meticulously analysed and reported on in order to make sure no cancers slip through the net.

This abundance of structured data and the repetitive nature of the processes involved lend themselves perfectly to deep learning algorithms and AI.

Yet, more often than not, the AI solutions put forward fail.

At Vara, we are approaching this problem differently from the AI systems that have come, and gone before. As deep tech entrepreneurs, we started from scratch; with a fresh perspective and a bigger picture in mind.

Instead of a standalone AI tool, we’ve built an end-to-end platform that re-imagines the entire clinical workflow with the goal not to replace radiologists, but to supercharge them.

The Vara platform captures real-world data while in clinical use — an invaluable asset to transparently measure real-world health outcomes driven by Vara, create a real-time feedback loop for clinicians, and improve our models over time.

Most AI solutions promise to reduce costs for providers who already run screening at high volume. While that can be a good starting point, it’s a zero-sum game. We are convinced the real opportunity lies in increasing access to effective breast cancer screening to women worldwide in the first place. Women that otherwise wouldn’t be screened. That’s when AI doesn’t play the zero-sum game but instead drives value to everyone.

Today, more than 20% of all screening centres in Germany are on the Vara platform. Vara’s performance has been developed and validated on more than 6 million exams. And we have already started to launch AI-based screening centres in areas women weren’t screened at scale before.

But we’re getting ahead of ourselves. In our next articles, we’ll explain our fresh, new approach in detail.

At the rate we’re seeing, and if something significant is not done now, the World Health Organisation forecasts that deaths from breast cancer will grow to 1.04 million by 2040. We can’t and won’t rest until we bring this down to zero.

This article is a part of the vara.ai Founder's Blog, where the leadership team of Vara share insights to their journey. Follow the Founders Blog here.

Healthcare is one of the most obvious ways technology is improving our lives. Since the industrial revolution, life expectancy has increased by about 50 years. Of their overall lifespan, people spend an increasing number of years on education due to more and more specialized work environments. These trends and other cultural and economic factors lead to ever-later childbearing in developed countries. In Japan and Italy, for instance, the average age of first-time mothers is now over 30; in the US, it has surpassed 26. Along with other health factors, delayed childbearing has led to decreased fertility. The global fertility rate has halved over the last 50 years. 

These tendencies are why fertility medicine has become a growing field over the last decades and will play a crucial role in modern healthcare systems. This growth has been driven by increasing infertility rates and demand for family planning, caused by delayed childbearing, lifestyle choices and environmental factors. Assisted reproductive technology (ART), such as in vitro fertilization (IVF) and intracytoplasmic sperm injection (ICSI), have been some of the most widely adopted fertility treatment methods. However, these treatments are imperfect, and there is ample room for improvement. This is where artificial intelligence comes into play. I talked to Felicia von Reden, who just co-founded Ovom Care, an AI-first company that takes on the task of revolutionizing the fertility space. 

‍

Fertility: A (very) short overview

The fertility space is a complex field, overlapping with many other parts of healthcare. According to Felicia, three layers constitute the framework of the area: General Healthcare, Reproductive Health and Fertility Medicine. 

‍General Healthcare: This is the foundational layer of the fertility space. It encompasses various medical disciplines that indirectly impact fertility by promoting overall physical and mental health. It includes nutrition, endocrinology, genetics, mental health, and more. All of these factors can contribute to an individual's or a couple's overall fertility health.

Reproductive Health: This is the second layer that focuses explicitly on the health and functioning of the reproductive systems in both men and women. It covers menstrual health, sexual health, pregnancy, childbirth, and menopause, among other topics. Also, it deals with preventing and treating sexually transmitted infections and conditions like endometriosis, which can directly impact fertility.

Fertility Medicine: This is the most specific and focused layer of the fertility space. It addresses issues directly related to conception and pregnancy. This involves diagnosing and treating particular fertility issues like ovulation disorders, male factor infertility, and unexplained infertility. The treatment of these issues includes medical techniques such as in vitro fertilization (IVF), intrauterine insemination (IUI), and third-party reproduction methods like surrogacy and egg donation.

In summary, the fertility space is a multilayered field within healthcare. It begins with general health and wellness practices, narrows down to the targeted area of reproductive health, and finally hones in on the specific domain of fertility medicine. Each layer is interconnected and integral to the effective functioning and outcomes of the other.

‍

How AI can transform Fertility

During the last few years, the progress in AI has reached the field of fertility. The technology can transform the in vitro fertilization (IVF) landscape, helping improve outcomes and optimize treatment plans. As a result, we see more and more companies popping up, developing AI-based tools for IVF clinics. 

‍One of AI's key applications in IVF is embryo selection, a critical yet challenging part of the process. Machine learning algorithms can analyze vast datasets of embryo images to predict viability more accurately than human embryologists. The automation of these processes frees up time for the staff in the lab, leading to a better quality of care. By analyzing a wide array of patient data, including medical history, genetic information, and previous treatment responses, AI can suggest the most effective strategies, such as the optimal hormonal dosage for ovarian stimulation.

‍Furthermore, AI has a significant role in genetic screening. For example, preimplantation genetic testing (PGT) can benefit from AI algorithms that detect chromosomal abnormalities more accurately, increasing the likelihood of a successful, healthy pregnancy. Predictive analytics, another promising application of AI, uses many factors like age, lifestyle, and genetic markers to provide a more precise prognosis, empowering doctors and patients to make informed decisions. In addition, automation of routine tasks, such as cell counting in an embryo, reduces human error and increases efficiency, allowing clinicians to focus on more complex tasks. Finally, AI plays a crucial role in research and development, helping to identify patterns in large datasets, which can lead to breakthroughs in fertility treatments. 

‍So far, IVF is still very expensive. However, if the technological opportunities get realized, this will drive down the costs in a way that makes IVF affordable for everyone. Moreover, fertility medicine will become more precise, adding enormous value to many people's lives. As Felicia put it: "We are all becoming parents at a much later stage, and infertility is becoming a mass phenomenon. Even if we could increase the success chances of IVF by only a few percentage points, this would already prevent a tremendous amount of suffering. And I believe that much more is possible:" 

‍

The Future of IVF Clinics

‍However exciting the above vision is, we have yet to see AI tools good enough to lead the transformation so far. According to Felicia, there are mainly two bottlenecks that hinder progress.

‍The first significant bottleneck is the lack of high-quality data. The companies that develop AI tools for fertility do not create their own data and have to rely on the data from the existing clinics. The issue, however, is that most of these clinics are not sufficiently digitized. This leads to the data on the market being scarce and of bad quality. According to Felicia, if you buy a dataset, often you cannot use large parts of the data due to quality issues.

Moreover, the data does not represent the whole patient journey from end to end. As a result, innovators are left with fragments and have to rely on those to build correlations. This makes the tools that are being developed imprecise and costly compared to what is possible. 

A second issue lies in the slow adoption from the side of the IVF clinics. Their business model typically involves selling IVF cycles, which hinders the adoption of better procedures and innovations in treatment. Because clinics often generate more revenue when more cycles are performed, there is a financial incentive to recommend additional cycles, even when the chances of success might be low. This model also discourages the adoption of new, more efficient techniques or personalized treatment plans that might reduce the number of cycles needed for a successful pregnancy. The focus on selling cycles is diverting resources and attention from investing in and implementing innovative procedures or approaches. The financial incentives are not aligned with the transformation.‍

This is why Ovom Care wants to lead the transformation by solving the problems from the root up. Felicia and her co-founders plan to build a new generation of digitized IVF clinics. The goal of Ovom Care is to first form the foundation for great tools by creating Clinics that can produce the necessary data and adopt AI-based tools. These clinics can then function as a data flywheel to build better fertility tools powered by AI. The promise of this pathway lies in an overall better quality of care through better success rates and more financial accessibility to fertility medicine and a better patient experience through improved digital support.

‍

Everybody knows a victim of a scam. Thanks to AI, scams are only getting more pervasive and more efficient. 

Can good AI triumph over bad AI? Company Shield, one of the newest ventures we’re working with here at Merantix, is putting that to the test.  

Julius Muth and Tom Müller met while working at German software unicorn Celonis, where Julius worked in sales and Tom in engineering. Based in the company’s Madrid office, Julius and Tom began discussing an opportunity for a new kind of AI-enabled cyber security employee training. They launched Company Shield, which works with clients to continuously train their staffers with automated, AI-generated attacks that are based on real-life scams.

We at Merantix recently teamed up with the young venture on its path towards becoming the next great cybersecurity giant. I sat down with Julius to discuss how AI is changing cyber security and the corporate threat landscape.

Our edited conversation is below:

How does Company Shield work?

AI technologies allow for super sophisticated attacks against the human factor. For example, fake phone calls, fake video conferences, fake invoices, and the combination of it. I recently saw that some scammers even used fax again, which is quite funny. Employees are the weak spot and with new AI technologies. It's even harder for them to spot the intruder.

What we do at Company Shield is protect exactly this human layer, but not by these lengthy online trainings, but by sending simulations of real circulating scams through all kinds of communication channels. For example, a couple of weeks ago Ferrari CEO was imitated by someone as part of a really good scam. Just moments before the employee really transferred the money requested, he got a little bit suspicious and asked the caller about a very personal thing: a book that the CEO just recommended to him. The whole scam suddenly became obvious and saved Ferrari from some painful financial damage. And it's exactly stories like these that we replicate in our training.

You showed us an amazing and scary example at a recent Merantix team lunch, where you used my “voice” asking for an urgent invoice to be paid. I would imagine this is a compelling sales technique when you’re pitching Company Shield to potential clients. You can actually demonstrate the CEO’s voice.

We need to be a little bit careful because according to the EU AI Act, you're not allowed to create a voice model without someone’s approval. Even if the training data, such as Youtube videos, is publicly available. But of course most of the prospects we are talking to give us permission on the spot, allowing us to showcase such attacks to them. And while reactions to such a live showcase are different, they all encourage us that a solution like ours is really needed in today’s world.

Cyber security training for big firms is kind of like “box checking.” Everybody clicks through as fast as they can to get it over with. You’ve used the “hot stove” analogy around here before. That the best way to actually train someone is to fool them when they least expect it and then train them in the immediate aftermath, right?

Exactly! Click-through online trainings, designed to let all employees pass, in order to check a compliance box, are a waste of time and don’t protect companies from real cyber attacks. Of course  people have realized in the last few years that a simulation-based approach works a lot better, meaning that you test employees in more realistic situations, and most importantly, on a continuous basis. This is where we see the “hot stove” effect. Right in the moment when an employee falls for one of our simulations, we have approximately 60 seconds of increased attention from this employee to teach him how to deal with this type of attack. 

AI is making attacks more sophisticated, but how is it also enabling better defense? 

That can happen in many ways. Whether it is on an infrastructure level through things like better anomaly detection. On the human layer, you also have a lot of possibilities, for example in email protection to say, “This looks 80% like a common scam that's out there.” Also, there are a lot of  AI detection tools [to verify] if some content is generated by an AI, for example whether a voice belongs to a human being, by checking for some frequencies that the human ear can't hear but that a machine could. You can detect the same for deep fake videos and all these technologies will for sure be used to also fight scams. But as of now, hackers have the advantage of being able to experiment with these new technologies without the wide usage of defense tools for protection.

Is there a particular AI-driven scam that seems to be gaining steam in the corporate world? 

A very prominent example was the fake video conference scam at Arup, which cost the firm $25 million earlier this year. Other scams I see evolving are often HR related, because obviously in the HR department non-IT people work with a lot of file attachments. That's one way through which a lot of hackers get their malware installed on company computers. Another AI-powered scam I have seen multiple times now is the deep fake phone call to ask for the payment of an outstanding invoice.

That’s the one you simulated for us at Merantix. 

Correct. It’s pretty much always the same pattern. If you’re targeted, you are a person in charge of transferring money. You receive a call from your superior, in most of the cases the CEO, saying that there is an outstanding invoice that has not been paid. The attackers are well-informed and user very realistic invoice scenarios based on public data. Sometimes this call only goes to the voicemail, but always the CEO doesn't have a lot of time but creates this urgency. In some cases employees fall for this and actually wire the money to a fake bank account. The damage is not always in the range of millions, but €100,00 or €200,000 also hurts the affected companies.

What’s been the biggest challenge growing Company Shield so far?

We have a huge pile of work consisting of new ideas, customers to follow up with, interesting trends within cybersecurity, so that could work for the next 100 years. Unfortunately we are still a small team, so we need to constantly make sure that we focus on the right things, which is very hard to do and we don't always 100% achieve that. 

What impact do you hope your venture will have?

We see our impact not only on a company level but also on a private person level. We really hope to equip humans with the necessary knowledge of how these AI technologies can be used in a bad way against them, so that we can reduce cybercrime damage within our world. 

What feedback from clients is most helpful for improving your offering?

Many things! Ultimately, we want to steal as little time from our users as possible, but at the same time we want to achieve the maximum learning effect and with this maximum protection levels. This makes us very dependent on user feedback on the right difficulty, the optimal frequency of our cyberattack simulations, etc. And of course it is super helpful to hear about any cyber attack that recently happened, as many of them are still not publicly disclosed.

Thanks very much for the chat! Check out Company Shield here.

Earlier this week, I received some nervous calls about the state of the AI world, with questions around whether the AI “crash” had finally arrived given the events in the market. The headlines from earlier this week were blunt:

Business Insider: “Tech stock sell-off may herald the long-awaited AI reckoning”

Bloomberg: “AI Rally Halts Abruptly as Global Selloff Sinks Tech”

Fortune: “Monday’s stock market crash shows that the AI revolution isn’t quite here yet”

So, has the AI bubble finally burst?

No way. 

And, as the dust settles, there are two reasons that Monday did not shake my confidence as an investor in AI startups.

First, the sell-off was about much more than AI. Investors got scared over US interest rates, inflation, a weak American employment report, the unraveling of the “Japanese yen carry trade,” all culminating with fears around volatility finally reaching boiling point. 

Hype and investor concerns about the ROI of AI had a role to play. It was a spectacularly bad day for big tech companies, who have notably spent billions on AI initiatives and investments. The “Magnificent Seven” (Alphabet,  Amazon,  Apple, Meta,  Microsoft, Nvidia, and Tesla) lost more than $650 billion in market cap. Nvidia, the AI standard bearer in the public markets, was particularly hard hit.

If you look closely, the corner of the AI world that is “bursting” is the part that we at Merantix always saw as a bubble: so-called “wrappers” without a fundamental business model. We’ve avoided these sorts of companies, and as The Wall Street Journal reported this week, some of those tools are now getting bailed out by Big Tech companies like Google and Amazon who are “licensing” their technologies and acquiring their teams.

Party like it’s 2000

So is AI over and done with? Far from it. I’ve said this many times before, but I believe this moment in time — this AI industrial revolution — is like the early days of the internet. If this is like the year 2000, the real category-defining companies of the future have yet to built. Or they are only being built right now. There will be bumps in the road, but I believe we are heading in one direction: forward. AI is simply far too impactful to think otherwise.

Some believe that Monday’s sell-off echoes the 2000s bubble. Did that dot-com crash stop the development of the internet, as some people foolishly warned at the time? Of course not. It was only the beginning, but there are always winners and losers. 

Indeed, in my opinion there will also likely be consolidation among the large infrastructure providers, with only room for a few major players as LLMs become commoditized. Tough times in this extremely high-profile and well-funded space would undoubtedly spark even more scary headlines about AI being just a passing fad.

Which brings me to the second reason I’m still confident in AI. I’ve never felt more sure of our investment strategy at Merantix: focusing on solid value creation and not simply imaginary AI forces.

It’s about value creation, now and always

Across our current portfolio and with the investment pipeline we have developed, I see companies that are leveraging AI to deliver quality products and services to their happy customers. Our venture Cambrium, for instance, recently announced a new investment from Google’s Gradient Ventures, a strong validation of their novel protein development work (with their first product, a micro-molecular and skin-identical collagen, already seeing huge traction in the market). Or take our venture Deltia, which recently announced 4.5 million Euros in seed funding. Deltia modernizes manual assembly lines with computer vision and AI, with customers like Viessmann and ABB.

That’s not hype, that’s real business achievement and real technological innovation.

There will be many more market life cycles in AI. I expect there will also be many more headlines written that exuberantly rave about AI’s promise — and later headlines rushing to declare AI dead. Rinse and repeat.

We don’t concern ourselves with either, keeping our eye on identifying the real AI value drivers of tomorrow.

But as an AI investor, I’ve never been more optimistic about the European landscape, and I believe there have been a few recent signs that show the strength we have here. 

Let’s take a few recent headlines. 

“AMD to acquire Finnish startup Silo AI for $665 million to step up in AI race”

US chipmaker AMD made a big move in Europe by grabbing Silo AI in a large cash deal. Founded in 2017, Silo offers AI products to large corporate clients like Unilever and Philips, helping them build models and integrate AI into their companies. 

After Google’s 2014 acquisition of DeepMind, it’s one of the most significant AI M&A deals in the region. Silo was founded in 2017 and has about 300 employees.

From my perspective, this also shows the strength of vertical integration in AI: By acquiring Silo, AMD will essentially have an in-house model creation team to help their customers and improve their offering, likely in the hopes that it will give them a leg up in the ongoing chip wars against the market leader Nvidia. 

I anticipate that we will see more deals like this, as model providers and infrastructure players try to enhance their products by getting closer to the application layer.

“Defense AI startup Helsing raises €450 million Series C”

Led by US investor General Catalyst, Germany’s Helsing landed a mega Series C round, with plans to expand into the Baltics and spend €70 million over the next three years to combat the Russian national security threat. It’s just the latest big AI round out of Europe (like Mistral’s €600 million Series B).

The news marked another strong round for Helsing. Like the AMD deal, it does show a certain elephant in the room: the inflow of money from the US underpinning European AI. One could be cynical about this and say that it shows that Europe can’t exist on its own without US influence and capital. But in my opinion, the investment shows that Silicon Valley recognizes the value creation happening here.

“Why Europe’s startup ecosystem is healthier than ever”

Finally, I enjoyed this fantastic column in Manager Magazin (in German) written by Katharina Wilhelm, a partner here in Berlin at Index Ventures. 

Katharina summed up perfectly why I’m so excited about the European AI ecosystem. She highlighted an interesting datapoint from Dealroom: In 2024, growth in European AI funding rounds is expected to outpace the US for the first time this year. AI funding in the EU will grow by 113% from 2023 to this year, compared to 47% growth in the US. 

To be sure, the US is operating on a greater scale. But the growth differential does illustrate the potential for Europe’s future. Katharina points to a few reasons why Europe is thriving, including our talent pool and our advantages on the application layer. 

Doubling down on the European advantage

I couldn’t agree more. While most of the attention in AI tends to revolve around chipmakers and large language models, I believe the most attractive investing opportunities lie in the industrial applications of AI. That’s our investment hypothesis at Merantix. 

I believe that in 20 years time, there will be many household name AI companies that were built in Europe, across robotics, mechanical engineering, healthcare, life sciences, fintech and tech-bio. 

This is where Europe really flourishes, since we have such a great concentration of industry knowledge and top-tier research. At the same time, our investment landscape is maturing, our entrepreneurial class is becoming more sophisticated and knowledgeable, and the ties and connections between the European AI ecosystems — from Berlin to Paris and beyond — are lifting all boats. 

Of course, the US is still ahead, and we in Europe need to be humble about what the US has been better at for decades: breaking down silos between research and implementation, flexible institutional investors interested in venture capital, and a pipeline of successful entrepreneurs that become investors themselves after big exits -but that’s a topic for another Substack post!

- Adrian

Connect with me on Linkedin.

Janette and the other judges heard pitches from 12 very promising early-stage AI ventures during the competition. They were battling for some great prizes, including a spot on the shortlist for the upcoming Forbes AI list and a free week of co-working and an event co-hosting slot at our Merantix AI Campus in Berlin. 

One presentation stood out for the judges: the winner was Futurail, which is developing an autonomous stack for self-driving trains. The company is working to remedy the train driver shortage, reduce costs, and increase reliability and flexibility for future train travel. 

After the competition, Janette sat down with Futurail’s Co-Founder and CTO Dr. Patrick Dendorfer for a quick Q&A. Before founding the company last year, Patrick received his PhD from the Technical University of Munich, where he researched human trajectory predictions based on deep learning methods. 

Janette: Congratulations on winning the competition! How does Futurail utilize AI to make a better future for travelers and rail operators?

Patrick: Futurail leverages AI to build autonomous train technology focusing on computer vision and deep learning for obstacle detection. Our product facilitates more frequent operations and reduces transportation costs for travelers and rail operators.

I would imagine there are some fears among current human train drivers about the threat of automation to their jobs and livelihoods. How do you address those concerns?

Acknowledging the industry's concern on automation, we view it as an opportunity to enhance working conditions for the existing drivers to focus on more complex aspects of their roles, i.e., high speed trains. Moreover, the increasing gap between available and required train drivers underscores the urgency for innovative solutions.

In 10 years time, what do you think train travel looks like and how will Futurail be involved?

We envision a future where rail travel is more accessible and sustainable with autonomous and eco-friendly trains connecting remote areas with frequent and reliable service. Futurail will supply the software ensuring the safety and efficiency of autonomous train operations.

How do you think winning this AI Pitch competition will help your startup?

Winning this competition amplifies our message about the significance of technological advancement in improving rail transportation. We hope this awareness attracts investments, catalyzes political will, and increases the prominence with the tech market.

Thank you for the interview, and we can’t wait to see where this will grow further!

If you want to get in touch with Patrick, best to reach out to him on Linkedin directly, or keep an eye on upcoming events on the Merantix AI Campus, where he will hopefully soon be speaking, too.

More impactful AI startups from the DACH region will be featured in the upcoming AI30 list by Forbes and Merantix. Be sure to follow us on Linkedin to be informed once applications are open!