The Future of AI Video Won’t Look Like Video: Synthesia’s Bet on Interactive AI | Victor Riparbelli, Synthesia

Victor Riparbelli:

Truly AI native video is not gonna look like the video we know today. It's not gonna be a broadcast that's sort of generated once. It's gonna be interactive and personalized and so on. Just like you can write a book from your laptop, you can make any song that you want from your laptop, you're gonna be able to make a Hollywood film from your laptop as well. It's so easy to imagine 10 jobs that gets, like, lost to AI, and it's so hard to imagine 10 jobs that gets created by AI.

Victor Riparbelli:

The DNA of a great company today is great people, and I don't think that's gonna change even ten, fifteen years.

Christoph Magnussen:

If you truly want to understand AI and get to the DNA of AI, you also need to understand video when it comes to AI, and one of the persons when it comes to video is Victor, the founder of Synthesia. So we are here in London having a conversation about what makes AI and video so unique, and he's not just any founder that started... Who started the company a couple of years ago. Synthesia is actually in the market for almost ten years. I think next year it's tenth year.

Christoph Magnussen:

So this is truly a talk that I have been waiting for and you can look forward to a conversation that goes really deep really fast. So grab your coffee right now and then we dive into that. Thanks for inviting us, Victor. I'm very happy to be here in your office. I was quite impressed when I came in.

Christoph Magnussen:

I'm like, okay. This is one of the the longer running AI startups. And then I felt like 2017 is not a startup anymore, but quite a journey. So maybe give us the the quick overview before you hit the ten years frame, what happened in the past nine years now.

Victor Riparbelli:

Yeah. It's it's crazy. It's been almost ten years. So the kind of founding story really is that I come from Denmark, grew up in Copenhagen, and in kind of my late teenage years figured out that I I love doing things online, started off doing e commerce stores for local businesses, building websites. Interestingly enough, think that at that time it was powered by Shopify, Squarespace.

Victor Riparbelli:

In a few clicks, you would like to set up a website, which an agency would charge, you know, like 50 times the amount the 17 year old Victor would charge. It's kind of similar to what are some of the stuff we're seeing in AI right now, right? But like a single person who's like at the forefront of what's possible can just... There's a huge arbitrage in that. But that's kind of where I started, websites, ecommerce stores, got bored of that after a while because I think the ceiling is just, you know, a bit capped.

Victor Riparbelli:

Went into Danish startup ecosystem, did that for some years, and figured I wanted to build my own company. I figured that I wasn't super excited about building accounting tools and more kind of business process operation software, which is mainly what I was doing at that startup in Copenhagen. I've always loved sci fi and kind of the Frontier of Technology, the kind of like the weird ideas, and that's really what I always got me excited. So I decided to move out of Copenhagen. It's a great place, but not really a great place to build a company like App.

Victor Riparbelli:

I went to London and spent a year basically trying to figure out what company did I wanna build. And in that process, I met a lot of interesting people, among them my co founders, professor Matthias Nießner and Lourdes Agapito, both professors. And Matthias Nießner had done some of the very earliest research in AI video. Looking back at it today, it's obviously, you know, not very impressive. But when I saw his research paper called Face2Face for the first time, I just felt like I I kind of see magic, you know.

Victor Riparbelli:

I got so captivated by it, especially from someone who loves creative things, like make music, love films, always like had that kind of side to me. I saw this thing and I was like, holy crap, at some point you're gonna be able to actually generate, you know, video content just from your laptop. Just like you can write a book from your laptop, you can make any song that you want from your laptop, you're gonna be able to make a Hollywood film from your laptop as well. That's kind of the eureka moment. So I wrote down a thesis, got the kind of founding team together, and that thesis really had like two big ideas in it, which is actually still what we're executing towards.

Victor Riparbelli:

The first part of the thesis is the one that today is now very obvious, which is that when you can use AI to generate actual video content, the marginal cost of doing that is gonna drop maybe not to zero, but very close to zero, not just in the dollars it costs to create a piece of content, but in the time it takes and in the skill required to create something. And that first idea is basically what we built the company around more or less up until now. The first three years we were in the lab, just trying to make the tech work, because it didn't. We had this like first big insight, which was that the first market for AI video was not gonna be high production, cool videos. It's actually gonna be more corporate, lower quality, but higher volume videos, more about turning like a text document to a video than helping an agency create something really beautiful.

Victor Riparbelli:

We build out the full value chain around that and we could chat more about that business, but I think we're definitely probably still in the early stage of this, but just the video models and how much they've accelerated the last six months, it's pretty crazy. Back then, we had this tagline, which was in ten years, you can make a Hollywood film from your laptop. And we're now almost 10 since we made that statement, and I think it is from a technical capability where we're really not far off. The second big idea...

Christoph Magnussen:

Depends on the on the Hollywood film. You define it, right? I would say For sure. Quality wise.

Victor Riparbelli:

For sure. The second big idea was this idea that what has always happened throughout all of history when we've invented new media technology is that media formats change. Right? The internet gave us things like a blog that didn't really exist in print before, right? The first websites kinda looked like a newspaper on a screen because that's kind of what we could imagine, then from there we learned that on our website we do a lot of things you can't do if you're using a printing press to distribute your message, you can update the page in real time, you can personalize the newsfeed and all the things that makes a website a very different thing than a newspaper, right?

Victor Riparbelli:

And we think the same thing's gonna happen to AI video, like truly AI native video is not gonna look like a video we know today. It's not gonna be a broadcast that sort of generated once, just like we used to print a newspaper once and distribute it, right? It's gonna be interactive and personalized and so on. And so we started investing in this eighteen months ago. Four or five months ago, we launched our first product in this space, which is Sessions, a platform, for two way interactive video.

Victor Riparbelli:

So rather than just passively watching a video, you're interacting with it. It could help you train a salesperson, for example, where the avatar can pretend to be a salesperson. You actually simulate a conversation, which is an incredible way to learn. It could be collecting information. So just like you would send, you know, a human interviewer or a UX researcher to talk to your customers.

Victor Riparbelli:

Or I would do one on ones with my employees. You can have an avatar that can survey people, but in a smart way, right? You can ask, you know, probing questions. You can't just accept yes or no for an answer, and for the person who's being interviewed, it's much easier to talk than it's to type, deliver five times the amount of information. We're launching more products in this.

Victor Riparbelli:

We also launched a developer platform. So if you wanna take the primitives, you wanna think like a talking head style avatar, you wanna build an app around that, you can do that. So I think we're just like sort of in the first phase of the next chapter of Synthesia, which moves from let's help people make video, as we all know video, way faster, way cheaper than has ever been possible before. So I think that's a big part of our business, but we're making a huge bet that two way video, and who knows, maybe we won't even call it video in some years, we'll find some other like name for it. Yeah.

Victor Riparbelli:

But that that's gonna be a really important interface for how we do everything from work together to how we entertain ourselves and so on.

Christoph Magnussen:

The first one is more this linear thinking of, okay, what we do faster and that's what many companies at the moment do with LLMs, for example, saying, okay, we we did this process now for a while. Okay. How do we make it more efficient? Instead of saying, okay, what is out there that we didn't even think of what we wouldn't do? Many questions around that, but before we dive into that, one quick question to like get a little bit of your take.

Christoph Magnussen:

There are a lot of people out there and I know from other founders that know you and and what I heard from you that you're a very people driven person

Victor Riparbelli:

Mhmm.

Christoph Magnussen:

And you don't believe in the agent only economy.

Victor Riparbelli:

Mhmm.

Christoph Magnussen:

What is your agent and people set up at Synthesia? How do you do that? How do you set it up and how important are people and how do the roles change of the people that work here?

Victor Riparbelli:

Yeah. So we're very distributed in terms of like how we change our ways of working. So of course every team, you know, has a mandate to go and reinvent themselves using AI and in every team at Synthesia, we're either, as you said, bridging a process, what we used to do manually, we can now do with AI or we're doing things we were never able to do before. For example, things like learning about our customers and what are they doing with the product that has massively been advanced by the agentic economy. The way we build product, of course, has also changed very rapidly, right?

Victor Riparbelli:

Like way fewer docs, way more prototypes. I myself create prototypes all the time, such a magical tool. But I think basically what we, our strategy has been, right, is to enable everyone to have the tools that they need to reinvent their process, help them when they need it. But I think if you hire really great people that are at the frontier of the tech curve, which I think we have done here, you almost don't even have to ask them, you know? People will create like the most amazing workflows themselves.

Victor Riparbelli:

That's what we're seeing. We then went through this phase of a lot of people call it kind of token maxing, which is like, just like go bananas, do everything you want. I think that's very helpful, and I would suggest any company that hasn't had a period of that to do that. But then of course it's like looking into, okay, if every time we send an email to our customer, it costs like $20 in tokens because we're running like 600 agents doing crazy stuff, maybe we need to like slice that down to something that's more manageable. Right?

Christoph Magnussen:

What do you look into people at the moment when you say like they they need to be AI driven but still driven also with with other things? Like, what what is the thing that you look at at the moment when you say this is the person we hire because there is... I I for myself experience a gap between you have a professional in a certain field or you have a professional who can also look on his field or like beyond and scanning and say, I know that there are areas that I didn't need touch with AI, not even making more efficient but look beyond and broader.

Victor Riparbelli:

You mean to be like kind of domain expertise versus sort of creative thinking?

Christoph Magnussen:

Because what I like about your story is like when you you started with something in your mind but you pivoted because you saw something works Mhmm. And you know, okay, we go from there and we don't go out with a big exponential vision right now for the moment, but by learning on the way, we get closer to that. Yeah. And I see I see a lot of people struggling with that if you like compared to their jobs because the only thing they can see is AI doesn't work here right now or this is not perfect right now.

Victor Riparbelli:

No. I think that that is really hard and I think especially when you're at scale, I think I think at the risk of sounding arrogant, but think it's really just because I've been doing this for ten years and I know so much about the business. I think there is this sort of sense that like six months before we need to do something, have this like gut feeling, we probably should be doing something. Depending on how strong that gut feeling is, I'm either gonna go out and actually do it. But I think it's like as a founder, these people have been there for a really long time, we should have a really good gut feeling of like what we should do.

Victor Riparbelli:

And I think it's kind of your job to bring forth those things, right? So instead of waiting six or nine months before we do something, you kind of put your fist down and say like, guys, we need to be doing this now, even though like we can't yet see it in the numbers. I think another interesting spin on this is that I think we've been going from this, I think every company, I think it was actually Tobi Lütke from Shopify, I think who kind of described it in this way, but I really like it, was the idea that as a company, you kind of swing the pendulum between being kind of very wide and exploratory, kind of like in ideas mode, and the other way of being much more narrow and being very sort of execution driven around something. And we've swung that pendulum many times in the history of Synthesia. Usually when you kind of reinvent yourself, you pivot, you kind of swing into the wide.

Victor Riparbelli:

And the last like three years, really, well, not the last six months, but last three years, really been about, okay, we built this video product, we found the right buyers for it, we figured out how to target them, we built a sales team around it. We're just like really going all in on scaling what we knew to be true and what we knew work, right? And now the kind of company is swinging a little bit in a different direction. Now we're just like wide open phase. We've launched, you know, three new products in the last four months.

Victor Riparbelli:

That naturally is a chaotic process in any company, but especially when you're like a bigger company, there is so many things you have to think about, right? We have an existing business, we have the new products, where do you invest, how do you make people buy into the vision and where we're going, and that is sort of a chaotic process in many ways, but it's really, really important, right? And I think there you can tell what mode people prefer, usually what I have noticed is that a lot of people who've been with it for a very long time, they love the wide exploratory, we don't have the answers, but let's try a bunch of stuff, let's see what works, let's get creative, they like that. And a lot of the people who, you know, maybe the ones who've been hired a bit later or generally kind of more sort of operational people, they struggle a bit with that uncertainty because you want answers, right? The sort of, you know, classical things you sort of see is things like a go to market team, right, are driven by incentives and incentive structure for like how they hit their quota.

Victor Riparbelli:

And that's of course what they care about, right? That's easier when you have a lot of answers. It's harder when there's more uncertainty. It's harder when the product is early, and when you're still learning. And so you have to manage a lot of these sort of things.

Victor Riparbelli:

And I think you need to figure out like who's good at what and make sure that the people who are good at exploratory stuff gets to focus a lot on that, especially when sort of that's the mode of the company. The people who are very good operationally are the ones who maybe work on some of the problems that are sort of less uncertainty about where they really can shine and go and execute and deliver on something.

Christoph Magnussen:

How about your customers if you look at that and because I could imagine that for them, they also need a lot of education. I remember I had a podcast almost ten years ago, let it be nine or eight years ago with Stuart Butterfield, the founder of Slack, who then said we actually have to educate the market, we had to like teach them that this is not email, this is not a doc, this is a new kind of product Mhmm. And he said it was very very intense to teach the customers to get that. What I get from Sessions, if you say, like, you interact with it, like, with an LLM.

Victor Riparbelli:

Mhmm.

Christoph Magnussen:

But you see a real person

Victor Riparbelli:

Yeah.

Christoph Magnussen:

You have this feeling of this is maybe something different as you said, like, it it... It's not video, it's not human, it's something between, it's new. So you need also customers like that. How different is that when you look from a product perspective on it to do that team wise? Because I mean, selling corporate onboarding videos is much different than...

Victor Riparbelli:

Yeah. Well, I think it's in many ways, it's sort of the... It's a similar challenge, right? I think now when we look back, AI video is much more mature. Like I think the market for it is like much more clear and so it seems, oh, well, was pretty obvious.

Victor Riparbelli:

It wasn't obvious back in the day, right? And we did a lot of things from a product perspective to make sure that we took our customers on that journey and we helped build the category. I think the hallmark of like great companies and great teams is that you have kind of an exponential vision, you have a really, really big vision, but that you can also sequence your way to it. And often people can, like if you look at companies that aren't successful, they're usually only good at one of them, right? Like you can have the one who's like the most philosophical and right person, who's like in ten years, you can make a Hollywood film on a laptop.

Victor Riparbelli:

That's a great insight, and back then not a lot of people had that insight, but if you just try to build for that, you would have failed, which is why we're probably the only company from that era left in this space, because everyone was trying to build use case that weren't mature enough, right? On the other hand, if you're only good at like the operational sequencing part, you might end up building for local maximum, you don't take any big swings. And so I think holding both of those things at the same time, I think is what makes entrepreneurship really difficult. And so for example, when we launched the Avatar product, right? Seeing the signal in like enterprise users, it didn't just fall into our lab, right, with a lot of like, you know, very intentional customer outreach interviews.

Victor Riparbelli:

I read a 500 page book about the invention of PowerPoint, which to most people probably see like one of those dull things.

Christoph Magnussen:

There is a book with 500 pages about the invention of PowerPoint.

Victor Riparbelli:

And it's very rare, like I had to find some like weird, because it's written like, you know, ten, fifteen years ago. But you have to really seek signal in like, you know, what's the history of technology? How do technology get adopted? Who are the right users for this? And we had this, you know, mantra, utility over novelty, which was this idea that, especially back, know, for the last, you know, ten years, AI video has been this like shiny thing, right?

Victor Riparbelli:

People wanna talk to you about, they get excited about, they dream about what they can do with it. And that's a great thing to have for our company that you get a lot of interest, but if you stay with just interest and people dreaming about what the tech can do one day, you're not gonna build a real business, right? So we had to figure out what is the signal, noise. And so we did that and then we figured out, okay, these users, right, they're not video editors, but if we just give them a clip of an avatar speaking, like that's not the video they have in their mind, right? A video is usually someone speaking, there's some text on the screen, there's an animation, there's an introduction, there's all these other things that makes a video a video.

Victor Riparbelli:

And so we made a very intentional move to say, okay, we're actually building for a PowerPoint user. So I want our editor to look and feel kinda like PowerPoint. If you're a PowerPoint user, I want you to go in Synthesia and within three minutes you feel at home. And this is like design principle called MAYA, which stands for most advanced yet acceptable that I absolutely love. And it's this idea that for people to really attach to a product, start using a product, it has to feel familiar in some way.

Victor Riparbelli:

And if you anchor it in something people feel familiar with, then you kind of get the sort of license to push the frontiers of what you can do. So if you take the very first iteration of the Synthesia, right, PowerPoint style editor, The speaker notes, for example, was the script, right? Because that's naturally, if you look at a PowerPoint, you would have the visuals and you have speaker notes, right? So that's natural place to like put the script, had like the, instead of having a timeline, which is what most people associate with video production, right? Where you kind of scan through it, we said, actually, let's do it as like scenes, which sort of represent a slide, right?

Victor Riparbelli:

So the mental model for people, very easy to understand, how do I use this product? And over time, of course, you know, got more advanced, now we also have a timeline view, but that was like kind of earned us the right to take our customers and take them from being a PowerPoint user, being a video user. For Sessions, the same thing, know, probably still very early, but, and experimenting, but one of the analogies I like for Sessions is meetings, right? So you watch a video, but you join a session. And if you think about the way that we have meetings, meetings obviously are very important part of running most businesses.

Victor Riparbelli:

There's a lot of meetings in the world that's very repeatable and a lot of people have wasted a lot of time sitting meetings they shouldn't be in or just reading out a slide deck, for example. Can we make that more efficient with a format like this, right? And then when you kind of deconstruct, okay, what does that look like? Then obviously people are now familiar with prompting because we all are using LLM tools. The way we kind of set it up, think there we gain advantage from like the macro UX that people are kind of getting used to.

Victor Riparbelli:

But then even when you think about these experiences and like how do you put them together, I kinda think of it as like a Zoom call has two modalities. There's like, you're seeing the other person, right? Which is the, you know, if we're on a call together, then I can tell if you're bored, if you're excited, my, I think under appreciated, your subconscious mind has something to look at. You you need to be occupied by something otherwise your mind starts drifting, you open a different tab. That's a huge part of why we like to see each other.

Victor Riparbelli:

But the other part of it is screen sharing, which is in many calls, maybe even more important actually than seeing each other. And so as we think about this product, right, like the avatar is one part of it, but showing you graphics, to pull up something on screen, drawing a chart, if the avatar is talking about like the pricing of a product, for example, that's what you wanna get to, right? So that people feel more like, oh, this is kinda like I'm joining a call, or this is kinda like I'm setting up a call that I would walk through with someone for like an onboarding or a sales thing. So you can help people with a mental model of just, not just like, hey, here's like an AI avatar and there's an interactivity, you can do something with it. So I really love this idea of trying to anchor new products in something people know, and then you kind of, you earn the right to sort of push the barrier and I think in many ways that's also why ChatGPT took off, right?

Victor Riparbelli:

Because back before, if we were playing around with like the earlier LLMs, know, before the ChatGPT moment as a lot of people were, and it was like API and it was like mind blowing, even though the model's not very good, but it's just like clunky developer tool people try to build apps around, and then the insight was just like, make it like a WhatsApp, right? Like a chat, we're all used to chat, everyone knows how to chat, that's every person on earth is chatting in some shape, way, shape or form today, right? And even the small things, you know, like the UX they built where you see like your messages on the right, the LLMs on the left, it may seem like very small details, but that's the stuff that means that like my mom could go into ChatGPT a month after it launched and understand, like, what to do. Right?

Christoph Magnussen:

Yeah. I love it. I didn't I didn't know that concept, but I love it because it gives this product focus that I think is underrated from people who say, oh, you can vibe code a certain application. Yes, you can, but you still have to think about how you set it up. And there's so much psychology behind it.

Christoph Magnussen:

I'm still struck by the PowerPoint book, but I love it because I I love to understand the details of the texture behind it and then build from there. How do you see it then? Very curious about that when more and more intelligence gets shifted to a model. Let's take the Frontier models and let's take Opus 5.5. I was quite impressed last year with Opus 4.5, how good it was to execute SQL code without making mistakes and stuff like...

Christoph Magnussen:

Okay, this Wow. Tried it again and again, which came to this phase then when all of us founders, I guess, during Christmas built something at home and said, okay, this works now. Yeah. And now you see more multi modality in certain models and models doing more things in the model itself. Yeah.

Christoph Magnussen:

How does this change product design and product development for you?

Victor Riparbelli:

So I think there's a couple of things, right? I think it has been tangible, like what kind of company do you wanna be? Do you wanna be a company that like trains model from scratch? Do you wanna fine-tune your models? Are you okay with just for now at least being an API call into these like big models?

Victor Riparbelli:

And to figure out what is it that you can win on and there's a bunch of different dimensions you can win on. I think there's still a massive gap between X and the people on X and the kind of worldview you get if that's where you're sitting all day and scrolling and the real economy, right? We're a company that, you know, predominantly works with what I think of as like the real economy, which is basically like non tech companies. And when you look at the needs of these customers versus what the big models can offer you, it's like miles apart. They want someone to call, they want a UX that's nice and wraps the work that they're trying to do into something that isn't just like vibe coded by someone on a random Sunday and put into production.

Victor Riparbelli:

The craft really matters, right? And I think we're moving into a world now where shipping code is gonna be, as we know, incredibly fast, it's gonna be incredibly high quality, the quality keeps going up and it's just absolutely magical what you can do today. I think we're gonna see a lot of products where you're gonna make the classical mistake that so many companies have done in the past, which is just that let's just build everything, right? Let's ship everything. Okay, someone asked for this feature, let's put it there.

Victor Riparbelli:

Then you end up with being like the local corner store that has absolutely everything on the shelves, but no one knows why they really go there, right? A lot of the best products in the world are the products who've really honed in on like doing something really good for a specific audience. I think that'll hold more true than ever. And I think the battle will be won, of course, in the business logic of a lot of these things, but I really think a lot of it will be won in the UX and the design and then the distribution. Know?

Victor Riparbelli:

Like right now, I think what you see in AI is just that distribution, being good at marketing is just like more important than ever.

Christoph Magnussen:

How do you see it from an infrastructure perspective? Because what I remember from the early cloud days, I think between you and me, there's like ten years when I grew up, and I can remember first time Internet and then the time that you were referencing to when when Shopify started, for example. Yeah. So I pretty much had a good idea about, like, there's a difference between the static classical website and then this magic button you can press and it loads and everything's fine and technically, what what struck me was how complex it was to keep these apps running all the time. When you talk about your new product, Sessions, that's a totally different product technically to keep it running

Victor Riparbelli:

Mhmm.

Christoph Magnussen:

At scale. Yeah. Like, who are the big players, what do you need to do and what do you see, what is different, I mean you're setting up new offices around the world, I guess to be closer to your customers, but there's also a technical perspective that I'm quite interested in to understand in order to explain people that it's not just a chat window and it's fine or you can just download a local model and you'll be fine. It's a totally different scale.

Victor Riparbelli:

Yeah. I think if you take something like the role play product, for example, right, You could go into Claude or ChatGPT and say, hey, I'm a salesperson and I'm speaking to a customer next week and I'd love to like practice it beforehand. And it would spin up something for you in voice mode. And at first glance it might seem like, okay, and you'd be like, why would I ever buy Synthesia's product, right? But I think that's where a lot of these like AI apps, I think they shine really well on the demo, but for most products, the last 25% is actually really important, right?

Victor Riparbelli:

So what we have found is that with a product like Roleplay, first of all, you need a good harness to ensure that the agent acts in the right manner. You probably wanna tune it to, depending on the use case, if it's a salesperson, if it's a customer support person.

Christoph Magnussen:

You pre tune it also for the customer then or do they

Victor Riparbelli:

change So you can select like different style, you can also custom prompt if you want to. You wanna make sure that from an analytics perspective that you have something that can be unified across like many multiple role plays, You need to have enterprise security. It's pretty sensitive, right? And you need to send it out to 800 people in an organization. How do you get that distributed really well?

Victor Riparbelli:

How to report back on it to the enterprise managers and executives who wanna see the data? Like all these things is like, you could say, yeah, could just vibe code that, but at least right now, I think that's like very difficult. But even if you could vibe code it in a short amount of time, then taking this out to market and selling it right is still really hard. And I think, I tend to think that actually that sales teams will be more important in the future, not less, because I think if you believe in a world in which software will more and more converge and it'll be easier and easier and the kind of quality differential between different products may kind of be lowered very significantly, then I think the teams that are winning are gonna be the ones that have like great salespeople, humans that can go out and help these companies who are not vibe coding themselves, the real economy, right? They want someone to help them, guide them through it, and teach them how to become like an efficient AI company.

Victor Riparbelli:

So I do think a lot of the value, it's kind of like whatever is scarce becomes more valuable in society, right? And up until now, code has been scarce and building a high quality product, the talent required to do that has been really scarce, but if that begins to change, then we'll like value something different, right? So, I also have this thing I say, what you think is that, the great DNA or the DNA of a great company today is great people. I don't think that's gonna change even ten, fifteen years. It's so easy to imagine 10 jobs that gets like lost to AI and it's so hard to imagine 10 jobs that gets created by AI.

Victor Riparbelli:

It's just how our brain works. It's a very like natural thing that's hard for us to imagine. And this has been true throughout every historical evolution of technology, right? People say exactly the same thing every single time. And I think right now to your point earlier, a lot of companies are in this like bridging phase where it's like, okay, we have these like 25 processes that run the company, how to make them more efficient with AI?

Victor Riparbelli:

That may mean you can save some headcount somewhere in there, But I think if you just continue doing what you've always done, you're gonna get out competed, right, by your competitors who will go much deeper on different products, they will massively build out their go to market marketing machine with much more automation than before, maybe they'll build new products, they'll spin up new entities. Like they're not gonna stand still, right? If you look at the pace of Frontier AI companies, I mean, we've launched like four products in five months now, right? That needs people to actually get that done. And yeah, people work in a different way, I think we...

Victor Riparbelli:

It's very hard for us to think in these like exponential curves. Right?

Christoph Magnussen:

It's almost impossible to think of that even if you know it in a theory, then you say like, okay A 100%. It spins off and then at a certain moment you think, okay, that was very, very fast. I mean, like, as you said, we we played around with the early platforms and and I saw, like, text generated by a large language model and wait ten minutes for the answer and everyone was like, that's magical, but it's a totally different thing if it works in that speed and I'm just thinking of

Victor Riparbelli:

We just launched this, like, actually this this Friday, I don't know when this goes on, but we just launched a product called Syren

Christoph Magnussen:

Mhmm.

Victor Riparbelli:

And it's a great example of...

Christoph Magnussen:

I've seen it on LinkedIn when you posted it.

Victor Riparbelli:

It's a great example actually of, like, the kind of need to always challenge yourself and reinvent the company, or reinvent at least some part of it. So historically, the way you've been creating videos, today the way you create videos in Synthesia we've built an editor that edits a bunch of like technical primitives. You can add like shapes and text and animations and a whole bunch of stuff that sort of makes up the video. And that's worked really great. It's been kinda like the PowerPoint type ish experience.

Victor Riparbelli:

Of course, you don't need to record people. It's a lot of like automation, help you write the script and do some simple design and so on. But then what we've working on for the last twelve plus months is our own rendering engine, which works from code. We're actually building it for code always, but twelve months ago we'd made a very conscious decision. So like, let's take our rendering engine, which essentially is web technology.

Victor Riparbelli:

So if you think about like all the animations, the shapes, the text, all sorts of stuff, right? How do you actually render that? Do render like a C plus plus backend somewhere or do you render it basically as web technology? So we began building out this web technology render and then we started experimenting with what if instead of having the kind of editor as a constraint for how we can help you create video faster, What if we just gave an LLM direct access to our rendering engine and it could learn how to use it, right? And because it's web technology, the LLMs will have seen a lot of different examples like how do I animate a square like because tons of examples of that online.

Victor Riparbelli:

And so the big breakthrough here was actually taking Opus 5.5, which is specifically good at this stuff, right? And say, why don't we try and just like give it access to our rendering engine, let it figure out itself? And the results is just, I mean, it's absolutely crazy. Like you can ask it to create a video about absolutely anything and it will create a video that I will, you know, if you took this video back twelve months ago even, you'd be like, does someone paid like an agency 10 k to do this or does someone who really knows video editing has been sitting for

Christoph Magnussen:

a while. That's what I meant by like, if more goes to the model Yeah. At a certain point, you need to change the product. So like, what happened then to like now the strategy of your rendering engine?

Victor Riparbelli:

Well, yeah, so I think basically the way you want these things to work, right, is that you have, we have a proprietary rendering engine. It's like faster, lower cost, works within some easier constraints, and very soon, right, we're gonna connect these to the two way video. It's not just gonna be the one way video, it could be, okay, you wanna train, I wanna train you on whatever topic in my business, how to sell the new product we've just launched. It can just figure out, okay, the way I'll do this, probably I'll just like create a video for you to learn it first, then I'll stop the video. You can ask the avatar any questions and the avatar can, you know, do it back.

Victor Riparbelli:

It can pull up like visuals if that needs to answer the questions. And after that, we'll throw you into a role play just to make sure you've actually understood the content right. That's where this stuff will eventually go. So this is more like you can use the LLM to sort of orchestration and it can pull like different strings. Okay, maybe I need a video over here.

Victor Riparbelli:

Of course, we also offer like third party video models inside the app. So we have our own avatar models, which are really good avatar style video. We use, you know, all the other video models out there if you wanna do something that's more like creative, like, you know, Seedance, for example, is one of really big models right now, which is super powerful. So I think it does move from kind of being orchestrating different primitives into an experience, and that experience is gonna differ based on who's watching it, when they're watching it, what you as the person setting this like automation up, how do you, what is it that you wanna use this for? And I think that's like the, there's gotta be like this, you know, period of time before this is like all figured out.

Victor Riparbelli:

There's a lot of experimentation right now, but when I think of this, what is truly AI native video going to look like? It's It's probably gonna look something like this. And I think it's gonna end up morphing into being almost more of like an interface. If you think about most agent experiences today, they're text driven, bit like we use command line in the eighties. Some of those are voice driven because voice technology has matured, you know, much faster.

Victor Riparbelli:

A lot of the experience will of course be better in those two buckets. Like if you just wanna know if your order is on the way, you don't want an avatar to pop up in your face, but a big fancy video around it. If you wanna do a quick customer support thing, maybe you prefer that as being just voice. But if I'm trying to train someone in my company on like how to conduct a multimillion dollar sale, it's not gonna be a voice experience, it's not gonna be a chat experience. And maybe parts of it, modalities of it, right?

Victor Riparbelli:

But how would you do that today? Well, you would have your enablement team, you'd probably have me, you'd have a CRO, do what? Jump on a one hour Zoom call and bring some slides.

Christoph Magnussen:

And there's a total difference in like, when I think about voice models, there were not very many real duplex models being able to do certain things at the same time, like researching stuff then answering at the same time, and if you think of a native video model that has the capability that you were just talking about reacting to a situation that is not foreseen or clear and not just by words but by a whole scene that needs to be much faster in generating everything in the background when it... Like, technically it sounds very very complex and at the moment I guess you're like... As you say, you orchestrate things together. Mhmm. Is there anything you can give out to the public saying, like, okay, this is what we're experimenting on because you mentioned stuff like c and c plus plus saying, okay, whatever, we give the model access to the lowest layer and it builds it from there.

Christoph Magnussen:

I don't know, how far do you go in experimenting?

Victor Riparbelli:

No, we go pretty far and I think we made a conscious decision not to give it like C plus plus technology because that's just hard to run in a browser, right? And if you want these things to be able to render in a browser at the speed of like a webpage, basically, you have to build a web technology. And so our bet is that web technology is actually gonna be the new video technology. For a very long time, it's been very hard to build like great creative tools in the browser, right? Like you've seen Canva did a good job.

Victor Riparbelli:

Figma was I think underappreciated by people who are not into, but Figma was like a technical breakthrough that was huge. True. Being able to create like such a smooth, nice editing experience was amazing, right? Yeah. But Figma was still constrained when it came to three d because three d is much more intensive than designing web pages, which are generally like kind of fairly flat.

Christoph Magnussen:

My kids want three d images all the time and my tries always end up that Codex is trying to use Blender and then make it.

Victor Riparbelli:

But you should try using ours actually. So we use Skia under the hood which is a web technology for doing three d rendering as shaders and everything And it's really, really good, right? So like the other day I made a video, for example, visualizing how to assemble a piece of IKEA furniture. It's a task, right? But I basically just gave it the instruction manual as a PDF file and said, give me a three d visualization of how to assemble this table and it did it because it has Skia under the hood so it can go in and actually create the three d pieces of it, it can put shaders on it.

Victor Riparbelli:

And so I think you, we just made actually like many, many years ago, I just had this feeling like web technology is gonna be all technology very soon. And yes, right now, back then especially it's hard to do things like three d rendering if you're doing it in, it's gotta be very simple stuff, but that feels advancing really quickly with LLMs beginning to, you know, massively accelerate development of those open source libraries. We're gonna be at a point pretty soon, I think, where you're gonna be able to do some pretty crazy things in the browser.

Christoph Magnussen:

What have you seen on the exponential curve reading a research paper where you said, holy smokes, if that works, that would be crazy.

Victor Riparbelli:

Mean, honestly, right now it just feels like everything is gonna work at some point. I've even gotten out of reading as many research papers. I used to spend a lot of my time and especially around like video. Yeah. But I think now it's just clear that these like very big and expensive run models are just insanely capable.

Victor Riparbelli:

And so I think, but a lot of the things we're thinking about is on the kind of cost curve, how do we take these models down? Do we need to, yeah, basically how do we run these things like at scale? I'll give you one example of like, when this works, it's gonna be crazy type of thing. So we launched our developer platform, which is real time avatar APIs. So if you wanna build any experience where you're speaking to an avatar in real time, you can take out primitives and build whatever you want around it.

Victor Riparbelli:

It's the same technology we use in our Sessions product for like role plays and surveying. This is just the developer platform.

Christoph Magnussen:

Right? So I could put it to my web app, the use of KI-Notes, for example, put an avatar out.

Victor Riparbelli:

It's very simple. You could just connect to an LLM and give the LLM a face. Nice. Alright. And, you know, that technology has matured like quite a lot.

Victor Riparbelli:

Mhmm. My sense is that if you look at the voice market, then ElevenLabs beautifully, beautifully captured the right moment in time where voice models went from being like, okay for your GPS, to all of a sudden being passing through the quality threshold and all of a sudden it's like, now you can do audio book, you do anything with voice, right? That coupled with amazing execution was a big moment for them. And what happened in that space was that people went from building these like much smaller models with lots of like symbolic AI within it. You know, you take like a hundred hours of like super high quality voice and you would train a model on that.

Victor Riparbelli:

There was a model called the Tortoise back then, and Tortoise was kind of somewhat completely outside of the research community. It was like completely unknown person. And he was just like looking at what's happening in LLMs, why don't we do the same for voice? So he just took, I forgot about it, back then it was like probably 150,000 or something, which back then was a massive amount of data to do these things. You trained a model on it to do voice.

Victor Riparbelli:

And I remember the first time hearing that voice, was just like, this is gonna be the answer. Like the level of like naturalness and intonations was insane, insanely good. But most people in the field were kinda like, yeah, it's kind of cool, but you know, every six word is wrong. It's completely uncontrollable. Like all the classical things you hear, right?

Victor Riparbelli:

Like why it won't work. And I kicked myself for not forcing my team to go all in on that, you know, nine months before we did, because I actually feel like I saw that moment, but easy to say in hindsight, but basically those kinds of models, right, then obviously became better and better, people with more data and that they learn how to train them better. And now you have, you know, state of the art models, like we have our own voice models, these are ElevenLabs, obviously, you know, one of the best in the world that are extremely high quality, they're fairly controllable, this is a great product. I think for video, we're still kind of getting there. The quality is a lot better than it's been before, but I think real time interactive video for like a really broad set of use cases, the latency is gonna have to go down a little bit.

Victor Riparbelli:

The cost is gonna have to drop dramatically if it's gonna be like consumer level and the quality has to go up. So I actually see the real time avatar space right now in this phase of a bit like the second chapter of Synthesia when we actually made something that worked for the enterprise, where the technology is now good enough for some use cases, but still fairly constrained. That's what we're building our own use case around, but very soon, and I think it's gonna happen within the next six to twelve months, we're gonna see these video models first, they're gonna like blow through the quality threshold where you're actually, they're gonna get so good that it's gonna be very difficult to tell if you're on a Zoom call with a real person or with an avatar, which has a lot of implications. And then after that it's gonna be the cost piece, right? And the cost piece is gonna, because it's a video, is still like a much, much higher than text-to-speech, but at some point, right, when the quality's gone up and the price has gone down very significantly, then I think we will start to see that like video, both avatars, but also in general real time generated video will just take over so many interfaces, which today are more kind of like text driven.

Christoph Magnussen:

What is the thing product wise from your own product lineup that you use internally heavily where people naturally are driven to and say like, this is what we use.

Victor Riparbelli:

So I mean, I would say all of the products, but the Syren product we launched, I mentioned launched it like five days ago from when we were recording this, totally insane. I think we're creating five, six, seven times as many videos internally as we were before. And what's exciting about it is not just that, of course when we launch new product, we use them ourselves, we wanna get feedback But on this is a fundamental shift in that we're now seeing like teams that before wouldn't really necessarily have a use case for Synthesia videos internally actively being like, I'm using this now. You know, like for internal comms, for customer comms, just seeing like everything pick it up. And it's because it was fast to make a Synthesia video before, but it says to make a good Synthesia video with the kind of core product, it would probably still take you maybe fifteen, twenty minutes or something like that.

Victor Riparbelli:

Now, I mean, it's like thirty seconds, you know, if you wanna visualize, just before I came into this meeting, I have two fun examples. Before I came to this meeting, our go-to-market team, obviously do these kind of quarterly business reports and so on, and someone just like literally just took a screenshot of the dashboard, dropped it into Syren to make a video that summarizes the data, and it creates this like sixty seconds long video that I watch on the way in here. Great, and I actually retain more of that information than I just like seen the dashboard once and like gone away. This Tuesday, I went to Downing Street to meet Andy Burnham, the new prime minister, which was a great experience and before I do that, of course, my PR team comes with heard like some things to chat about here, like some recent perspectives and they come with this like long document that I didn't read, right? And then this time they're actually like, here's a video instead.

Victor Riparbelli:

And I watched the video instead and it was generally like a much better experience for me at least, because just remember more of it, because I'm seeing a video and it's more engaging than like reading for a two page document. We're seeing so many of this internally, like Martin Tschammer, our head of security at Synthesia, he has like a fitness program and he just took his fitness program, like in verbatim text, dropped it into Syren, and then Syren created this like video with an avatar, like actually doing all the exercises, showing how they're, how they should be done, and it just looks like it's like a video that's created by someone. There's lots more examples, but this, it feels like one of those like, you know, very genuine breakthroughs where like it's now so, a bit like ChatGPT actually in the early days, it's like, it's insane how I can type five words.

Christoph Magnussen:

And you get something that you actually use.

Victor Riparbelli:

It's like you say, build me a research report on something and it comes back with what would have taken yourself like ten hours to do. In a

Christoph Magnussen:

in a field where a lot of people consume video all day long and are very used to that.

Victor Riparbelli:

I think it's just like video is becoming table stakes. In in our private lives, for most people, obviously the younger you skew, the more true this is, it has become table stakes. People consume video, podcasts, way more than they read, right? And there's probably pros and cons to that, but that's the reality, but in the business world that hasn't really happened yet. And I think, you know, what we are starting to see is just like once that, once the kind of the barriers truly collapse and becomes actually as easy to do video as it is to create text, which is what we're seeing with this new product now, that massively changes kind of like that equation, right?

Victor Riparbelli:

And the other example I'll give is we're launching a lot of new products internally right now, as I've launched four new products in the last four months, and so using role plays as a way to get, we have a very big go to market team, right, who have to learn things all the time. It's an incredible tool for people to practice before they actually go on a customer call. And it's an amazing tool for the enablement team and for our CRO to get a view into, a dataset into how enabled are people actually, because this is a question historically you've never been able to answer, right? And there's so many products or companies out there, right? Where you build a new product and then you go out to the market and try and sell it, and it's not working.

Victor Riparbelli:

Is it because the product is not good or it because the salespeople don't know how to sell it, right? That's usually a question that companies ask themselves. It's very hard to know, right? But with this, because we actually built you a graph of how enabled are the people in your company, when they go on these role plays-

Christoph Magnussen:

After the role plays.

Victor Riparbelli:

Yeah, and these role plays are generally like, I think the average session length is like eighteen minutes, so quite long, right? Then you can actually assess like, do people actually know what they're supposed to know at least as a baseline? But as before, right? It's like, hey, watch this video, do this course, do this multiple choice thing. I don't think anyone believes that a multiple choice at the end of watching a video tells you much about if people understood the content or not, but if you're sitting and you're speaking eighteen minutes with an avatar, like you cannot really cheat on that, you know?

Victor Riparbelli:

Like that's... It's very hard because it's like being in a real conversation.

Christoph Magnussen:

So could you also like work with the developer API and keep your own training platform continuously improving with that if it's strongly connected and as a living product because what I'm thinking of like when you read this old book about PowerPoint, you see that struck with me. Yeah. I watched very old videos about first training IBM machines, telephone back in the day. Really enjoy that. Yeah.

Christoph Magnussen:

And they are long. They are two hours long. But there are many small details in there where they put a lot of effort in these details because it was like very intense to film these things and do all that stuff. Yeah. And I would be curious to to know like if you would have the developer API, could the platform itself then keep improving if I build it and say like, okay, there's new stuff coming and I see what people like, I see the graph Mhmm.

Christoph Magnussen:

And with that, my own videos that I produce as a customer of yours would improve itself?

Victor Riparbelli:

100%. So that that is... We're launching that like pretty soon where the idea, I think what you wanna move to is a world in which it's actually, it's almost like you go an abstraction layer lower than the video. So here's like an artifact of some knowledge, right? And this knowledge will change over time, but it's an artifact.

Victor Riparbelli:

And then when you wanna visualize that, you could do that like in real time. So maybe you want like the thirty second quick nugget before you go into your sales call because you've, you know, seen the videos before. If you're onboarding, maybe you want the five minute version of it. If we know you're struggling with it, maybe we wanna put a role play at the end of it to just make sure you've actually understood it, and it becomes more of a sort of agentic upskilling loop, right? I think that's for sure like where the world is heading.

Victor Riparbelli:

And I think what's very exciting is that we've had these breakthroughs now in of course LLMs, but also like now in video models, both in like the video editing, which is kind of what our latest product essentially is really good at, is doing the editing for you, the core like pixels of what you can create, and then just like interactive formats and real time formats where if you take that knowledge artifact, right, and you wanna use it for something, the agent can then figure out like what way to use it. Should I build a role play? Should I first show a video and then a role play? Should I first survey you maybe to collect more information around, hey, we know you're struggling with this particular thing. Tell me more about what you're struggling with and I'll put together the perfect materials for you afterwards.

Victor Riparbelli:

Or if it's a sales context, say, let's connect to your call transcripts and look at what are you actually struggling with in real calls and then say, Hey, you know, next time you're pitching a customer on Synthesia for automotive companies, let's just do like a quick five minute exercise to make sure you're prepared for that, right? Maybe there's two case studies you really should know. The test you've actually done them, and I'll visualize them for you and how this could be used in this call. So then it becomes this like communications machine that's always on, and right now we're talking a lot about like upskilling and workforce training, etcetera. That's something that's always been very close to Synthesia, but I think this also kind of gonna work in external use cases as well, you know, with a customer.

Victor Riparbelli:

Like we're right now building, I think a very interesting, like one of those things that I feel is like on the frontier of the curve of what today is, a lot of people are doing it, but in like five years is gonna be the default way of interacting with the website. Today website is like a brochure basically, right? What these guys are building is an experience where you come in as a physical product that they sell in stores and you talk to an avatar, right? So they ask, you know, like for that particular category of products, a bunch of questions and based off that, it'll show different products. So So kinda like if you went to the actual store and you spoke to someone, they would ask you a bunch of questions and say, well, we have like this thing up here, or we have this thing over here, and it becomes more of like a continuous conversation, right?

Victor Riparbelli:

And I think right now the technologies are like early, but very soon they will not be early anymore. And this will truly feel like you're just sitting and talking with someone, everything is snappy, the latency is low, you can get to see the product. If it's a couch you're buying, right? Like, okay, great, open your camera and show me your living room and boom, let's create a picture of the couch and live, like the whole experience I think of like shopping and using many digital tools, it's just gonna become like this sort of video first experience. It's hard to imagine what it's gonna look like today, think, but I really think that's gonna happen, right?

Victor Riparbelli:

And if you look to Asia pre AI even, right? Yeah. It's just, it is really interesting how video is just default for so many things. Right? And live video in particular.

Christoph Magnussen:

I can... Like, I can see it with my kids. As I said, like, ten years between you and I and I know for them, it's so normal that they have a video image always, all the time. They don't wanna... They want to talk, they want the one video.

Christoph Magnussen:

And what I still see at the moment and then that stuck with me is this the adoption and the upskilling part in companies is a huge thing, but not only in companies, and if you can if you can create an experience that actually evolves and you combine that with the intelligence of a model and you get that, I I definitely see that. And I...

Victor Riparbelli:

Even with kids. Right? Like, I think Yeah. That's not the market we're targeting initially, but I think how amazing will it be when every kid has not just an LLM, which today is like an answering machine, which is already very helpful as you use it correctly. But if you actually have more, someone who's more like a tutor, Yeah.

Victor Riparbelli:

They'll also be like, hey, let's test if you've actually understood this stuff. Did you just prompt it and I gave you the answers or have you actually understood it right? Is a big shift

Christoph Magnussen:

of doing a heavy, heavy video program and LLM program in schools at the moment and upskilling education there heavily and they combine it with people. They don't do like you just do that and in Europe I have the feeling we always say like, oh no, we can't do it because it will replace teachers and it's not. It's like taking part in the whole journey and it's a much different experience.

Victor Riparbelli:

Sure. It's so interesting the whole... I mean, I think the EU is just... I think it's just depressing like all the stuff that's... I really think the current politicians are just destroying our chances of success, unfortunately.

Victor Riparbelli:

I think there was like last year, I think it was a bit better. I think now they're back to like, let's over regulate everything and kill anything before it even starts. But it's very interesting with something like China, right? Because it's such a tech positive society and I'm not saying that's like purely a good thing, but that's the reality, right? They really love tech and they don't view tech the same way they would do in the West.

Victor Riparbelli:

This is like guilty pleasure, you shouldn't really be doing it, etcetera. But they're also very good at enforcing rules, I think actually matters, right? So there's this like, how long can you use TikTok? What kind of content can kids see on TikTok? So on one hand, they're like very tech positive and they want people to use a tech, on other hand, they also seem to be much more forward in like, let's shape how we use these technologies.

Victor Riparbelli:

And I think our generations, I think it's gonna be part of this generation we look back at as the kind of strange middle generation from like tech started moving really fast and we've got all these new tools and we had to like figure out how to use them and what are the right modality for using social media, for It's just because everything's moving so fast. I think we're now struggling with how we use these tools to the best of our ability and it feels like a country like China is actually ahead of us in many ways in thinking about like, how do we actually use these technologies in there?

Christoph Magnussen:

They're regulating at the moment also the social interaction of avatars, for example, for kids. Yeah. I think that would be very interesting to continue at a certain point, but we're hitting the hour now and your your hard deadline, but we should talk about that at another point, and I will also definitely look into the developer side for sure because I mean, we are in adoption for more than fifteen years, but I know behind the scenes that when my team and your team were talking, were already like, okay, we gotta do something so I will play around for sure. Amazing. And, yeah, last word maybe from you, you were talking about this book, not the PowerPoint book, but the MAYA-like method.

Christoph Magnussen:

What would you recommend for someone listening saying, hey, wanna start something and it should be product based, where to start? Not to get lost in vibe coding and and the avalanche of things happening, where to start?

Victor Riparbelli:

I think it goes back to the basics, which is what I think people often forget when tech becomes as exciting, which is like, what is the actual problem you're solving? You know, I have a lot of friends, a lot of friends are not tech people and they all come to me with these, like, I could build this app, it's a great thing. It's like, okay, it's cool, but does it actually solve a problem? Are you inventing a problem that you don't even have yourself or you think other people have, right? So making sure you're building something that's valuable.

Victor Riparbelli:

And I think often a good way of thinking about that is that people are creatures of habit, right? And we, most great products tap into something or some behavior that exists. It may be that the technology morphs it into being a different, a much more exaggerated behavior, but it taps into like some existing behavior that we already have, right? And I think you wanna make sure you really understand like, what is that? Or are you dreaming up that every human on the planet is gonna like massively change their behavior to something that they're not doing today?

Victor Riparbelli:

And I think that's where the MAYA principle is really good because that forces you to then think about, okay, what does that mean from a product perspective? For us, it meant that people create PowerPoints. That's a very tangible thing I could then attach my product to, right? Another example of this is like the early iPhones, all of the icons, like was what you call skeuomorphic, which means that they were kind of depict, like the phone book for example, was an actual picture of like a analog phone book, right? That's not true anymore, but back then that was a part of like, okay, people need to learn how to use this like UX of like these small apps, which today is like clearly obvious to most people, but there's gonna be a generation of older people.

Victor Riparbelli:

And so they built it as in like the actual physical objects that they were making a digital version of, right? I think it's just like such a good, I'll give you another example, WhatsApp. There's a lot of like people building messaging apps before WhatsApp, you know, really came out. And I think one thing it did really well is just like your username is your phone number. What a stupidly obvious thing, but it just means that you're like, it's an SMS, It's not like I'm going on this, especially back then, this like weird internet messaging app, which is like people associated with being like, for like strange geeks in a corner who's like playing Dungeons and Dragons.

Victor Riparbelli:

Oh, it's just like a free SMS. Most people don't even know that it's a username, but it is your username. Has nothing to do with your actual phone number, right? But I think these are the small things that like in design just matters so much. If WhatsApp hadn't done that, I don't think WhatsApp would have been what it is today.

Victor Riparbelli:

You take something like a Slack. I think they are actually building a game first, right? And I think what they did was they basically took IRC, I don't know if you've owned IRC Yeah, I've heard When I was a teenager, everyone was on IRC, right? For the game shit. We have these channels, have these different servers.

Victor Riparbelli:

And I was like geeks sitting on that, but I think what they realized is actually this is like a very powerful modality for communication. And they basically made the corporate version of IRC, and so they could anchor it. So I think it's just like all good products have some level of insight that like taps into this.

Christoph Magnussen:

I love that. Thank you so much, Victor. That was very... Not only very insightful, I loved it. I love the speed.

Christoph Magnussen:

I love the intensity and energy, and I hope that we repeat that.

Victor Riparbelli:

Me too. Thank you. It was great. Thank you.