Founder and CEO, DeepL SE
[EN] re:publica 25: Jaroslaw Kutylowski - DeepL Dive – From hidden champion to German AI flagship
DeepL Dive – Vom Hidden Champion zum deutschen KI-Aushängeschild DeepL wurde vom Kölner Startup zum globalen ...
About Jaroslaw Kutylowski
Jaroslaw Kutylowski, CEO of DeepL, appeared on the Big Technology Podcast on July 8, 2026, to discuss the rise of specialized AI models. He argued that purpose-built models can offer better accuracy, lower latency, and reduced costs compared to large general-purpose systems, and noted that companies are increasingly using model routers to select the appropriate AI for each task. Kutylowski also highlighted real-time translation as a tool that could help businesses expand across borders, and described voice as the next frontier for AI. Kutylowski stated that AI translation tools like DeepL can reduce the upfront investment needed for companies to enter new markets by handling documentation, sales communication, and customer service in multiple languages. He described the ability for every person to talk to another person in the world as a "beautiful application of AI" that is worthwhile from both a business and human perspective.
Transcript (27 segments)
I
Interviewer0:16
Hello Republica, hello Yarik. In the best case we get to know each other a bit. Basically, there are three of us here: Yarik, me, and DeepL the company. It's a typical company in translation with a lot of aspects. Their main headquarters are in Cologne. But it's only one of five private, not publicly traded companies above five billion—a so-called unicorn. There's no classical Silicon Valley narrative; it's quality before growth. And a bit in the background, the founder has stayed a bit in the background. This interview is something really special, so thank you again for being here. As I said, I am here as a professional, as a researcher who has dealt a lot with digital infrastructure, data, and AI. And as a former supporter of digital technology, checking what successful development needs, but also privately—I have one parent who is not German, but I was raised only in German because in those days they thought that children could be confused by growing up bilingual. But a lot of things have changed since then. Privately, I have experience of exclusion with other languages because adults switch into them if you shouldn't hear something. So to come back to the topic, I did learn English after all, and when I became a researcher, that's quite international. I work a lot in English, and that's how I became a DeepL power user in the last few years. I'm going into this as a researcher, as a citizen, and I would like to channel the audience with my questions what they might want to know. So, translation as a competence, the diversification of language. The earliest texts we know are clay tablets from Mesopotamia which have synonyms. In 2017, when you founded DeepL, could you already foresee that there would be a market for translation which is steadily growing? And your private experiences with language integration—what do they have to do with your successful product development?
J
Jaroslaw Kutylowski3:32
Well, thanks for the introduction. I believe it's quite clear that language is a large topic for this world. That technology can help and how it can help is something that wasn't all that clear. And if we can be successful as a company, that's also not that obvious. Would you have asked me in 2017, I think I would have had a question mark about that. But the need for all of us to communicate, to talk with each other—that is there. And we as a company focus more on the business world, that it's very necessary there, and that quality and fluidity of language matter a lot for companies. That's something that was always clear. We just needed to check how can we develop the product and the company in a way that it helps people in that area—from somebody who is self-employed or privately on the platform, up to DAX, Fortune 500 company in the US that needs to internationalize or localize something. Privately, I was never a language genius in school, I have to say. If you would have asked my teachers in school—German, Polish—they probably would have never believed that I would have had anything to do with language in the future. But the topic is actually mostly mathematical, it's powered by AI. Nevertheless, with my Polish and German background, I think I understood quite quickly how important different languages are. And Cologne is close to France, so I think that also helps the employees at DeepL to understand what the problem really means.
I
Interviewer5:35
Since it's a digital product based on data, where does your instinct as an entrepreneur play a big role in how you design DeepL in its different functions? Data analysis—what can you learn about your users or other users, about what they really are interested in, how the product should work, and how language works in a professional context?
J
Jaroslaw Kutylowski6:00
Well, that's a very broad field. It depends if we look at a single user or if we look at companies—the needs are often different. If I use DeepL, you use DeepL, or someone of you, you want a quick translation, a product which is easy to use, something where you can actually work with the translation and maybe improve it a bit because each one of us is a bit special and you don't really want to take on what the AI has generated for us. But if you look at companies, the answers are different. It's about how can we make one shared, common language across the whole company. The company has a brand which is at best embedded into the language of all of the employees because that gives a homogeneous picture. And thus the research methods, how we find out what our users or customers need, are different. We do everything from A/B tests—we try to check what works well with our users—we do interviews, we talk with hundreds, thousands of people, but also the big companies—we have a vivid exchange with them.
I
Interviewer7:23
We already talked about Cologne being the main headquarters, but to show the distribution a bit more clearly—if I researched this correctly, more than 120,000 companies, institutions, and private people use you in a number of markets in which language works quite differently. Maybe you can say something about it—in how many languages you can use your product.
J
Jaroslaw Kutylowski7:48
I think it's 33, but it's growing a lot. We are putting a new focus on Asian languages. We started with Europe, but now the Asian markets have become quite interesting for us because the language barrier there is quite large. And the economic exchange between the countries within Asia, but also between Asia and Europe and the US, is quite big. And we have to see that we have local teams, that we have teams there on the ground that can talk to the customers, that can represent us, and that can also give us back the feedback—what works well in Japan, what doesn't. German we can check here, with other European languages also, but when it becomes a bit more exotic, it's also more difficult.
I
Interviewer8:45
Yeah, and also maybe not all the markets are structured in the same way. I can only say for my own use case from science—there are good glossaries for law or administration. Things are a bit different where the requests come from. So you are in Europe, we are at Republica, and you could play a bit bingo with the word digital sovereignty. There's strengths and weaknesses in founding here. GDPR conformity, EU regulations—all of these things are probably factors, also on your website in the outreach. What about trust in AI? Cultural questions—what role does that play for you?
J
Jaroslaw Kutylowski9:43
Well, it's as much a technical as a cultural question. That is demanded by society, by our customers, by companies, and the technical follows. Trust is very important when it comes to AI. I think the more powerful AI is becoming in the future, the more important it's going to become. Because depending on what AI does and how powerful it is going to become, the use cases are going to become more and more valuable. There's a difference if you want to translate an email quickly, check it and send it to someone, or if you want to translate a contract and sign it without checking it first and keep working with that. So there are of course very different rules for the different use cases. And that means what follows is—we as a company are orienting ourselves towards matching up to this trust.
I
Interviewer10:49
Yeah, you're a European unicorn for machine-driven translation. Maybe it's also valuable to differentiate—are you an LLM? Do you have an LLM? Are you in the same business model?
J
Jaroslaw Kutylowski11:04
Yeah, well, it's a complicated technical question, actually, because the way how we generated models and the AI models that we trained starting in 2017 over the years—they are not very different from the LLMs that we see nowadays. The technology, generally speaking, is quite similar. The details are different, but the area is the same. In the past year, we switched over our platform towards LLMs that we built ourselves. They're not generalist. They're not for everything. They can't do everything. They're actually built to translate, or if it's about DeepL Write, our writing product, to make language better or be a creative help when writing. So the technology underneath is the same, but it's trained for a different use case.
I
Interviewer12:08
And when you check infrastructure a bit more, we already heard a bit about the technical specifications. There is the recommendation to have a model that's also financed publicly. With you, it's a bit more interesting that you chose to be financed through international VCs. You didn't take any other money. Was that a strategy or necessity?
J
Jaroslaw Kutylowski12:35
Actually, I have to say I am a fan of the private market for financing. I believe once you have a good idea, once you have a good team, and once you can actually prove that more or less well—proofs can be different depending on the size of the company, sure—then you should be able to find your financing on the market. If you can't find it, it's also a statement how good you are placed as a company. And I think the markets work quite well. VCs are experienced in making companies grow and help them grow. And that was a conscious choice. And we also consciously chose the investors that are a part of us. We picked well. We tried to choose consciously. It's actually similar to hiring employees. You want investors that are good for you, that have similar values to your own company, and perhaps also have a similar style, because possibly I need to work with them every day on the board. Well, not daily, but work with them. And you want to have a person that you can trust and with whom you can work together well. And when you look at that on a global scale, you have to say the investors come from Europe, from the US, a tiny part also from Asia. The market is quite big, and if you have a good standing as a company, you can make use of this market quite well.
I
Interviewer14:36
That's the financial resources. We already heard the digital minister this morning, and I think the plan is that incentives should be created for more founders like you. So let's talk about other requirements for this. Maybe you can say more about the requirements for innovation or the prerequisites.
J
Jaroslaw Kutylowski15:04
Hardware is a question of money mainly. And the big hardware producers—well, Nvidia is the leader there. You can't get around them. They're usually happy to help a company if you bring the necessary financial funds. So I think financing is the most important question in the beginning. When it's about the side of regulations, yes, you can build a company also in Europe. We're a great example for that. A bit more regulations or bureaucracy in Europe does not impede that, but you do have to think—the more a company is subject to regulations, the more it will probably go towards other markets because there innovation or progress is easier. And AI is one of those fields where you have to stay on the ball the whole time, or else it doesn't really work and you're falling back.
I
Interviewer16:21
Since we've all come together here, and maybe some of the co-workers of some of the ministers that are still here—is there something that you would wish for?
J
Jaroslaw Kutylowski16:29
I think my biggest ask would be that technology or the tech industry should be elevated to a higher level. We've always said we're an industrialized nation and we have a very export-oriented economy, but we definitely need more technological companies for the future, and these companies need to be enabled to apply AI in the future. Thus I would ask that it should gain some more spotlight—what we build in the realm of AI and around that whole ecosystem—again, some more public exposure within society. That would help a lot. And I think France is really showing how it's done. Paris, for example, is now viewed as a technological epicenter. These days it's also kind of a cheap way to have some effect. You don't need a lot of money. It's mostly about publicity, about PR, about the press, and you can still achieve a lot.
I
Interviewer18:01
I would, as a scientist—with the expected boom and since growth and scaling isn't always the greatest thing, there's always pressure there. You've doubled in size since the last 15 months. How are you dealing with that expectation and the pressure that is upon you? You've not been traded publicly yet, so...
J
Jaroslaw Kutylowski18:29
Well, I mean, we're not traded publicly, but that doesn't mean that there's no pressure. There is always pressure. And it's not always clear how to do this. I still feel a lot of responsibility for the investors and the money that they invested in us. We usually talk about invested, but in the end—this could be our pension system that is investing in us, so we're in the end responsible for pensions, possibly. And we speak to everybody, to customers, to interested parties, other stakeholders, and that is actually hard for me. But as a founder and as a CEO, that's something you just have to do. It's part of the job. And it's nice to see myself grow in this role and to walk that path. And I actually would recommend it to anyone who's interested, but don't fool yourself—psychologically, emotionally, it's not easy.
I
Interviewer19:32
And what does it feel for you that you're the AI poster child of Europe?
J
Jaroslaw Kutylowski19:38
I mean, clearly that doesn't alleviate any of the pressure. It just makes it worse. But obviously, we're proud of it and it's nice to help out and to be part of that process. And if I hear from the other founders that they read about us and they've seen us—'you were courageous enough to go up against the big companies in the US in the technological arena'—that makes me proud. That makes me glad that we can give something back to the ecosystem, that we can help them out.
I
Interviewer20:20
Are there also connections to politics, for example, or to other countries that you may need for building up infrastructure? Has that—is that something that you've grown in the last years?
J
Jaroslaw Kutylowski20:38
I'm not too active in that area, but clearly the company growing in size has become more visible. But I've always enjoyed being a small company. There are a lot of small companies, but also as a startup—we just know that a lot of them die, and that's okay and that should be that way, because if that's not the case, then we're not daring, we're not building risky things. So not everybody can deal with all of the small companies, but the moment you grow, you become more visible. And we actually do like to step into the spotlight, and we want to help Germany out and Europe to become stronger in this arena.
I
Interviewer21:26
And sorry, my time is running out, but I need to be quick. And I've still got two questions that I want to ask. One is very dear to my heart because I'm really a proponent of open source and that's very important to me. You're using data that is publicly available in a freemium model—for example, company home pages, maybe social media networks—but you have not published any open source models. DeepL is a proprietary product and it does have APIs, but there's a limitation about the features and it's not very open. So why are you actually very strategic about having it closed? And is there no aspect to growth if you would open it?
J
Jaroslaw Kutylowski22:17
Well, the market that we're in is dominated by big players that have a lot of power in that market and that have gigantic resources for research. And it's not just about the research, but it's also the market entry that we're looking at. Every company is basically tied to Microsoft, Google, Amazon in some way or other, and these other players we need to hold at bay. We need to be very deliberate about what is good for us, and we need to see where we can still ensure the economic success that we need. And if we would really open source our models, then this would be very challenging. And that could lead to a point where they reach feature parity in translation, and then they would just switch to our open source model that we put a lot of research and money into. So this would be very difficult for us. Whereas if we have a proprietary system, we can be very deliberate about what product we offer to the market, what the limits are to the usage—for example, how many words can be translated every day. So this is just economic calculus behind that. And it is important to be clear about—if you do research at the forefront, then you need economic success. Otherwise, there is no basis for the future in order to reinvest and to stay at that frontier.
I
Interviewer24:09
Okay, so now let's look back at the very large questions of society. For example, if you have geopolitical changes, the role of languages are changing. You're working with text at the moment. There are also technologies that work with audio. Are we losing some understanding of nuances and the finer points of language the more we hand that over to the machines? And what is your vision for a world where there is real-time translation, for example, the Babel Fish from literature? Where does that lead us?
J
Jaroslaw Kutylowski24:53
We are actually doing that already. With DeepL Voice, we can translate audio, and that is actually a little bit more impressive even. For example, when I sit in a meeting with a customer in Japan, I easily can follow what is being said, and I can talk to the customer. There is no barrier, and that is absolutely impressive. I find this really fascinating, and for the global economy, this will be another spark of growth, and that might work against the closing of the markets that the US is currently leading, and that allows for more cooperation that we all really need around the world right now. And whether I believe that Babel Fish—or DeepL Voice as we call it—do I believe that it will take over all of communication? I actually don't think so. I still enjoy talking to my old colleagues from school back in the day in Poland, to talk to them in Polish, and it still allows me to connect to the culture. It gives me understanding. It shows my appreciation for other cultures and countries. And I find this really fascinating—that this technology for translation that we're building, we're actually helping people keep the teaching of languages in their own countries, because otherwise some societies are under pressure because they're very small and their language is not relevant enough. And for example, my kids by now speak better English than German, and that may not be something you want. Whereas if we help them keep their local language, then we can actually help them out.
I
Interviewer26:51
So, thank you very much. Everything is about communication. But sorry, we won't have time for Q&A, but maybe you're still around and people can ask you questions. And people now know a lot about DeepL, the mythical origins, and also a few numbers. And I'm very grateful for you taking the time to talk about this ever-changing and very dynamic technology. I would have loved to go towards how the labor market is moving, and I heard that you're working together with professional simultaneous translators, and so improving the collaboration—so that may be a very nice high note to end on. Thank you for your time, and thank you for this talk.