All right, listeners, welcome to another episode of The Edge, the podcast of Cherry Ventures about product and AI. I am Jasper, partner at Cherry in Berlin, and I’m joined by our venture partner Lutz Finger, who is in the Silicon Valley, and he’s just welcoming our guests. So we will interview Ferdinand Therne, or Ferdinand from France, who we met at Station F, nice place to be. You should be there when you visit Paris and many other places, obviously in Paris.

And he will tell us about his startup, Lasco AI, which is in the creative space and is following one of our thesis of the age of super creation and how to create tooling around generative AI and make it productive for their part in B2B context. So. Let’s go. Hey, this time from San Francisco, the Cherry office in San Francisco, together with Ferdinand and Jasper.

Yeah, Jasper, how is Berlin? It’s amazing. It’s 35 degrees today. I think we still have 32.

So I’m also overlooking Berlin, but I’m on the fifth floor. So that’s enough in Berlin to overlook the city. Yeah, very much looking forward to the conversation after we met you, Ferdinand, in Paris. Exactly.

Looking forward to it. I’m quite excited. Very excited. From Paris to San Francisco.

Well, Ferdinand, what we like to do is obviously when we have guests like you, just give a quick intro. I learned that you love Stefan Zweig, which kind of was like, wow, that’s an interesting fact. But who’s Ferdinand? Or like you pronounce your name correctly in French.

Don’t tell my wife that I can’t say it. So the French way is Ferdinand. But we can do it like the German way. Like Ferdinand.

It’s probably easier to pronounce and to get for like English speakers. The reason why I said don’t tell my wife, my wife is French and she always makes fun of me not being correctly able to talk. I think it’s better probably for you to say like more useful things in French and Ferdinand, right? On the like, must learn with Jay Lister things earlier.

For us, it would be mostly interesting. Obviously, there’s a lot to tell us about your life, but you come from a consulting background. So it’s interesting. I come from a consulting background.

But you’re from a different firm. So interesting to understand how from there you come to this totally chaotic, high risk, low salary world of building a company and what’s driving you there. So basically, actually, when I joined McKinsey as a consultant, I already knew that I didn’t want to become a partner. Basically, I knew that I wanted to launch my own company at something.

And if I joined McKinsey, basically, I joined the Casablanca office, which was the entry. For like, French speaking, Africa, and because I was like, okay, Africa is a super interesting continent, which I still think is true. And like, why not launching something there? Because I had to spend a bit of time in Kenya, spent six months in Kenya, working as a chief of staff in a startup.

I was like, wow, so many things are going on here, like, you can really create life changing products, which I felt was less the case back in the days in Europe, like for reason of, like, maturity of the ecosystem. So I was like, okay, let’s do it the safe way. Let’s try and see how it goes with joining McKinsey Casablanca. I spent two years there and I was like, wow, actually launching a business in Africa is extremely difficult.

It’s extremely difficult, especially if you’re a foreigner and so on. And it’s very difficult to make a successful tech business. And at the same time, I think the French ecosystem had made huge progress. And I was like, wow, isn’t it a good time for me to go back home and to launch something to contribute to the economy of my country?

But also to launch something in a more thriving ecosystem. And at the time, I tried a few ideas on the side. For example, I made a nodular guide of points of interest in Paris. And I was like, no, I didn’t forge a strong enough conviction for me to launch something.

So I decided to make an intermediate stake and join Alma, which is a kind of competitor of Clona doing binary pay data. Then I spent two years at Alma contributing to the hyper growth of the company. I joined, we were around 50 employees. I left 400 employees.

And this kind of confirmed that I wanted to be an entrepreneur. I also learned like what hyper growth means, the pros and the cons. And I was like, okay, now I feel ready to launch my own thing. And how did you get your own thing?

So it has been a bit of an active way. I mean, it’s a good thing when you join a startup. It’s like, especially like in that hyper growth phase, you see a lot of things. You see a lot of pains.

So I left Alma together with one of the two co-founders. And we were like, okay, we both want to launch. We both have a bunch of ideas. Now let’s just explore and figure out what exactly we want to do.

And so to be honest, like we spent six months working on FinTech ideas because we were like, you know, a bit like Bayesian book. We were like, that maybe comes from the consulting background. You know, like you want to follow the framework. I was like, okay, we have consulting.

We have like FinTech backgrounds. Let’s figure out a FinTech idea. But for some reasons, probably like our fault or the market’s fault or whatever. But at the end of the day, it’s always your fault.

We didn’t figure out like a good FinTech idea. And then we were like, okay, what other pains do we know? And my co-founder had launched a product like for designers before. I had worked in the marketing team.

So I knew all the pains around like creative assets. And we met the third co-founder who had been working on generative models for images for three years. And basically at the beginning, it was a bit like, okay, it looks like we know the pain. It looks like we have a good founders market.

Let’s give it a try. You know, that’s a bit of a mixture because like own experience and opportunity. It’s really… It’s really interesting how you saw different business models and deliberately chose that one and also that area.

Maybe for all other founders listening that were thinking about Africa and were thinking about FinTech, what is something you would recommend to these people, what they should really understand and consider before building something in Africa or building a FinTech from someone who deliberately says, no, I want to do something else? I haven’t built anything in that. It’s one of those two areas. I think like it’s just what I will say is to be…

I haven’t built anything in that. It’s one of those two areas. I think like it’s just what I will say is to be… I think so.

Africa, my view, the conclusion basically I came to is that being successful, it’s very hard to build a successful business. And most of the successful businesses are there, like traditional businesses. So if you want to make furniture or services, it can work. You can have very profitable, very beautiful businesses.

But it requires to be hands-on, to spend 10 years in the country, build local relationships and so on. So I think the first question is really a lifestyle question. And this is basically the choice I wasn’t ready to make. And of course, you have some good fintechs and so on, like Nigeria fintech scene is awesome, a Kenyan fintech scene is awesome.

But most of the business that actually work are non-tech. So that’s my learning for Africa. And for fintechs, basically the conclusion I came to is that the space has been extremely out over the last four or five years. So I think either you go on something like very close to the rails, like payment rails, like you have Numero, for instance, in France, like they’re doing like connecting the, like really banking APIs and so on.

Or you do something like very verticalized. So for example, like one of the last ideas that finally we didn’t pursue was like payments for like healthcare in Europe, like connect all the insurance refunds and so on. I’m sure they, I mean, no, not sure. Otherwise maybe I wouldn’t be doing that.

I don’t know. I think there is something to do here, but yeah, that is like either very technical in depth or very verticalized. Otherwise I think it’s super difficult to create something that works in Finland. Yeah.

I like the thinking a lot. And I think that’s a good leeway to talk about business models. You spoke something very often when I’m in class and I talk to students, there are so many students who say like, oh, I want to, I want to be an entrepreneur. And then I’ll look at them again a year later and they are at McKinsey.

What would you recommend people who want to become an entrepreneur? You said just a second ago, and I knew that this is the right thing for me. It sounded like, it sounded like a declaration of love. And I met her and I knew that she was the right one for me.

What do you recommend for somebody who says, oh, and many of our listeners actually think about entrepreneurship, right? Like how do you, would you do the same journey again? How do you figure out whether you have the itch that you want to become an entrepreneur? So basically I think that’s a bit of two questions.

The first one is like, how do you figure out you’re made to be like, entrepreneur is a good path for you. And the second one is, would I do the same thing? So if I think the first question first, I, think I always love building things or leading projects. So for example, back in business schools, in business school, I led the like a student, we have like something in French, we see like student led consulting firms.

So basically you’re like your own CEO. We made like around 500K revenue. So with a team of 10, 10 in total. So like I was managing my own business back in the days and I was like, wow, that’s exciting.

You know, that’s super tough. And sometimes we work till like 3am and so on, but that was cool. This is, I think the first step, do you like that type of uncertainty? Do you like being at the end of the day accountable?

It’s up. And also the second thing is like, try to build things yourself. So for example, that’s also when I think I realized I wasn’t maybe made sure enough when I built this side project in parallel with McKinsey of these like city audio guide. That was like, wow, actually it’s, I have no clue how to do this.

I think I was expecting a bit of an overnight success. Because when you’re in consulting, you have very, very fast validation signals. So basically if your slides are very bad, your manager will call you and say, well, what did you do? Like, it’s really, really bad.

And so you have an instant answer for the market. When you are an entrepreneur, it’s way harder. You need to figure out to deal with a lot of more, a lot more uncertainty. And I think these two things like made me feel, okay, actually I liked that, but I saw also like what was missing.

And this is also why I decided to join Admind and discover like what’s for real entrepreneurs. I can really resonate with that because I guess at a consultancy, it would be comparing it to a startup. You would just do the slides yourself and you go directly to the customer and present them, get the direct feedback loop, which could be positive or negative, but you don’t have that. So it feels a bit numb.

And then you also, with your recommendations, you just don’t know what the customer’s doing with them later on, which feels annoying. Yeah. I like the building part. As you can see in the background, I do a lot of building in my private time.

Cool. Shall we jump into your product? So, and the motivation for doing it because it’s totally different than furnitures in Africa and FinTech. What are you building?

Yes. So basically with my two co-founders, we’re building Lasco and Lasco is AI designers for marketing team in small and medium sized organization. So these designers, they produce on brand visual assets. They are available to, anyone in the marketing organization, even without like any design skill.

And they are accessible within the marketing workflows, starting with a Slack integration. So basically doing that, we bring an additional capacity to the company, to the marketing team, and they can express their brand like without limits. And they can basically with that, like compete against their largest competitors who have like huge design teams, brand, like content teams, and so on. So this is a track in a minute.

Just to make it more clear what is in there. It’s not just the asset creation. It’s also the asset variation. It’s detailing, but it’s also a database that has a structured overview of what assets are out there.

So I could search it and then I could even alter the assets based on maybe customer feedback, et cetera, without needing to open any product from Adobe and the like. Exactly. So like, the product is still very early. Like we launched, but we should do MVP in April.

So it has been just a few months, but yes, you have this image generation, which relies on a fine tuned model on the brand identity of the company. And there is like what we’re currently building, like for example, a gallery. So you can access and anyone can access to any visuals that has been generated in the past. And what we will be, we build around that is for example, like what we already have also like editing, like if you want to make a variation, we have like within the gallery, you can search for past visuals and so on.

Nice. And for us, this is a little bit in our thesis of the age of super creation, where we say it’s, it’s more building the tooling around these amazing models, how to tame them, how to control them. So I think this leads definitely to a technical part for Lutz to understand how you can tame this beast. Can you walk me through a typical user flow?

Like I have a logo. Think about the cherry logo. I like beautiful color text. And now I, yes, for example, and what would I now as a user do?

So I’m in the Slack channel. All you have, it’s a cherry logo. And now, so I think that basically to get it right of products works in two steps. The first one is not like, is basically we do a call with you.

Like we haven’t productized actually the first part of, of the value proposition you share with us your brand identity. So you give us access to your Figma, to your like Google drive. We want to understand also what your brand is about. Then we get your assets and we train a custom mobile on your assets.

So we train a Laura, as you said, can be several low rise, depending on the complexity of your brand identity. So the example of the logo might not be the right one because usually you always want to keep the same logo, but let’s say you have a mascot. So for example, let’s say you are duolingo. You have to do like, fun green, all we train the model on building.

That’s the first spot on the dual mascot. And then the second part is that you can access within Slack or on our platform, an image generation solution where you can say, okay, actually I wrote an article about, I don’t know, holidays. So it’s where like holidays, summer holidays are starting. I put the text of my article here and I generate a few visuals of the mascot and joins a son or having fun by the beach.

Got it. So essentially for cherry, we could give you the logo and suddenly you would place the logo in San Francisco, in Berlin, in like on the or as are in different. And you would show logo generation. Now, what is the important part?

Let’s take cherry again. There is the font. There is the color. Maybe we like the logo to be placed on a sailboat.

Like we are always discussing to doing. Yeah. You’re a single event, this pounders, but like, like, here we go. Um, but it’s on a sailboat could be nice, but we might as well recreate the logo in water style.

So now the color is missing and also the, but we keep the font. How do you decide what is important for us as an organization? Today? Most of this is done with a human in the loop.

We have had an open discussion with you. We, we look at your brand guidelines and usually like we only work with brands. We’ve had an already like existing brand book, brand kit. And so if you haven’t like figured out what is important for you, like if you’re just thinking about creating your brand, we’re not the right solution.

Usually we have this discussion with the client. Then we make a test. So we change the model based on our understanding. We give them access to the solution based on the feedback.

We adapt that to ensure we are constantly on break going forward. We, we’ll probably productize that part. So LLMs now are very good at understanding an image. And this is exactly the type of thing.

And this is where, you know, we’re starting to go from a feature to a product. But he was like, Oh, okay. I need to add an LLM here to understand, maybe to re interact with the diffusion model, which is generating the visual itself and to make a loop between the LLM and the diffusion model to get to the right asset. Okay.

Understood. I would, I would love to tell the model that part different color, that part larger, et cetera. So I would like to split it up versus just having a prompt that then changes everything again. Is that already possible or not?

It’s always that, is that actually important in your workflow to be able one day to do this? One of the other key parts, which again, from the future to the product. Yeah. So basically the diffusion models, it’s model, it’s safe.

Plus the editing modules on top of that doesn’t understand, like, wouldn’t understand if you’re like making blue. Yeah. That’s what we’re building is a conversational agent. So we understand that when you say, Oh, make blue, it means actually the background should be blue or it means, you know, right.

That instead at this point. And so today we started doing that. We’ve just understand your inputs. We build this.

We build this conversational pattern for the input, not the iteration yet, but this is definitely where we’re going. We have two challenge, right? And you tell me whether this is correct. Challenge.

Number one is we need to extract what are the correct guidelines. If I say we take this video here now, and yes, this is a podcast, but do your podcast listener, we actually recording a video. So now if I take this video and say, I want Ferdinand and myself sitting on the, on a vast plaza, we had burning men, just, just burning me. And so we could be on the plier and a huge Sherry logo in the background.

And we have continued off this discussion. That is, we now need to extract that information, bundle it with the brand guidelines. And then the second problem is we need to do the rendering and for the rendering, you’re using a diffusion model and you train the diffusion model on the brand equities. And, and for that you use a Laura.

Do you want to say something about what is a Laura? Basically a Laura is a retraining of the model, but instead of retraining all of the model weights, you just do it on a few parameters. Yes. And so it basically enables you to have extremely consistent results, but in a pretty short, less than an hour training time.

So it doesn’t need all the computer free training, the world model. Exactly. Now, meaning, you have a neural network, you have a lot of weights and biases and they are all pre-trained. And now Laura is doing a low rank adjustment of those weights.

Now those adjustments needs to work with the right data. So essentially the first step extracting the information is the technical key part here. Okay. So now we have a product.

Now, do we have a feature product already? It’s the most annoying feedback of a VC, I guess, kind of saying, ah, we think you still have a feature. Bottom line, a product is something you can sell in the package with a certain price and then at scale. So you could say it’s just a combination of features that are solving a very concrete problem of a very concrete persona, user persona or end buyer.

So it could be different or same, but if you have a feature, it might be part of hundred things somebody has to do during a day, but a product is definitely more five, 10, 20 things. And I’m covering a part of a workflow. I’m not doing one thing here than 10 other things in a different tool than two things at my tool, et cetera. So that makes it more a product.

And then when you really come to the company, obviously around it, that’s how you sell it, how you support the customers, et cetera, et cetera. The very simple version of it. Where are you right now? I guess very early, so that’s fine.

But how, because we always come from this pain point, urgent pain. I mean, that’s kind of what everybody loves, right? An urgent, urgent pain in a market that is changing and growing and that exactly needs, needs your product. But where is the urgent pain?

Where’s the wow, where you can convince someone, yeah, I really want this, although it’s not perfect, but I still want to test it. You can get your user feedback. Can you describe a little bit your thinking there? My first thinking is, I think this question is nothing new with AI.

Like my last company, for example, Alma, it was please payment. Basically. And we shot, he was a CEO used to work for Stripe and he saw this demand coming from clients. Is it also, I want to speak payment solution and strike could have done that, right?

They had to be, they had the data and so on. We didn’t decided to focus on something else. I’m thinking also closer to our space about photo room for the room is like background removal. You could see it’s a, just a feature, but like both these businesses are on track to a hundred million AR.

So the last two years of like Jenny, I wave this question, applied. And I think like, also why is that questions that new is that you have two probably things that are changing. So the first one is like AI creates a ton of opportunities and nobody understands like how fast incubants will be able to ship AI features, which could indeed turn a lot of AI startups into just like features may take care that it’s pretty difficult to be able to good AI product because you have to deal with probability and uncertainty and an established player doesn’t like to deal with that at all. It’s like, it’s like, You already preempted the second question because we have a question on product, but that’s the way.

And the second part, which makes this question also quite relevant in this stage of AI. So new thing to that question is that there is some, like you have a lot of AI products or features that replace the job of a human today, like of a white collar job, you know, and there’s no benchmark. There’s no, like, we don’t know how much people are ready to pay for that. Like, how will it compare?

How will it compare with the willingness to pay with a, like a, with a print designer, for instance? Yeah. And it’s also pretty bad pitch. If I come to you and say, I will replace your colleague, you need to want to replace your colleague right now.

Exactly. Actually, I think the replacement like pitch doesn’t really work. Like nobody wants to hire half of the team, especially in Europe. I think in the U S the mindset might be a bit different.

No, I also think there is the question later on is only the cost reduction. You set a, a brand image because you sell a product, you have a huge mark. That’s the reason why you’re working on your brand. Right.

And now when somebody says you can save a little bit and he was like, yes, but I have a higher risk not driving in my margin because my brand sucks. Yeah. So that’s wouldn’t do that. So that’s why you said you go to small medium companies.

Yeah, exactly. Not yet have that capable. Yeah, exactly. Exactly.

Only start that working by the way in industries where there’s high churn, high turnover, like call center, customer care, because then there’s definitely an effect of automating. But the biggest driver is then the onboarding of the people supporting the people make them work easier. Right. So it’s again, not really cost cutting at the end.

Yes. I think the way we see today, I think there’s like several things to be qualified as a product, not a feature, but at the beginning, the lines are very blurry. The way I see it, like we bring today an additional capacity to a company. So they don’t have a brand design team or they have a very limited bandwidth.

And so we enable them to express their brand. Like they couldn’t never have imagined that. And so they are like, wow, actually it feels like magic. That’s a bit.

The first point, like we bring an additional capacity. I’m just not, I’m not saying just, I will increase your efficiency by X percent. So I feel a gap basically in their current capacities. That’s the first thing.

And I do that by targeting very precisely which type of clients I’m going after. Like they have a brand identity. They know how they want to communicate, but they have limited resources. And this is why my pitch resonates with them.

And then like going further, of course, like we need to have a strong vision because to, to be like what basically wouldn’t have been possible before. Generally, basically I just use my feature to build what Adobe has done before. Like my business will remain a feature business for. So.

Yeah, that’s true. That’s true. One thing we said in our founder coaching where you were also a part of at station F, and you already alluded to it. It’s probability.

So the big question with AI products is how can I make the experience for the customer good enough? Because I don’t have all the data. I don’t have trained the model best, like well enough at scale, et cetera. So this initial customers, how do I make sure the feedback is relevant?

Because I will have to make sure the product experience is good enough in comparison. When I just write software, it’s not that easy. I know, but I’m not a computer scientist, so I can say it. I know exactly what I have to write.

There are rules. I built the output fine, right? I test the experience. I actually want to add a word to this, the minimum, not viable, but quality product because Adobe cannot ship lower quality compared to you guys as a startup.

We have seen this before, right? We have seen this by, for example, word top of the line. And then Google came and says, well, it’s good enough. If I take a viability, which is lower, but quality is lower, like response time was not as fast and so on and so forth.

But like you got other things for it. Yeah. How do you define for you this minimum quality? What is it?

Today? Basically, we have this two step process. We tried the mobile. We test.

If we think it’s okay, we ship it to the client and then we see if it works for them. So they have the human in the loop. It’s a humanism. Yeah.

Is it? And that’s a good thing of our business. That’s at the end of it. It’s, it’s hard sometimes because you’re not design beautiful and so on.

Like, does it match what I had in mind? It’s a bit intangible, but at the end of the day, there’s a human in the loop and they’re able to say, okay, actually this makes a cut on us. But if, okay, got it. Human in the loop.

And how do you manage, like how, like you go to later on a venture capitalist and says, guys, I have the perfect solution. I save a lot of money and scale a high margin product. At the moment, you have a low margin, product that we just not scalable because you have the human in the loop. How do you measure?

What’s a key metrics here? The cremated for what? Like for, to make it scalable or like you have to write, make it scalable and as well, don’t have the cost, right? So at what point in time do you feel, how do you measure?

Okay. I, I created now a cherry logo and I checked it. I shipped it. That’s fine.

At what point in time do you ship it without having to check it? So basically it’s the cost of the human in the loop. It’s not that high for us because we have a recurring product and we sell in B2B. So to price basically the payback on the initial cost, if it was outsourced to like freelance designer that costs us like 400 euros per day, for instance, or 500 euros per day, the cost here would be paid back quite fast because it’s a one off cost for us.

And then the rest of the human in the loop job is done by our client. So if the, the initial check is done is good that we make the cuts often enough, then the human in the loop check made by the client is minimal enough. So they are happy with the output and don’t churn. Cool.

Okay. Now, so let’s say you have the lower margin. It really works and people are happy and you actually produce higher quality. Couldn’t the big corporations who are out there who have customers offer this as a service?

Yeah, well, if tomorrow Adobe puts a hundred million dollars on the table to do exactly what we’re doing, I think they will figure out at some point given the advancement of my product today. But first, when you look at just Figma, like one of the best companies in our space, Figma released their AI products. They had to roll it back because they got like complaints about like the usage of their own clients data and so on. So first, it’s not easy to ship an AI product for any comments.

Second, Canva, which is probably the most inspiring player in our space. They have a ton of different segments to address. They have all these small designers, these small companies that want AI to facilitate. I don’t know.

I’m a restaurant. I want to do a menu. They try to target large enterprises too, and so going more and more after Adobe’s clients. So they will have specific needs for the product.

They will need want to do some like graphic designs. They will want to do like video. So they have a ton of needs and they need to prioritize. And our advantage is our speed and our focus on the exact pain points of our clients basically.

And if I specify it, you’re basically saying they would need the feedback loop of all these customers, understand the feedback loop, incorporate it maybe in the same model or different models for different segments. That’s a huge, huge task. And you can just focus on your one product. Yeah.

You can focus on your one segment with your limited product with lower expectations. And from there, from the ground up, you can build it, which is basically how Figma also started and Canva started. If you think of exactly. And I think we should, and this is for everybody out there who thinks about, oh, should I build something?

And then things about like, but like the big corporations could eat your lunch. It is not as easy because you have a higher quality. Number one for all of your constraints. And number two, and you glanced over it very quickly, but I would like to add my experience here to this.

You actually said it ain’t that easy. And it is true. I just recently published an online course for AI and generative AI and it’s five courses. We roll it out.

And one of the things that I did there is we integrated open AI into the course. So the students have a co-pilot teaching assistant essentially that can use for analytics, for building, for coding, for whatever. There’s a lot of stuff they do with them. Now, the problem I saw was a student’s it.

Yes, everybody theoretically has used chat GPT, but using it in a way so that you do summarization correctly or coding or analytics correctly is actually a skill I needed to train. So I actually recently built a module just teaching the students how to prompt correctly the assistant. And that shows you, there is a per year. You have to jump over to get actually to the right setup.

So everybody looks at chat GPT and says like, oh, you’re a startup who said just wrapped around chat GPT. But in reality, it’s more complicated than that. Exactly. I think that’s one of the paradoxes of AI.

It’s very easy to get the proof of concept, right? So it varies you. Wow, that’s awesome. But when you want to productize and when you’re a large company, you want to productize for hundreds of millions of users.

Things become tricky. I think that’s, that’s very well said. And the second one, what we discussed a lot, Lutz and I is it’s also a different way of interaction with the product from an interface perspective. So when you think of products, we mentioned Adobe a lot, but they’re obviously others paintbrush, but there are a lot of buttons out there and you use them in a less iterative way.

You might scribble something, then you fine tune it, however you call it as a designer. But maybe your interface would look very, very different to that. You wouldn’t even show all the buttons at the beginning. Maybe never.

Maybe you interact, you talk with it in a totally different form. You are just upload pictures as form of input. Yeah, exactly. And basically, when we think of your Canva or Adobe, you cannot say, okay, let’s start from scratch.

Let’s just rebuild everything with these AI main flows in mind. It’s hard for them also to get the right UX. So, I mean, that’s another interesting point. Like mobile first companies were a thing for a while.

Exactly. So, I have a VC question. Go ahead. So, I want to ask about your mode because we were now from feature to product and company, but the mode is boring, right?

It’s just something around your castle and it’s protecting you. You’re in offense mode, right? You’re not defense. You don’t have a castle yet.

It’s maybe a small village or a house there. So, in ancient times, we called it unique selling proposition. But if you think of your company in the future, you’re going to have a lot of different things. So, I think that’s a really good question.

If you think of your company in the future, obviously, there will be something defensible, some mode that you’re creating because you did it in a special way. But how would you attack the others out there? And where would you eat someone’s lunch? Where is the unique selling proposition to customers?

Why they should buy you and not others? If you want to share it here in the podcast. So, who do you define as others? Like others competing startups?

Incubants? I mean, it’s probably larger agencies. It’s probably tools that you might replace partially. It could be even a person, obviously internally, but there has to be a reason why the 10x or maybe it’s just a 2x, but why you are so much better than what is currently being used out there.

So that’s convincing people and the mode is obviously more defending your position in the market. I think like the convincing people to get in, we discussed it a bit already. So, you know, like which specific pain point you tackle, like which ICP and so on. If I think like five years from now, if you go fast and you’re successful, you will quickly get a lot of data because of the typical modes, but that’s true.

And one advantage that we have, for instance, targeting only B2B is that we have clean data. So we don’t suffer from this like garbage in garbage out that can happen with like prosumer or consumer products. So this is one of the first things. The second thing is that you have a lock-in effect.

With clients, because when they have all your basically the model works well for that. Plus they have all their data stored within your platform. Why would they change? I think there’s a kind of virtual cycle of the zero because the more like this locking effect makes your product better for your own clients and also makes it easier for you to create a lot of value for new newcomers.

How is your feeling about agencies? Because I’ve been investing in a couple of businesses that were replacing human agents. That are so much so creativity, I think, will remain. I think I believe in the super creative thing because the cost basically for a very creative person to ship a super creative idea will be reduced.

But this is great. This is your strong idea for your new marketing campaign and so on. But in the day-to-day of marketing teams, there is a need for production. And here, I think one of the big pain points that we’re solving is that it’s easier to have the marketing guy think about, okay, what do I need?

And make their own visual than having a brainstorming or discussion between two human beings. And here, yes, it will be less creative, maybe less awesome in some cases. On the production part, I think the agency has way less value. I think agencies will keep thriving on very creative ideas.

Yeah. Could also be for consultants, right? When you think of slide writing and you don’t send it somewhere to design it. So Lutz and Ferdinand just left to have a, I’m pretty sure, a very, very good coffee in San Francisco.

So they left me here in Hort, Berlin. Summing up what we just learned, I’m pretty impressed by the approach of Lasko AI and Ferdinand and his team because it’s not really easy as you might have to tame these diffusion models and make them productive. So I think it’s really important in a B2B context. So having customers with pretty specific wants and needs and then a model trained or several models actually trained in a way that I can say, hey, these are my assets, my visual assets, and I want a certain style of output, quality, scenery, and the like.

So the approach they’re using, the thoughtfulness around the need, but also pragmatism, how to scale this with feedback loops from the customers, avoiding high costs, but also bad user experience because as we know, these models are not perfect. It’s probabilistic distributions. That was very, very interesting. We didn’t even tackle the part of AI agents because again, these are different models and also they are applying them in certain workflows and the agents play a certain role.

But yeah, just give Ferdinand a call and maybe he’ll tell you about it. We have other founders coming up with whom we will, we’ll have deep dives on the application of AI agents, which is quite, quite interesting and also very challenging. So subscribe and something will come up. Thank you for listening.