How Higgsfield Reached $100M ARR: Alex Mashrabov with DVC's Nick Davidov & Marina Davidova

We use AI to make educational content about AI.

Very high bar. The goal is to build a successful company. Winner takes it all.

Last effort was 50.

What were your criteria?

The group was really small. Since then, over last 10 years, I work in this space.

Good to be Switzerland. So, this is not about time. This is about attention.

And your team is small, right?

Which is the best. And Meta is likely to be a loser. Welcome to our podcast. So, this is Deconstructed VC. We're trying to deconstruct companies, you know, like lots of engineers when they're children, they like to take things apart, radios and other stuff. We like to take businesses apart and try to reassemble them back together and see what spare parts we got

and maybe inspire other first- time entrepreneurs to also do startups. and also maybe inspire somebody to join your team. So today we have Alex Mashabuff. Is it Mashabuff or Mashabuff? Both works. Both works. Okay. That that's you know how how you tell a founder is humble. So Alex Mashabou is CEO and founder of Higsville AI, an absolute breakout company that is taking the world of AI video by storm. and also other weather effects

as Nick mentioned usually we deconstruct current businesses of our portfolio founders but your story is very inspiring even before that and we maybe can start like from from from the beginning from from the previous journey.

Oh yeah. So you started I think the latest I found you worked at at Yandex and you were working on machine learning for natural language and for search like beam search that was a pretty a pretty cool thing back then. now you know all of the AI agents when they search they look different things and they try to split queries. but you you did it way back. how was it at Yandex and how did you how did you end up at Yandex?

This thank you very much. It's it's a great pleasure for me to be here. Yandex is a natural choice for all the X ICPC guys to join the company because the code base is written in C in C++. It requires really low latency and people with like Olympia background can really thrive there. So that's where I started. I started really in the space of like fixing misspelling in the search queries. So it's an overlap between language translation and search. Interestingly, I very quickly saw that most talented people across both departments are very interested in something new. Not like cut boost

but really neural nets. I tried to be part of this community. the group was really small like maybe tw up to 20 people. back later in 2017 2018 this technology was heavily used for Alisa and now these people are primarily out of Yandex and working for anthropic meta and open AAI. I think I didn't realize back then how neural nets are powerful. In 2014, it was not clear, but we got some breakthrough results where we were able to surpass Google language translation from Russian to English with neural nets while it wasn't possible with gradient decision trees. And then I think the major like enlightening moment for me was in 2015 when all these systems for face recognition got deployed widely in airports in in banks and ultimately those systems are trained in a way where they learn to distinguish AI generated contents and real contents and that propelled the space of AI generated imagery and this completely took over my and then since then over last 10 years I work in this space.

So before doing AI factory you worked on another startup first. So how did you be become a founder?

I think in Yandex it was not fully clear how the company wants to transition to neural nets and I think even back then many people were leaving the company. Some of the people who were pioneering neural nets they were probably not happy very happy as well. So I thought I need to look for opportunities outside. I think back then there was it was relatively clear that the space of the mobile apps

is growing. I I think back then it was relatively easy to get to a few million AR and like cash flow pots if with mobile apps and we decided to do some of that. So the first app was which we developed back then I think the group the team was just maybe 10 11 people and the app was very simple it allowed let's say schoolers to take a photo of their homework

and get solution that's where I learned the first rule of venture really and the and there and the rule is very simple winner takes it all

the company called Fatamath

the company from Croatia they were number on and let's say our app was like number two or number three in this category but they were at like they were probably 10 20 times bigger than we were and that showed us that this asymmetry which is created in the technology markets where like winner takes it all and I think we essentially in 20

can we can we maybe dive into this concept like why does winner take it all

I think I I am I'm not very good in in in economic theory I just think this is this is just sort of an empirical rule like I think this happens with all the markets even like Uber versus Lifts is another example and I think the only sort of an exception here can be probably cloud markets where Amazon has isn't in dominant position but has less than 80% of the markets that's probably the only exception but probably the reason of this exception is that Google and Microsoft were burning cash to really catch up so they because they could do that in venture I think it doesn't work this way

I think also what might work there is that this companies that are already first they have momentum and it's cheaper for them to attract capital it's easier for them to hire better people and it's cheaper for them to do marketing so a lot of like word of mouth goes like so I think this this is true this is really the case in consumer so what happened to to your app like you you felt it was too expensive to compete with photomath

photomath was doing very well and I think we kind of come to the came to the point where we have to go beyond that. We had several M&A opportunities although it felt like there is an opportunity to kind of to expand and there were several directions. First is there is an app called Google Classroom.

Mhm. It took really well in 2017 and it felt like oh there should be an opport and and there was a I think it was back then it was was Obama administration they pushed really hard with iPads and Chromebooks in many schools and really make the whole process more electronic at least it felt this way back then and we were trying to see if we can be part of this ecosystem the answer is no we we we couldn't find a way then the second direction was to really partner with companies like CH

Mhm.

which is famous for textbooks rental and we explored many ways to copyright including M&A it didn't work out and the third way and the third direction which we're really excited about is to build AI to not just solve equations but really focus on the word pro problems. So back then transformers were not very strong. There was no GPT3 like model to solve those problems. So we just essentially were not able to deliver on R&D sides and had to switch and this was the reason why we had to kind of start another company.

So you started AI factory.

Yes. With Victor

with Victor and with Greg and

what inspired the idea? So we did a lot of research related to how to run neural nets on device.

Mhm.

this was truly unique. I think no one else had this experience and you know that Victor is he's definitely into all these face effects all these viral growth and it felt like the right symbiosis between like the experience which we have and his business and product experience and I think this materialized very quickly in our first product which was the SDK for stickers with people faces. Mhm.

I think we had a lot of I think it's called Tailwinds.

Mhm.

Where Jify was growing very quickly. There was another company called Jifyat and since Meta tried to acquire Jify, it created a lot of momentum on the markets, a lot of interest in services like ours, Tennor was acquired by Google a few like in 2018 I think. So that's eventually we got this nice exit to snap

that was 166 million in 2019.

Yes, I think it's part of that is cash, part of that is stock. Interestingly Snap stock went from 15 at the time of the deal to almost 80.

Oh wow.

During the zero interest times peak and now it plummeted down to maybe seven. So the stock is

part of the reason I think Snap is trading where they're trading is is their like Evans stance on AI and so you were leading generative AI and my my eye product at Snap. You you kind of created it. What do you feel you know it was like at at the at the time starting something within this relatively large enormous company to launch like a product used by millions of people. We invested a lot in running neural nets on device and I think this was a really hard realization that we need to start from square one like snap didn't have the same level of cloud exposure and cloud infrastructure to train you know or run models at really high scale. M

Snap had a very very good infrastructure and I hope still has very good infrastructure to serve recommendation models but with the recommendation models the patterns are quite different compared to LLM image or video models because ultimately when the when we run these personalization models for ranking purposes one of the key parts is actually calculating the embed embeddings for each user and really plugging this embedding really effectively in the personalization engine to produce personalized results. So this is a different type of the workflow compared to what's compared to like let's say LLM or diffusion transformers. The shocking moment for me was that I learned that back then bidance had tens of thousands of H100s

Mhm.

Nvidia GP

in GCP. GCP only in GCP.

Oh. So I just learned that many companies they did this infrastructure buildup even before the charge GPT era with this was truly surprising to me that many companies were actually investing already a lot and they already have an infrastructure to run this. So that ultimately whenever new stable diffusion model from stability was coming out, Tik Tok was naturally positioned well to run this in clouds as they had the infrastructure ready but also tens of thousands if not hundreds of thousands of GPUs even back then. We talk about early 2023. I think it was it's still an important question is if AI is going to disrupt their business and the answer is probably yes as humans more and more humans feel comfortable connecting with AI and Snap purpose is to connect friends and I think AI is taking part of that and then the second question if Snap thinks this is valuable then how to do this in cash flow positive manner given that the company financial performance is let's say below meta like Snap has gross margin maybe 55 60% while social media companies typically other companies they're at 75% plus I think those are still important questions for Snap which they are working through

so you decided to quit around end of 23

yes it was September 2023 it was an interesting moment of time where like other maybe AI believers some of the other AI believers left I think notable that it's it's probably notable to mention that Jack Brady left who is now CPO at Suna.

Mhm.

A lot of other companies in the video space Hen Creatify Captions are all xamp alumni.

So the story continues same as with Yandex. It's like a bunch of people who are early believers in AI and then generative AI leave a bigger company and start lots of amazing things. There is definitely a little bit of that and I think but it also really comes down to the maybe problems which you all experienced. I think the key takeaway for me from me was that Tik Tok was the first to build this seamless video playback experience and recommendation.

Mhm. and I think Tik Tok within like maybe a year or so showed to the whole world that we moved from images to videos while for snap you there was massive investment to improve the playback but but it took year or even more to provide the same level of user experience as Tik Tok had even in 2019 2020 and I think we all kind of are traumatized by that right because I mean you go back 2019 19, Tik Tok was actually advertising on Snapchat.

Mhm.

And spending a lot a lot by any means. And fast forward 20 end of 2022, average teenager in the United in the United States spend more time on Tik Tok than on Instagram and Snapchat combined.

So I think the stats the public stats are like 40 something minutes on Instagram, 27 minutes on Snapchat, and 90 minutes on Tik Tok.

And that's in the United States. in the United States, huh?

Per day.

Per day.

Wow.

And I think in other parts of the world, Tik Tok can be over 120 minutes while like the numbers for Snapchat or Instagram could be lower as they have less relevant content. I think so. Tik Tok another thing which they did really well is like kind of covering various niche subgroups of users. So that this each niche subgroup feel really connect with each other and there are like hundreds of micro trends originating on Tik Tok and maybe dying the same day. But they were the one who really who really made it work and I think just recommendation engine content supply plus playback experience and aggressive growth spend on user acquisition allowed them to get ahead of Instagram and Snapchat within just two years.

That's insane. talking about the future. So you left Snap and we know firsthand that you were particularly trying to look for a co-founder. So why why a co-founder? Why not doing a loan? Why not doing solo? And what were your criteria for co-ounder?

Yeah, I it's so funny that you ask this question pretending like you didn't help me with that, right? You're like, "Oh, Alex, we heard about that." Obviously I I appreciate that you allow me to tell the story. So maybe you kind of allow me to kind of hide some maybe problems which were there right now not talk publicly but I guess yes. So you you remember that u ear in 2023 we were talking about like various companies how you see the space and I think back then I had a strong conviction that this is a completely new era and it's going to be just hard to like learn everything from scratch and I was asking you for any references anyone whom you know who really had this experience of working with either reinforcement learning or with these sort of diffusion transformers and there were several people whom you connected me to.

And I think it was truly exciting to see that companies like OpenAI and Meta they were very much aggressively going after the same talent pool. Ultimately I I realized that the space is is just getting accelerated very quickly. So I have never seen this much investment going to R&D project in my life.

It's also scary. But so the question is why in the first place why did you decide to work with a co-founder on on a new startup and not just solo

and I think that that question is very important for for a lot of founders because they trying to make the decision trying to understand like benefits from like

also can you find a co-founder like

it really comes through network as you help me to connect with right I think we we're connected to from many different angles but I I think definitely your effort was critical to make it work. Finally, back then it felt to me that placing the bats at the right layer is going to be critically important and it's important to have someone who actually is training those models dayto day

to appropriately position ourselves in the whole ecosystem. But you didn't consider just hiring people like like not you know co-founder level just saying okay because you had so much experience you had a super successful exit like a and stellar career I think lots of VCs would just back a startup with you as a solo founder. Yeah. And I think we we all probably can acknowledge that there is a tendency in Silicon Valley to fund people who are in Silicon Valley. So definitely a lot of our seed fundraising was really like driven by my past accomplishments. and I think although I although fundraising is important, it's the goal is to build a successful company and fundraising is a function of the quality of the team the the market, the vision and the actual progress. So that's it's it's good that I checked some boxes on the team sides, but making real progress and mapping out the markets and picking the making the right bets is still more important than just past credentials.

So that's why I felt like I need to partner with someone who really deeply understands the space.

That is that is a really good answer. How many let's let's switch to talk to to Higsfield now. So can you tell us in like a oneliner what what what is Hickfeld?

The oneliner it is like our goal is to build video video engine for content creation on social media. So why and that's me maybe like go word by word. So why I say video? Because video is way more valuable than any other part of content than any other type of the content on social media as it's easy to consume and it drives higher ROI, higher sales, higher engagement than images. Why I say engine? When I say engine, I think it's important to acknowledge that the market is just not consolidated in terms of the tools which smaller players and bigger players use. We all know a lots of famous mobile app companies coming from let's say Cyprus and those companies they can run thousand of ads

every day AB test was performs well and really continuously iterates every day and the media buying for them happens fully automatically it's like it's like similar to trading meaning that algorithms do most of the trading most of the operations and like but the key decisions needs to be made by humans. So like the same type of trading terminals also exist for these performance marketing companies and they also have the creative machine to make like thousands of videos every day while smaller players they really struggle to make any content for social media. So for us really making an engine so that smaller players can also make let's say tens of like maybe experiment with thumbnails maybe experiment with hook in the video

and and really connect the content creation process and the metrics. I think that's what we call engine and we think this is a net new product which never existed in the space simply because content creation was never possible with AI before

taking out a camera and actually filming something.

Yeah, exactly. And I think the third piece is which is very important is social media. We do believe that fundamentally social media is driven by I mean it has just very different rules. First of all, it it requires to produce several pieces of content a day. It requires to publish content daily and it and the lifespan of content is just maybe one day at max three days. So it all is very different from let's say so this like professional social media use case social media marketing is completely different from consumer use case which is just trend driven and it's completely different from TV ads or Hollywood's where the timelines timelines are still measured in months regardless of if AI is used or is AI is not used the timelines in this se segment of the market is measured in months and I think so social media marketing is a trillion dollar industry today and it naturally has this like let's say pattern of daily usage that's why it's a very compelling market to go after in our opinion.

Yeah. And as the user usage of web diminishes people still spend lots of time in this old garden of you know Meta and the old garden of Tik Tok watching videos and you know being bored so being ready to learn something new. see some like a new product. Lots of video AI companies were started around the same time. actually even slightly before you started

Hixel you already mentioned.

So going after the same field after five companies have raised like tens of millions of dollars in it was was it a hard decision? Was it calculated? Yeah, that's by the way a very good point and I think this is a key a key learning for me and I think that that really talks a lot about the early stage investment right when I look at early stage investors a lot of people told us that look it's going to be very competitive and you need to pick like very differentiated space you need to have very differentiated product because it looks like a red ocean and this is part of the reason why we try to do some other things as we try to come up with like some differentiated product which so that in terms of like so that's from the eyes of VCs we are in an we are in a completely new market but I think it's important to acknowledge that these type of the thinking didn't work well for us and there are other types of VCs who really just believe in the team and believe that they will figure this out and I think you are this type of early stage VC right and I think so for us it was very important to sort of just while the while there are like so many companies in the video space.

We just realized that no one actually goes after the social media can content creation. No one really goes after social media marketing and we can well differentiate. But it definitely took us like over and and over last month since we launched on March 31st, we got to over 100 million run rates.

100 million now.

100 million runs.

But the last I've heard was 50.

Yes. But it's Yes. Last month is 60% month overmon growth. So

Oh my god.

Yeah. Yeah. So we we we really grew very quickly and I think we just sort of rebuilt the whole team internally to really be so that's everyone is passionate about about the space.

And your team is small, right?

It's not small anymore. I think the team is approaching roughly 70 people today.

Oh wow.

Still quite small.

Yeah.

For

for your numbers is still very small.

As long as you are over a millionaire or per person.

Yeah. That's good.

It's it's very good.

But yeah, it's I doubt you can hire as fast as you can grow.

you're absolutely right. So I think as we are maturing as as a company today is the day when we launch our team plan. So it's not just a B2C offering, it's offering for teams to collaborate.

Mh.

So beyond so that we give more reasons for users to use the platform.

Mhm. beyond just content creation because content creation burns credits and credits and we provide limited amount of credits and then which leads to user some of the users churning. So by building these collaboration workflows we give more reasons for teams to stay and use even without producing new content because a lot of work is not just to make content but also to review together discuss the plan do basic project management. So while we have seen companies like frame.io O succeeding in preAI era in that space we hope to upgrade the workflows and make them native for for teams which are AI first. I think the complexity of the business is just increasing and there are going to be several more B2B products which we will be releasing by the end of the year. For us, the complexity of running the business increases and definitely hiring people who have this direct experience is important.

So now I think everyone clearly realizes that the space you're after is much bigger than other spaces that people you know envisioned when they like when when you think about a video company you think okay Hollywood visual effects like doing conf content for Netflix. but this market is incomparably smaller than social media and social media ads. Do you think a lot of your competitors are going to be switching? Do you think a lot of like incumbent companies are going to be going after the same thing? No,

that's a really good question and I think this is where I'm I'm glad that you are saying that and u because I think a lot of other companies in the space let's say runway they attracted more capital than anyone else and they really try to reinvent the Hollywood business while there is no technology company successful technology company built on top of the Hollywood usage really

in terms of the usage in Hollywood and the interesting part is that One of the argument is that Adobe has Adobe Premiere which is using Hollywoods while I think most of their revenues really is accounted for Photoshop and I think even Adobe is not does I mean like Hollywood doesn't drive most of the value for Adobe.

Mhm. And I think for us it's important to really appeal to the state of the market today where as as you said social media marketing and social media at like ad layer are 100 times bigger than software for Hollywood and I think today in the space of social media marketing we're already number one in the space. So I think and I think Runway is in a completely different type of the business because they go after Hollywood. They train own models and they are incentivized to spends a couple hundred million dollars to catch up with Google and OpenAI in terms of the pure model capabilities

which is a weird thing for a startup to do but that's my opinion. can you can you give us like an example of like a creator like a user journey like is it where did they come from? What did what do they normally do with the product? Is it mostly web? Is it mostly app?

Yeah. So I mean another key decision for us was to abandon the mobile app initiative. We are not seeing the patterns of usage changing with AI meaning that mobile apps are famous for really high churn. We see the same thing. Most of the usage is coming from desktop. We have mobile web version that's maybe less than 20% of usage.

So most of the usage is driven by professional social media marketers. There are sort of two types of customers. First is like smaller marketing teams maybe up to 10 people today and our goal within the next two months to make sure that Hicksfield is a good destination for larger marketing teams as well. And those teams they very often do storyboards with AI. And last week we released our the first model in the world which can actually do the full storyboards which is called Hixfield key frames. Sora is doing everything at once. It basically takes it basically is doing a lots of things like implicitly meaning that they take a prompt they produce a script they make like let's say a storyboard they generate all these segments they compile them together and and then they add audio but it's it's all it's amazing consumer products but the quality is not really high for professionals and there is not enough control as well.

Mhm.

We think that Sora is in a in a different market.

They're trying to go after, you know, Tik Tok.

Yes, they're going after Tik Tok and I think so for for these more like professional marketing teams, Hixel provides the model and also the workflows. and then there is and I think this accounts for like 60% of usage roughly 40% of usage is driven by AI enthusiasts. So by AI enthusiasts I mean like people who just try to make sense out of AI try various tools. I bet they try midjourney as well. What happens is naturally midjourney has access to only their model. We provide access to all the models and this is a big differentiator.

There's u in a market with lots of participants and lots of channels it's always you know good to be Switzerland. It worked well for Perplexi and Corsor. Yes. it's important to acknowledge that we take a lot of inspiration from these two companies. See, because they operate in a way more mature space and they are clearly one of the key winners just pure out of the traction and we take a lot of inspiration from their work. Before you got to this incredible exponential growth which is like happened in the last what 9 10 months. You spent about a year working on a product figuring it out. you tried a bunch of different you know avenues can can you tell us a little bit more like what what did you try what what learnings you got from there

there are several things which we try to do first and foremost we really tried to make sure that maybe I'm maybe I think it's important to kind of start with maybe general overview that a lot of VCs and people in the market are going to say oh we don't understand who uses midjourney midjourney is really it's not it's not like clear who actually uses that. It's like hobbyists. They come one time a month and leave.

It's not a VC. It's not like it's not like

it's not a venture capital.

It's not venture capital.

Not sustain.

Yeah. And I think we got very much and that's probably on me. We got very much distracted by this talk simply because in the end of the day the data which you get allow you to learn very quickly since the space is evolving so quickly there is no better way to learn than getting direct feedback from your users.

Mh. So I think ultimately the rules of building companies do not change with AI as you need to build something which users need and sometimes the adoption may be driven first by hobbyists which we have seen in 2024 today it's driven by these very small groups of professionals people who are making money with social media and it's obviously going up the chain and I think we just were trying to Look at we were we were we were looking at some other areas where we did not get much commercial traction but we learned a lot. One of the spaces is the performance marketing and also performance marketing specifically for the space of the short form dramas because I mean we learned that the value of content creation itself is relatively low like short form dramas they learned how to make a minute of content for a cost of around $1,000 which is two magnitudes lower compared to TV and Hollywood while they can spend maybe 30k 100k to make 100 minute long drama.

They the leading companies are going to spend tens of millions of dollars to promote the best performing dramas

and they're going to b they're going to burn money very quickly on like Facebook, Epavine, Molo, Tik Tok very quickly within a period of like maybe 20 30 days and they will be making tens of thousands of ad creatives and there are going to be one which takes it all which is going to take like tens of millions of the ad budget. Interestingly enough is this process of like improving end performance is pretty much data driven. So like if you see that the drop off is like on the second number eight then you can like use AI to improve the conversion rates. And I think these all gave us a lot of insights on like how AI video systems will revolutionize social media advertising really connecting data let's say platform management because today it really has to go cuz I mean content today has to go through all these v various social media networks and the content creation.

It was never possible before to make like one pipeline combining all these three aspects and it's and it's going to be exciting to see in the next two years someone that someone is going to do that hopefully it's Hicksfield and yeah then we will be then this company is going to be legit participants in this trillion dollar markets a very popular objection to that is to say oh like look how like Hicksfield whoever else you bring any startup name is going to even remotely compete with Bidance and Google historically Google and Bidance have the strongest AI teams in the world, more distribution than anyone else. And the answer to that is that they are going to be the winners, but they're going to be losers. And Meta is likely to be a loser because they have very well optimized ad marketplace today, which is going to be completely disrupted in the worlds where most of the content is generated with AI.

And while there are going to be winners and losers in this space, there is a place for Switzerland AIA video Switzerland to really like orchestrate and really help smaller team to make sense of maybe 10 various ad networks which exist out there

using using AI. While Higfield is focused today on the video generation, we are we do a lots of work behind the scene on aentic workflows and video understanding as well. So you're collecting the data that Google doesn't have. like they have YouTube, but they don't quite know what videos how people make videos with AI, how they perform, and they'll have to like learn pretty much from the same point as you. You you're absolutely right. I think so. Also, what's what's important to acknowledge is that Google is clearly number one company in that space today. They completely turned it around. While they had a lots of problems with initial launches of their AI chatbot, now they are they have pretty good Google like Google flow is a very good product for v 0ero for a large company. They have very fast iteration speed for vo model for nana for like image models like nana banana.

It's all truly exciting. We see although we see the usage across all the plat across all the various use cases and I can tell you while it's probably true that none of like across like 11 domains which we track subdomains it partially overlaps which you see at the LM arena or artificial analysis the data on these public benchmarks is representing like human choice although we have real data we have data like how often users actually download the video. So this is a very this is way stronger feedback loop compared to anyone else and we see that maybe across five domains out of 11 Google is the leader but it's not the leader across all 11 domains and we collect this data we actually know which model is better for which use case and how it should be like prompted or like how it should be chained to get really high success rates

taking into account your strateg Y that it's would be way beyond just video creation. who you will be looking for your team?

Absolutely. So I think we we are looking for first principal thinkers and very energetic people who are excited to figure this out how the space is going to evolve. Mhm.

It's very interesting to see that the whole marketing approach is also being completely different from the way how it was done before. A lot of content really has to be educational as the as the interest in video AI surges and there are and there is just not enough educational materials how to use it. So it's so interesting that we use AI to make educational content about AI to talk about how to use AI, right? So it's

that's the loop.

Yeah, that's the loop, right? So the goal of marketers today is not just to understand the distribution channels and the hooks and the topics which work out well which is still very important and all of that is getting automated through AI platform AI platforms for social media listening

but also chain this all in aentic workflows. So like hickfield marketing today is run very heavily with agents. It's true that sometimes it backfires when we work with like some top profile creator and there is like this charge type of the response and then they become pieced but that's really what's happening is like the that the marketing is getting fully automated and completely reimagined and as we build this marketing machine internally we are going to to push more and more of these workflows in our products. So in some sense we are the users of our own products

which is the best

which is the best. Yes. So at Hickfield we very quickly amassed like billions of reach on social media more than all the competitors combined. And I think we just try to make this self-reinforcing viral growth loop where we use AI to make content about AI. We we use AI to do the social media listening to get all the insights and find the best distribution channels. Again the there is a downside of like also connecting to some of the creators using AI which people are not happy about but then and then we and then once this content is getting distributed this helps to this content goes viral it helps to increase awareness it drives more people on the platform and it closes these viral loops and I think over the time as like our own marketing workflows mature it's going to be part of the main product and I think historically the marketing techn technology companies they took a lot of time to take off really a lot of time and it's so interesting to see that the platforms like open AI agents seeing that NA10 while they were not designed initially for marketing purposes they are being heavily used for marketing purposes so the the like the rules how marketing comp marketing technology companies are being created th those rules are being completely overwritten today and I think we all also see a very interesting opportunity in front of us where we actually train models even today we train models to drive higher download rates and higher engagement rates as we like kind of pull the data back

like of content generate with Hicksfield we can run reinforcement learning to actually improve the performance of contents

it's really cool so your team is now 70 people

roughly breakdown who does What how many people in like which department?

I would say in GNA we have maybe three people.

Okay. Not too many.

Not too many on GNA. Recently I did an analysis by the way most of the companies even Proplexis 11 Labs Corours they all have like GNA team very lean like typically way under 10 people. It's surprising to see and I think AI automation and all the procurement process invoice automation is also tremendous and I think we we are scaling our growth team very quickly and we like to have like very young we really like to hire young people with founder mentality that they are willing to take ownership over some channel let's say Reddit, LinkedIn and so on. and then figure out this marketing workflow and then really automate that and work across the whole across the whole company to build end to end workflow. This has been very productive for us. So we had maybe two people on the growth side in July and now this team is 10 people

and and and I think we typically try to have maybe two three people on each channel but the number of channels is actually getting to 10 as we are going from like more like this as we are mixing viral marketing alongside with tutorials and educational materials. So the number of our initiatives grow and I think we are always happy to work with founder with people with founder mindsets who really want to think from from the first principles and reimagine the workflows for a specific channel.

Mhm.

So your next one to play also will be somehow connected with growth.

Yes. yes I think that the next 10 hires out of the next 10 hires more than half is going to be related to growth as we are building all this automation and this automation eventually flows into the core product

and I think marketing teams are by definition are the users of the product internally and

the beta testers

yeah they are the beta testers for sure and I think marketing team is also becomes a very useful channel to solicit feedback from other top creators but also from the creative partners to keep improving the product.

What's your take on remote not remote work?

Yeah, for sure. That's that's a really great question. So, I think that my stance completely changed. I think it really worked well for Snap. We had offices like I think over 20 major offices. So, it was in LA, San Francisco, Seattle, Salt Lake City, New York, Texas, London, Ke Sydney in Australia. and so on. So like it was really distributed team wi-i which was surprising for a company which has just 5,000 employees and I think today given that the pace is just so quick it's very hard to be fully remote it's very hard to keep everyone in sync because I mean part of the best of Hicksfield is that we are in terms of the model implementation into the workflows Hicksfield is going to be faster than anyone else right while Google and by dance they will definitely benefit benefits from having like a vertical stack.

None of them is going to have like monopoly on the market and we are betting that Hicksfield is simply the fastest company in the space. So whenever a new model comes out, we are piecing this into a workflow like the V zero of the workflow were those templates which make those models really easy to use and our VFX templates but most importantly camera effects got a lot of adoption especially for let's say e-commerce use case so that someone can just upload one photo and transform it into a video

showing the product. I think this drove a lot of engagement in like V 0 stage of the company and now we're progressing to V1 where we enable to make like the final where we enable the teams to collaborate with each other but also actual making like actual video which can be posted on social media and then measured on social media. We have these workflows obviously internally today and the goal is just to wrap it in a nicely looking product not just a bunch of like this u lovable coded internal prototypes right but really built on top of tools like NA10 but then but but to a real product which is ready for like adoption at the scale of like millions or tens of millions of users.

It's really cool. So you mentioned lovable and NA10 for like internal prototypes. How big is your engineering team and how much you know the split is between vibe coding and actual coding?

I think the engineering team 90% of them they do the real coding. So I think the most the popular tool today is clot code to be honest. I think the the engineering team is total 30 people and maybe 28 of them do like real coding. Yeah. Yeah. And there are maybe ones of people doing like prototypes and integrations and some internal automation.

That's really cool. How how do you split them like between how big is the like a team? Cuz I know Meta and Snap they they had like seven, eight people teams per like feature that would deploy a feature and test it. Perplexity now has three people teams. do you where do you gravitate? We definitely lean towards the best practices like perplexity. so thinking about the teams of three four people.

Mhm.

All technical or like

I think design is just very important for us. But our designers they're not just like thinking in terms of the pixels but in terms of like frameworks and components. So the designers are rather technical.

Mhm.

In terms of like they understand how like the primitives and architecture of our front ends. So yeah and we are basically rethinking this concept of forward deploy engineers.

Mhm.

This is I'm hearing this more and more from you know founders of top AI startups. Everybody wants engineers that understand how you know architecture works that have great taste so they can design like vibe design and with founder mindset so they can take agency and understand growth. So yeah

very high bar. Yeah I think the same applies to us. I think we're a little less skewed towards the design because we have amazing design team. So we can trust them that they are going to produce something which is which is great. I think for I think with what's important for developers is to understand the usability patterns. So they don't need like much explanation be beyond the mocks and what becomes very important is for as we operate at like meaningful scale. So like we talk about millions of users and then it's it becomes important for engineers to understand input and output metrics to make sure we deliver performant experience which is critical for video heavy workflows. You also appear to be a very successful angel investor and well LP in our fund. Thank you for you know being a part of this. How did you start investing as an angel cuz you know your list is truly impressive.

angelist bra solve intelligence both early stage late stage

I think most of the successful angel investors they are actually very active in early stage and they can spot they can spot great companies I think you have great traction I think other someone else who comes to mind is Ellen Gil and so on and I think I'm not even remotely close I think I just had some access to and like anthropica 2 billion it's a steal

like scale AI with all government contracts at like 1 billion is a steal as well. So I think I just saw those opportunities quite early. I think Bre's adoption across the startups was absolutely crazy and I think for solve intelligence I saw that they sign big customers very quickly and it's so all these decisions are datadriven although successful angel investors they typically can understand they typically make a decision based on the founding team quality and their vision about the future and I'm definitely not good at that I think I make my invest investment. I used to make my investment decision decisions based on that actual traction and this data. I think this worked well over although since I started Hicksfield I I'm definitely not active. I invest in friends companies maybe once a quarter and help them with fund raise but that's that's it. I I'm not active anymore.

It's very hard to get you know enough time when you're running your own startup. A lot of people talk about the time but I think it's important to for for every startup I think it's important to do the right prioritization and then taking action on the most important things. So, and it takes a lot of mental capacity because I mean we all have a little bit of the confirmation reaffirmation bias where we tend to prioritize things which were important for us in the past or prioritize things where we kind of maybe had like some immediate failure and you think oh we must figure this out right so there is a lot of these emotional components and I think just really trying to apply how fast and slow thinking So like making sure that the prioritization across the company is right. I think this takes just all the brain capacity which I personally have.

So this is not about time this is about attention. Beautifully said.

Yeah it's it's about attention but really also making making sure as as I'm quite active person and I strive for like daily progress on the most important things. I mean sometimes like it just creates a notion of like because because it's it's it's it creates a little bit of of of a different effect where we start to prioritize things where we spend more time while it it must be different. we need to identify the important things and spend time but not making sure that whatever heavy process kind of got created in the company becomes takes even more and more resources over the time I think just making sure that we can apply like this kind of maybe like as we rep prioritize and learn every day and that's the question which I ask everyone like what did we learn today to be successful it's important to collect these data points but also do this holistic revision every weekend as Well,

that's I think the problem with angel investment. angel investment has very long dopamine loop. So you can make a decision and the result will be only in 10 years and you

you need to be patient. Yes.

Yeah. Dopamine does not survive that long.

Yeah. And you cannot like truly like see was successful or not before that.

This this was truly inspiring. Thank you so much.

Let's do the takeaways. what did you learn? well, I learned first of all the way Alex phrased it was was beautiful that a lot of time syncs that are created in companies just tend to have momentum of their own and attract more and more resources just because people spend time and start believing rationalizing that it's important if they're spending so much time on that about investor investors thinking that you know they they follow they follow trends and they like see competition where there's no competition

and customer feedback loop is more important than oh yeah

like overall trends.

I really like the way you built your and I I feel like it's cheating because building a product for your own use is probably the best way to build products because as a like they always say talk to your customers but if you are your own customer that's even easier. Don't even have to talk. definitely the way to go for introverts.

Yeah, absolutely. And we and like we made sure that the creative team and data annotation team are all have marketing backgrounds. They worked on some commercials either like TV commercials or social media commercials but like we we we definitely made sure that we have this close internal feedback loop. probably the last question what's your strategy to be more like Tik Tok in terms of like the situation when they are prepared infrastructure for the things happening for the future compared to snap what you have to do right now in your company to be prepared

yeah I think like the quality of our infrastructure team is way I I I cannot compare against like open or meta right so Tik Tok I didn't work in these companies but definitely better than what I've seen in Snapchat. So I think the quality we we definitely invest a lot in the infrastructure as we have to as we operate at like meaningful scale in terms of the GPUs as well and we are constantly working with leading companies in the space like Nobus to make sure that we are up to date with the trends and we leverage the we leverage the like infrastructure stack built by these cloud companies as well.

Thank you. Can you one one thing that Marina usually asks but like maybe

any advice for

Yeah. Any advice for a book?

Oh, okay. Yeah. Two question. Any advice for founders for first- time entrepreneurs? And yes, any suggestions what to read maybe not not particularly book maybe to something to listen to like your favorite podcast, your favorite books.

Yeah, I think the most important book for me was Charles Ferguson book, right? the partner at DVC

the inside job.

Yes. So I think just long story short is we all hear about the success stories. Each startup goes through many downturns and all decisions related to money

are not typically taken lightly by some of the party. I'm not sure there is any other book except Charles book which really goes through the complexity of building a company as yeah I mean we all understand I mean we all understand high level that different investors have different priorities depending on the size of the funds maturity of the funds returns of the funds I think this high level concept is well known in in the valley and there are a lot of stories which are which people tell each other during the dinners. At the same time, I think in terms of the book, I think Charles did a great job trying to outline some of the complexity.

That is a good really good recommendation. Thank you. And also, incredibly privileged to work with Charles. Thank you so much for joining us today and, you know, sharing stuff. I'm like, I'm glad we caught you before your revenue hit a billion. that would be like our podcast would need to grow a little bit more to earn your attention but I mean like with this kind of growth you'll be there very very soon

doing our best and thank you for your support.

Oh thank you very much.

You were obviously DVC is the first funds who invested in Higgsfield and you supported us through each stage of the journey. We're very grateful for that.

We're running out of money at this point. Yeah, that was that was was this moment as well. Yes.

Thank you.