27:06
I wanted to introduce our next two guests, Anastasis and Anatoli, if you want to come on up here. Um, so I'm I'm really excited about uh this conversation. Um, these two companies have some things in common, but I think they're they're actually quite different. Um, including sort of like the stage of life in which they're in. Um, anast I'm sorry, is it anastasis? Anastasis. I'm the sass. Thank you. Um, artist, engineer, researcher, co-founder, and co-CEO of Runway, uh, video generation AI company that was founded in New York in 2018 and has become very highly valued by its investors to the tune of more than five
billion dollars. Um, and we have Anatoli Kazanov. Thank you so much for joining us, too. uh also co-founder and co-CEO of Holy Water Tech, an AI first entertainment company founded in Kev uh in 2020 and has raised $22 million from investors including Endeavor Catalyst and Horizon. Um I'd be thrilled to sit down with either one of you individually. So I'm so happy to have you both here. Can you just talk for a few minutes because people are hearing like AI and video tell the crowd a little bit about how your companies are different and what they're doing. Um maybe Anastasis if you want to start. Yeah. So uh Runway for those who don't know is a applied uh research company.
We focus a lot on video generation and world models. Uh we've been around for seven years now. Uh so when we started actually AI video was not quite a thing. It was a thing in our imagination but not quite. uh there was some very early video generation but uh obviously over the past few years like the quality and uh kind of performance of those models has completely dramatically improved. Um and you know a lot of what we try to do is figure out like how do we both make those models better? How do we train those models from scratch? We build kind of foundation models based on visual data uh but we also deploy them to products. Um and those products are used by a wide range of uh filmmakers, by uh marketing teams, but also increasingly
from gaming companies, robotics companies. Uh so we see AI video as a very general category of models. Uh and I think you're underelling it. This is a very buzzy company with big ambitions. Uh and tell us a little bit about holy water, please. Yeah. Actually, we're using runway to create EI movies and they perform very well. So basically uh what we discovered that people want to create content they want to consume it and AI lowers barriers to entry. So like anybody can be filmmaker, anybody can be author, anybody can create comics and we build this kind of platform where we sell and distribute hundreds of
different type of content. We started from books then we create AI movies and we create live action movies and we're biggest in short drama space. We have more than 100 million users and I really believe that in few years it will evolve towards Roblox where you can create content consume it and our goal is to build on top of foundational models not to create ones but to build this kind of orchestration of models where you can create any type of content you want. So both of your companies have evolved pretty dramatically since you started them. I mean you focused of course on AI generated video. Now you're sort of focusing a little bit more on role models and getting more into robotics which I think is really interesting. as you mentioned, you
started in romance, I think comic or book publishing. Um, and now you are working with, you know, studios in Hollywood on these sort of micro dramas, which we're going to get into. But, um, you know, Elana asked the last panelist this, and I think it's it's a great question, but when you're starting a company, you know, there has to be sort of an insight. You have to sort of, uh, see white space perhaps that other people don't. So in your case Anastasis you leaned into video generation at sort of a very early time. This is 2018. This is well before you know chat GBT sort of you know blew everybody's hair back in 2023. What sort of need were you trying to solve at the time?
Yeah. So I think obviously today the applications of uh of generative models are obvious and there is uh but at the time it was actually generative models were considered to be a bit of a toy back in 2018. Um but we a lot of the AI of the time was about classification was about uh you know uh models that recognize specific items in language or models that recognize specific objects. Um and the idea that generative models which at the time could only generate you know very low resolution um not high quality images would get to this point where you could generate entire you know entire films uh entire campaigns using
those models uh and the those models will be able to simulate the world. Uh it was a very it was a bit of a far-fetched idea. Uh but I think for us like the company started in art school uh was is not where AI companies tend to start. Uh but and a lot of the initial motivation and vision was it was very clear if you look at the trajectory of how like the technology was improving that at some point most of uh what we uh creative tools would dramatically change as a result of uh AI uh that being able to generate uh images, videos um kind of content uh will dramatically accelerate how quickly you can iterate and test ideas and turn ideas into actual our outputs uh and as a result it will change like how we build those tools. Uh and so
a lot of the original vision was let's um let's figure out like what does it mean to build creative tools assuming you have this AI kind of superpowers and assuming you have this these magical models uh that are able to kind of generate something from very little uh input. Anatoli yeah I mean we started business in Ukraine. In Ukraine there is no venture capital, no money. So you need to learn how to be profitable and consumer apps is like the best way. You can start fast, you can build fast, you can distribute and that's how we started. We started to building I would call it like random utility apps. We made it profitable. We scaled. But then we decided it's not really our ambitions to
build those small application. We exited sold them and we found this kind of content ecosystem where we saw a lot of potential. We saw as well how GPT2 is evolving how uh AI models as well evolving at that time and we started from book kind of content and that help us basically to build expertise build distribution and I would say it's also about like pivoting at the right moment because uh you can see how other market are growing how technology is evolving and uh I think like this culture of pivots I think we did like three times uh at holy water and each time was extremely important for our growth since we've been growing 2x year on year for like from the start of the company.
You know, I kind of wonder how fixed or not your mission or your north star is because tech has evolved so much since you started your companies and it's continuing to evolve. So I wonder how you think about things. I mean, how do you sort of are you constantly sort of is your mission in flux or do you know what the end state is that you're going for? Um, I always like the quote of, you know, the best way to predict the future is to invent it. And I think that was a lot of like our philosophy at Runway is, um, like to really try not to just push, you know, uh, you know, grow the
company, but also develop this technology and figure out like what is it like what does it mean to actually build tools on top of this technology? What does it mean to like those how does this technology democratize um, kind of human creativity? Um I think planning is difficult in AI companies. Uh I think from the very beginning we did not believe in very we believed in having a very long-term vision but very short-term execution uh and figuring out like what is you know what is the next stepping stone that we need to uh succeed at in order to actually further uh grow the technology grow the company.
Um and so that has always been the mode of operation of the company is you know when we started we did not have you know the if you start an AI research company today you immediately might be able to raise a ton of capital. We did not have the luxury of that at that time because nobody really believed in generative uh generative models. Uh and so we needed to be incremental in how we developed the product how we found like applications at every stage of capability of the technology. And I think that built a good muscle where whereas now you know the field is moving so quickly but it's a mode of operating as an AI company that we had to enter very early on before um you
know back in 2018 where nobody was building tools around those models. Yeah. Uh I mean for me I would say it's about just like persistence. So for example when you believe into something and it just matter of time if you persist enough you can achieve basically anything and in our case that uh we already saw that how AI has so much potential in terms of content and uh we just like put that idea and we actually started creating AI movies two years ago and even though the quality was not there yet and uh we just persisted we believe that it will evolve and we experiment we were finding what work doesn't and then I was like the vision became more clear where it's
going that u basically anybody at home would able to create like Hollywood movie and or like anime or comics and that kind of content would reach like 10 or hundred of million of users. So and at some point we saw okay that's how the future will evolve and then we just decide okay we will try to put as much effort as possible and see how it happens. Well I'm sure this is probably true of you know a lot of people in the audience um but especially in your respective businesses and it doesn't matter almost how much money you have you probably have more ideas than resources to apply against them. So how do you think about R&D and prioritizing the many different things you want to do?
Um we intentionally at every stage of runway we've we've uh we've intentionally try to keep the team smaller than it should be is one principle that we always follow of like how do we uh like there is the luxury of having too many resources or too many people is that you end up pursuing too many ideas at once. uh and it doesn't force you to like you know really have conviction over the idea or like the approach that will succeed. Um I think prioritization it's it's always like a mix of uh kind of intuition about what the technology is and what products are enabled and actually work with this generation of technology. Uh and then a lot of experimentation. Um, one of the nice things coming from, we came from
this um, this grad school program that was kind of at the intersection of art and technology and it really encouraged this idea of like prototyping and demo culture um, and like you know not trying to convince via long uh, specs and PRDs but trying to convince by actually building the thing uh, and seeing whether it actually works and I think it's always very helpful like to and now I think with VIP coding that has become I think more of a practice of just you figuring out like what does the 80% version of this look like and that's that's always very helpful I think to get that intuition of like can this actually is this actually worth to like maximally kind of pursue yeah I mean for us because we have
this culture of pivots we launch multiple consumer apps many of them become very successful and biggest in their niches so we build this kind of like small lin teams where we try to hire best entrepreneurs and basically bring them inhouse give some them like equity or some motivation that tie to the product and with all the AI agents what Anashia said that you can have relatively a small team of like two developers one like I would call like product engineer or product manager with vision uh and they can basically build the product in like two to three weeks launch it uh try to find unit economics product market fit so and we try to keep that structure present where We have those kind of like founders inside of
holy water who has stake skin in the game and that's how we managed to launch so many new product and actually make them succeed. Um well you know despite the advantages in the tech there are always challenges and I wonder if you wouldn't mind sharing for again the founders here and for my edification um what are your biggest sort of challenges right now? What's maybe not working as well as you would like? So at this point we've gone through maybe uh four or five different generations of our base models. Um and it's been interesting that like we've we've almost um with every generation of
of those models the challenges are very different and we need to restructure how we think about our research, how we think about the product. Uh I think right now the biggest challenges that we face is we're we're now in the process of um expanding the company uh from the video generation which is purely you type in a prompt or you provide some image and you get back a video uh towards simulation and towards this idea of world modeling. Uh so something we've long believed is that video models like video gen models that generate video also understand the world and understand like you know gain an intuition about physics gain intuition about like how humans perform different tasks and those
models can be useful beyond uh artistic expression and uh more towards actually uh systems that can drive uh robots and they can drive autonomous systems. So, a lot of the challenge that we're facing now on the research side and the product side is how do we actually uh evolve our models um and build the right kind of commercial structures in order to sell the same kinds of models that we've been building for the creative industries to um to companies that are working robotics um or companies that are building kind of infrastructure for uh for physical AI. So, so have you hired just a like your sales team must be very different or I mean that process is very different talking to Hollywood studios versus robotics companies where you're trying to strike
partnerships right now. Completely yeah completely different uh go to market motions. So it's it's a nice it's an interesting um it's an interesting structure where you know if you look at the models themselves kind of 90% of the compute that we spend on building those models is shared between the stuff that we do for film making versus the stuff that we do for uh gaming and advertising with the stuff that we do for physical AI but the way we sell uh those models and products on top of those models is completely different. So we've we figured out ways in which we can build kind of small um essentially tiger teams that are focused on a particular vertical and try to figure out very quickly iterate and kind of work with
different partners to figure out what kind of commercial structures actually make sense as you're working with all these new verticals and industries uh and learn as quickly as possible. uh which is uh but some things are very common uh between the two like we have you know forward deployed motions across both our um kind of uh our Hollywood studios that work with runway uh that we have with robotics labs uh it's just like when we work with a Hollywood studio we embed our creatives and artists and help them with a particular project particular production where we work with robotics companies we kind of bring researchers kind of on the same table as those robotic sims and figure out like how do we post train kind of our models
to fit your specific hardware or specific tasks that you want to solve. Uh so quite different but we figure out some like common aspects as we evolve the company. Can I ask how many employees do you have right now roughly? Sorry a number of employees. Uh so we're around 160 people 160 just okay. Um and what are your sort of what's a challenge for you right now around like so basically because we started this kind of like preI era right now we have more than 200 people and my main like challenge how to make this team AI first team so how because right now like people can 10x 20x do their performance and uh because of that you can basically out compete Netflix any big Hollywood studio because your team
is just more efficient and that's I would say my biggest challenge how to transition the team properly from like using AI towards becoming operators of AI agents where they can like anybody can build agents each agents can like basically replicate their work how they can build twin twins with AI how to build this kind of knowledge access system where you can have like knowledge of the whole company and people can just let's say focus on outcomes and instead of focusing on input or hours they put into work. So I would say that's the biggest challenge and the biggest growth for us. Basically making those like 200 people work like 2,000.
Um you know I think another thing that's obviously very interesting with AI generated video uh is the risks the deep fakes. Um how are you both thinking about I guess sort of safety and these sort you know kind of ongoing challenges Yeah. So, you know, h having put out to market um I think it's the first text to video uh kind of tool out there. I think we had to deal with this challenge very early on and I think we had to evolve a lot of like our policies and how we thought about things like moderation. Um especially as the models improved quite a bit. Uh we have a trust and safety team that focus a lot on like different aspects of you know both like harmful uses of those models but kind of
using uh trying to generate content with public figures like all those using those models generate uh known IP um and especially because we're uh very enterprise focused so we kind of working with all those uh large companies uh we had to you know it's it's very important to them that they feel that the platform kind of u has a strong policy around um kind of what it allows to generate in the platform and what not. Um for better or worse like I think runway has built a bit of like reputation of being overly conservative in its moderation. Uh but I think it has it's it's been the right kind of approach because it's kind of a lot of the risks of those models can be unknown. Um and we need to discover them
and kind of figure out them as we go. Um, and a lot of what we do also is like with every new model, we do a ton of kind of red teaming and a lot of preparatory work to figure out like where like to kind of understand kind of mitigate potential risks ahead of actually releasing the model. Yeah, I mean for us we don't train any models. So we just rely like on foundation models on their policy but we also have moderation in house QC to make sure that we are consistent with quality with policy with all the things that is happening but again our goal is mainly on orchestration rather than training or building foundation models um you know on that front I mean again very different businesses so you're
building models you're you know relying on other people's models you need tons of compute Um, not everybody in this room is in that position, but there are probably people that want to be in the position of building a company like yours. What is it like? I mean, I guess from a funding standpoint, um, it almost seems like as soon as you raise a round, you need more resources to kind of feed the beast. So, how do you think about those conversations? Yeah. So I think to start off we are big believers in scaling loads then the more you scale the data and compute uh of those models that the better the performance of those models will be and we've we've seen that again and again as
we as we scale those models. Uh at the same time there's a lot of research techniques uh that you can use and the way you think about and ways in which you work with the data that can make that training a lot more efficient. Uh so as one example like our latest model uh gen 4.5 um you know when we uh when it was released it kind of topped the leaderboards and it topped the leaderboards against you know the models that uh Google released or the V v3 was Google's model or like uh Sora uh Open AI's uh video model um and it was not before because we had more compute uh we are definitely disadvantaged in terms of compute uh we're probably 100x less compute and like 10x smaller team that's working on those models. Uh so a lot of it is has
to do about focus about kind of data uh really um really pay attention to how you uh you train those models in an efficient and data efficient manner where you can learn more from less data um and really the taste of the research team. The ability that the fact that we've been doing this for many iterations that we built all this like internal knowledge of how to train those models effectively. Yeah. I mean for us it like human costs more than compute we spend on producing content. So and it's like 10x more efficient and because it's predictable cost we can pass it to our consumers and we scale our content once we are confident that we
can like make it break even. So compute doesn't really affect us again because it costs more to produce live action than create a movie. Um, you know, on that last point, Moes, I mean, everybody's, you know, I'm sure you get asked this like five times a day, but, um, you're both up against, you know, a lot of, you know, big companies. You especially, you're going up against, uh, VO, you're going up, you know, Google's got video generation, it's got world models, it's a five trillion dollar company. Um, what is your sort of sustainable advantage? Um I think in AI the only real mode is the team and the culture of the company.
Um like model modes are very temporary, data modes are temporary. Like we expect every model that we release uh is not going to you know even if it if when it we release it's it's the best model available that's only going to last for a few months and we need to get to the next model. uh but the ability of the research team and the to continuously innovate and our like affinity with uh with the verticals and the industries that we operate such kind of you know for example with the creative industry um I think those are the real modes and they're more about the team rather than the product at any given point in time uh so as and you know if you're a researcher you want to you know you're not just motivated obviously by
by uh by comp and like the you know the perks of joining a company but ultimately you if you're a good researcher you have the choice to join any uh any research team. Uh so what do you look most uh in like a team that you join is like who else is part of that team like what is the culture of collaboration uh and what is the focus and the vision of the company uh like if you look at runway we've been focused on the same thing and like and those mo and we've been living those models for years and years uh and it's not like a side project for us it's like the bread and butter. So if you really care about making an impact in this field like it's uh it's a good argument for us that you know for joining runway. Uh
so ultimately I think that's that's the only real mode is like the team and the culture. So that's interesting. So for all this advanced tech it kind of comes down in some ways to human relationships and culture. Yeah I think so. Yeah. And I think the one of the luxury in caris is like we're kind of very we're very early on in this field and I think for a long time I think it was we were not a very um we're not working on a very popular field uh of kind of generative models. Uh but that allowed us gave us kind of the breathing room to basically evolve the culture and figure out like who are we as a company. Um and I think that was really helpful and it's it's very difficult I think if you're starting today to just both like
you know compete at the pace at which the field is moving and at the same time step back and think through like how do I actually want to build the company what are the values in which are important uh for how we work together and like how we collaborate. Yeah. Uh I mean yeah I definitely agree with Anastasia uh because for us we compete against Hollywood and I would say those companies have like ten of thousand or even hundreds of thousands of employees and I said they really bloated they have a lot of process bureaucracy and for us to basically out compete them it's about how efficient you can build your team for example like I personally run like four agents right now that do some part of analytics some researches while I'm speaking at conference and I think
That's how you can out compete them basically being like 10 20x more efficient with smaller team and with less process with less bureaucracy and just moving much faster and with all the AI tools that available right now you can actually build that very fast and that's what we are seeing already how fast we are growing in our entertainment industry and how much we can achieve with small team in terms of creative part in terms of marketing part distribution So yeah, I mean there's so much that you're doing that's not traditional from, you know, using these agents to the size of your teams. One other thing that I think is interesting, you're both co-founders and you're both co-CEOs, which is something
that we haven't really seen that much of historically, or at least when it's happened, it's been I think we've both been CTOs as well. I'm sorry. We both we've been both CTO chief technology officer. Um yeah, just wondering about that if it that's sort of like a permanent thing or uh how you think about that or if that is some reflection of how you're building the company. Yeah. So the way um the way we think about runway and I'm curious to hear uh your take as well. Um I think uh this really this change in role happened a few months ago and it was really kind of reflection of like how me and my co-founder coco Chris have been
uh working together for a long time. Uh the reason for the timing was um one of the you know as the company is expanding towards uh world models and uh and those new industries um like one thing we like to say about runways like it's many companies inside one u and it became clear that like if we wanted to kind of capture all the opportunities across both the current parts of the uh product and commercial uh traction that we're seeing in the creative industries and on those new areas of the business uh that I needed to uh also be more involved in the commercial aspect of kind of uh of building out those functions. Uh and that was kind of the motivation for that for the for that change. Uh but in practice not much has
changed I would say. Yeah. Uh actually I would say very similar for us as well with my co-founder Bton because we have again multiple basically like separate companies inside Holy Water. one for books publishing part from AI movies short dramas comics anime and basically I've been involved a lot into like product side operational side and been basically managing them and we just decided to split like responsibilities he's more responsible for fundraising investor relationship for uh admin functions and I'm responsible more for product side for operational side and basically managing that our business can grow and all our like small companies side can grow fast as well.
Um I know we're out of time. I wanted to ask too sort of on behalf of the audience uh you were born here. Uh you were born in Ukraine. You both have business dealings obviously in the US Runaways headquartered in New York. Where is Holy Water headquartered? So I mean in Cyprus but we also have office in Portugal, Poland, Ukraine as well. Yeah. Okay. I just wondered how important the you know I guess ties to the US are and how people should be thinking about sort of like their own land and expand strategies in 2026. Yeah. Um so I think I mean for Runway's case uh I think the company could not have started anywhere in any other city in the world I would say. um because it's not just us, it's New York specifically as this intersection of so
many industries and intersection of technology and the creative industry. Um we came you know we when we started the company we had zero network um zero um kind of we had to basically like it was not on the basis of kind of our credentials that we raised our first rounds of funding and I think that's always been kind of very unique aspect of the US is that you can um you can you know you can start something with like very little uh very little kind of evidence of kind of past accomplishments just on the basis of like what you're building and what you're putting out in the world. Uh I think the VC climate is somewhat different between the US and Europe. Um so I don't have experience
kind of as much with kind of European VC world. But I do think the long-term conviction and kind of bigger kind of riskier bets. I think the US uh kind of VC world is a bit more um uh more keen towards kind of uh placing uh bets there. Um but ultimately I think today like you can start a company from anywhere and I think you can really um choose to expand to the US uh kind of later on. Yeah, I mean for us I think it's also about network because we are producing movies. We are competing in entertainment. We definitely need to have some presence in LA and that's how we managed to sign Fox Entertainment being there been uh talking to them and
building this kind of network and also 60% of 60% of our revenue comes from US users. So you definitely need to have also understanding of culture having team there having like script writers showrunners from here. So I would say it's definitely very important and the amount of like networking you can get there I don't think it really other like any cities that can compete with SF uh in terms of people talent networking that you can get and very fast. And last question, and it's a little bit corny, forgive me, but um I think for both of you, these were your first companies. So, what do you wish somebody had told you? What advice can you share that you know now that you didn't know then?
Um, I think if I look at like most of the mistakes we've made and most of the times we weren't making as much progress as we would have liked, um, it's been because we uh we pursued an idea that seemed kind of the most likely to succeed. Uh, versus an idea that we deeply cared about and we saw a lot of excitement internally with what we were building. Um and so I think listening to our kind of not just to like our um kind of intuition as founders but also the internal momentum of like a new idea and a new product a new tool that we're testing uh instead of listening to the this thing that makes the best narrative externally. I think that has always uh that has always worked well for us. Um
so if I look at you know for a long time very early on the company we're building this uh this video editor product on the web which sounded like it was kind of a very clean narrative around you know you can uh bringing like a traditional video editor to instead of having it like a native application making it a web application and adding collaboration on top of it. It sounded like a very clean story versus something like, you know, those new kinds of generative models that um that were very difficult to explain at the time, but there was so much excitement internally about using those models, testing them constantly, sharing outputs. Um, and I think every time we listen to that in like the
external world kind of figures itself out and eventually becomes synchronized with our internal uh temperature. Yeah. Uh I would say like it's all about team. If you can hire right people, you can get the best people and you can basically do achieve anything and you can pivot as many times as you want. And I think because it was first company that I founded and uh that's probably what like biggest learning that I took. If you can get the best people uh and persuade them to join you, no matter what failures or issues you have, you will overcome them and you will become much stronger. So that's probably like biggest uh learning that I took and I'm trying to do it as much as possible right now like
networking with people building the pool of people I want to headh hunt and I think like that's biggest part and second biggest part I would say it's about like building also relationship with investors with smart people just to have them and have them to challenge your ideas because at the end I mean you can learn from your mistakes or you can learn from others. and building this kind of network really helps you to learn from others mistakes much faster and evolve much faster. So those are the biggest learnings for me. Learn from those who have seen it before. That's great guys. Well, thank you so much. Really pleasure to meet both of you in person. Thank you for being part of this event. We really appreciate it.
Um thank you all for joining us. Thank you again team Endeavor for being just amazing uh and supportive of us and uh the rest of the community here. Thank you guys. Have a drink.
