AI and the future of video

​​As quickly as AI changed the way knowledge workers write and do research, it's changed how we make videos.

To learn more, we teamed up with Sandhill Markets to host a panel conversation about AI avatars, text-to-video models like Sora and Midjourney v6, and AI-powered editing & production with:

Key points from the conversation:

Transcript below.

Adam: Alright. We're here again with an awesome panel, help from Sacra. I'll give them some thank yous later as well, but super excited about this one. It's close to home for us because we've been kind of messing around with video for a while now. If you've been here and been a subscriber for a long time, you know that before with Sandhill, it was Stonks, and we were a live demo day platform. So we've been kinda hanging in and around the video space for a long time. And definitely believe in the power of it from, you know, building trust, selling things, you know, its role on the internet more generally. So super excited to host Chris and Hassan who, I think, agree with the role of video and the importance of it, and we'll get their take on how they're building with AI.

Wistia has been around for a little longer, founded in 2006, compared to Tavus, which is a YC’21 company. So I'm really interested to hear about timeline stuff, you know, how they're using it, how they're not using it. It was one of the interesting questions, but we'll get into all of that in a second as I kill time and let everyone jump in.

So before we do that, the typical and normal stuff that I do, this is me. If you've never seen us before, sandhillmarkets.com, host a bunch of this stuff. We have a newsletter that goes out, not even close to daily-ish anymore, as we focus on what we think are the highest value sends, which are, you know, telling you about events like this, telling you about when events like this are happening, and we have a big syndicate. Which goes back to what I talked about from back when we were a demo day platform. And now we do it more kind of traditionally async, you know, sharing deals with that accredited investor group.

I have to, as always, oh, actually I think I missed it. But sacra.com, if you haven't seen them, there's it's missing a slide today. Great crew, Walter's been great. Always helps us pull these together. And if you haven't checked out Sacra.com before, I would encourage you to.

But as we get into it, our guests today and our very lightweight bios, I'll let them give the real kind of intros. It's always more productive that way. But Chris is the CEO and co-founder of Wistia. They founded in 2006, over 350 customers, 375 is a number that I saw. So I won't undersell him, multi-million dollar revenue is an understatement, but 100+ employees, I think, is about right.

And then, our second guest here with a tight panel of 2 is Hassan, CEO and co-founder of Tavus, and I'll let them kind of explain what they do, but they were founded much more recently in 2021. They're an API, more developer-focused vendor which is more enterprise video focused. And they were, yeah, like I said, YC and Sequoia backed, and then recently actually raised a Series A a few months ago. So, again, newer kind of more AI-first, let's say. I'll have them kinda comment on that, but really excited to get their take on this wave that we're all experiencing.

So with that, the first five minutes are killed. Chris, Hassan, the floor is yours. Chris, you want to give us your brief?

Chris (CEO & Co-founder of Wistia):

I don't think you want the long version even though I like to sell it. Yeah. So, basically, we're a video marketing platform. Our goal is to make it really easy for anyone who's in marketing to get to take advantage of the power of video and do that confidently.

So for a long time, that meant for us that we were video hosting analytics. That's really what we were until, like, 2022. We'd stayed hyper-focused on that. And really built like the world's best analytics and integrations and stuff to give you the insights on, like, how your videos are performing? You know, so who in your audience is watching? What are they skipping? What are they rewatching? All stuff. Like, a player, you have total control over SEO benefits that go to your site, all that kind of those types of things.

In the last couple of years, we've been adding in recording tools, creation tools, editing tools, editing so you can edit videos via editing the text. We have a webinar project we launched about a year ago. And the goal across all of this is to make it really easy and simple to work with these different tools in one platform.

So for example, you do a live event with us. It's recorded in Wistia. What we've seen is about 100% of webinars are edited. People take out the, you know, mistakes. They take out the beginning. They take out the end. They take out Q&A, all these different types of things. Also wanna repurpose it. So it's definitely gonna download and put it into an editor and then save it and render it and then upload it back and put it back in something like us.

The goal is to make it so that you don't have to do all that extra stuff. It's just in there and it's easy. So that's what the product is. The company was in-person pre-COVID, is very remote now. We actually have about 180 employees and millions of customers, so it's an old bio, I guess. And that's fine. 

And then the thing I would say that is, like, might be interesting to this crowd is, so we never raised venture capital. We raised angel money in the early days, raised $1.4 million. And then in 2017, we had the opportunity to sell the business, decided not to, and did a leveraged buyout with debt to buy back control.

So we raised $17 million of debt, bought back control from our angels, converted all their shares from preferred to common, and have since paid off that debt. And so we just we, you know, we're really right that video is gonna happen and business is gonna happen. We're really wrong on the timing. And that was a way to get us into a better position, so that we could own our destiny.

Adam: Yeah. Awesome. Super cool. Hassan, you wanna give a similar rundown?

Hassan (CEO & Co-founder of Tavus): So I'm Hassan, I'm the co-founder of CEO here at Tavus. And Tavus we're a generative video research company specializing in creating digital twins. And, what that means is we essentially build really cool models to allow users to leverage their own likeness.

So whether it's generating a video from text summaries or even real-time interactions, we essentially facilitate developers to build these really cool human-like video experiences. And, that's what our vision has always been - how can we help people leverage digital likeness at scale?

But the story actually starts 4 years ago before generative AI was super cool. And it started in the world of personalization. We saw that personalization was really effective at increasing engagement, you know, and basically increasing someone's likelihood to want to interact and watch content. But personalization is very difficult. Right? So sending a one-to-one video, you know, just took time and effort and people didn't really like being on camera.

And so we built AI models to personalize at scale. So you can create one video, and then we could personalize it for a lot of different people. And, you know, the evolution of that is, well, if you no longer have to get on camera to record anymore, you could just put in text and it creates videos for everyone.

And now we're really focused on empowering all sorts of video platforms, video editors, or software suites to actually allow people to create videos without having to record it anymore. So that's a little bit about us. We are based in San Francisco. We're a team of about 30.

We are in-person now. We recently got almost everyone to come back into the office. And, yeah, we're in the Mission in San Francisco. And, yeah, it's been it's been a fun ride. We went through YC in the summer of '21.

We are venture backed. So, you know, we raised our Series A, you know, late last year. Yeah, that's a little bit about us.

Adam: Cool. One of the one of the first kind of big things in, like, the theme I wanna talk about is on the company building side, like, within the frame of AI. And I think it's really interesting. The different perspectives here where where Chris obviously, you know, '06 when you started Wistia, I don't think the word AI was probably, like, in your, like, what you were thinking about. It's really it's a big part of it now.

Whereas Hassan, right, like, it was always, like, about, like, that was always something that you thought could be a big part of what you're powering on, like, how you're getting concerned?

Hassan: Absolutely.

Adam: So so, Chris, as from, like, a company building perspective, with AI? Like, how is AI a part of Wistia? And how is it not? Like, is it a tailwind that you're experiencing? Is it a "we had Wade from Zapier and and Des from Intercom on, and they were very much like it's about the company moment for them."

How do you feel about it, Chris? And and how have you, yeah, been been taking advantage of it?

Chris: Yeah. So I think it is - I think if we were not, maybe I'll describe what we're doing first. Yeah. And then, yeah. So, basically, we had to make a call. You know, how should we how should we think about AI? Like, we were starting to actually play with GPT-3, I think, before ChatGPT. This is some close to fear, but is this is the market ready for this? And then ChatGPT happened. Okay, the market's ready. People understand what this is. Yeah.

And we decided that, like, our strategy is gonna be we we really need to use it everywhere. And we told every team, like, we don't have an AI team. We've told every single product engineering team, like, you you need to look at AI as a tool. You need to use it wherever it makes sense. And there's gonna be a bunch of places where we think it's going to save us enormous amounts of time or let us do things would be able to do otherwise for our customers.

And so it's been incorporated all across the product, like, in foundational ways in the sense of, like, AI has changed transcription dramatically. So we've always had transcription in Wistia, but you know, you used to have to pay for it and it, like, it was hard to guarantee really accurate transcriptions without it without requiring people, which we actually worked with a partner to do. So you could go with to click a button 100% accurate. It'd be like, AI plus people.

And we realized that the cost was gonna get there. So the cost would get low enough, and it was fast enough and good enough that we could actually start including it for free for every video. And what does that mean? It means from like a metadata perspective, we're starting to be able to give you insights where we understand what your video is. In a way that we couldn't understand before.

And it's enabled things for us like, you know, we were already working on this editor. We've wanted to work towards having text based editing, because we thought that was just a much easier way to edit a video is have the transcript and you know, delete words. Right? And so it was actually very fast for us to enable that.

And there's there's just a lot of different places where we're adding AI through the product, and we to do it more. I think if we weren't doing it, I would be worried. I think that would be very stupid. It's not yet a moment that I've said, like, you have to bet the whole company to do it. But in a sense, like every road map, every every plan has some AI in it. Some some have huge amounts are doing things they couldn't do. Otherwise, it's and some have less.

Adam: And and that's both internally and and, like, in terms of company processes, like, you all building products, faster, better, as well as

Chris: It's everything. It's how I build being

Adam: experienced by the user as well. Correct.

Chris: Yeah.

Adam: Yeah. Now I have another all the follow-up because save it for after. But so Hassan different, right, like the AI in terms of and I think you were you were in AI before Tavus. Correct. Right.

So this has been, like, your kind of world and and built off of that. Obviously, when you think about, like, building a company with, like, kind of this AI first in mind rather than adding it on. What when you just heard that from Chris, like, what stands out as different, or is it all kind of well, obviously, you didn't have, like, the product ready and then AI moment happen, what what jumps out as as kind of, different approaches or or or is it all pretty similar?

Hassan: Yeah. I mean, I think it is important to to really step back and try to determine, like, where does AI actually deliver a better experience, and where does it not deliver a better experience? And I think that, you know, there's we've seen in the market there's, like, these directives, like, okay, we have to be, like, an AI first company now, for companies that traditionally weren't. And, you know, and so they sort of just, like, throw everything at the wall.

And I think there is a thoughtfulness of, like, when do we, when should we leverage AI to build these better experiences? And I think there is a lot of opportunities. And, you know, at Tavus, a lot of our partnerships and the customers we work with is helping inform them of, like, what are the possibilities and where can we actually you know, provide, like, meaningful improvement.

And so I think - I think, like, the the approach of, like, being thoughtful, I mean, especially if you have users that have workflows that work. It's, it's maybe it's maybe not, like, in your face. Like, oh, we changed everything and now it's, yeah, it's maybe actually, like, things like, you know, Chris talked about, which is like transcription, which is like, oh, it's it's just it's making things a better experience that, that, like, that, like, just sort of seem like it just happened rather than, like, in your face.

Now you don't have to, you know, create a video at all anymore. You you don't have to join an oven hour anymore. There's a place for that, I think, but it doesn't have to be, like, this, like, complete turnover. Right?

Adam: Yeah.

Hassan: And I think from - we build lot of products and models to support a lot of different things there. Right? We have, like, you know, especially earlier on our days, you know, we were building tooling to help, help you blend videos if you, like, cut in a, like, a piece in the middle. Right? It would automatically blend it over. So it didn't look like there was a cut, right?

We would automatically replace words. And so, you know, that, that was sort of the, hey, if you have people that are way more, involved in video editing, then you might wanna use that model versus we have the stream of, like, you can just create an entire video of Adam saying whatever. And so we sort of supported both sides of that.

Adam: Yeah. Yeah. Well, I think that, you know, obviously, you're both talking about, like, different ways that's being used or or one of the interesting questions I like to ask in these situations is where you've maybe been the most surprised or, like and the easy one is, like, where have you seen it move the needle the most?

Like, transcription has come up a couple times, right, where it's, like, It's just better now. It's cheaper. Like, you don't have to involve people anymore. Where are our places where you maybe you you think that are the you know, highest usage moments, right, where you'd see it really actually move the needle and and Chris and this on. I think that you both are gonna have different kind of perspectives on this where it's you were, you know, Chris had the product before Hassan had been building with AI from the beginning. And then so biggest usage and then most surprising I guess, would would be the the the second part of the question.

Hassan: Yeah. Sure. I mean, from our perspective, I think that whenever we think about, like, in our world, replicas and avatars, was both surprising, but maybe should have been not surprising, but where we've seen a lot of usage is in the concept of using your likeness for skills that you have. And specifically, we've seen it in language. Right?

So we have, like, a healthcare company that has physicians sending out instructions ahead of, like, you know, surgeries in the patient's native language. Right? And they've seen across the board that a lot of these, like, translation use cases where, like, it is still you, but now it's you speaking in some native language. Like, I think translation use cases have been taking off and we've seen a lot of demand for that, and that was certainly a bit surprising, but it totally makes sense.

So I think, like, these leveraging your likeness beyond what you would have been able to do yourself.

Adam: And and where it's being used, so that's the most surprising one. In terms of where you see it used the most, is it still like kind of classic personalized sales outreach? Because that's where my mind goes, right, where it's, like, you're gonna click it more. You're more likely to click if it's like, "Hey Adam, you look great today. You should buy this thing." Right? Or?

Hassan: Yeah. So definitely from a volume perspective and from a frontier perspective, I think sales and marketing is definitely where we see the most volume, most usage, because that's also where the barrier for someone recording is usually higher. Like, there's just too many people. There's a scale problem, and there is also a problem of, like, people don't typically want to record.

So definitely sales and marketing use cases - whether outbound or even like middle of the funnel - is where we see a lot of usage today.

Adam: Got it. Yeah. Chris, same to you, but in terms of, you know, most usage, and I guess with the, you know, I guess with an AI tilt on the answer, and then most surprising.

Chris: Yeah. I mean, I think I think - I think anything, the most usage is the place where you have to do the least amount of work to get the benefit. So, that's shown up for us. Like, pretty quickly we saw, like, actual significant productivity advantages with our engineering team using Copilot with GitHub.

And that was funny. I remember having a meeting with everybody. We're talking about AI - who's using AI? Well. And someone's like, "Actually, I've been doing this for 6 months. I think I'm 30% more productive." We're like, "Oh, okay. Alright."

Adam: Without himself. Like, I just thought you were a superhero, man.

Chris: I mean, it turns out it was someone who was basically a superhero already. So it was, like, not surprising. But I think - I think a lot about that in terms of there's, like, so many incredible AI tools already in so many different pockets, but you have to know that they exist. And you have to believe and test and not - you know, there's a huge problem we have with everyone overpromising and underdelivering.

And I think what that's trained most folks to do who aren't, like, extreme early adopters is, like, they've stopped trying as much stuff. Right? Because it's just like, "Is this really gonna be the thing that writes the perfect email? Is this really the one? Like, I tried 4 already. Come on."

And so I think - what my gut tells me is that we're gonna see it in a lot of places where it just gets turned on. You start to receive the benefits, it changes how you think about a product or it changes behavior, and then we add more layers of that on top.

And so transcription is what we already talked about is a good example. You turn it on. You have the transcriptions. Now you can edit them. But then you have the metadata piece, which is like, should we tag your videos for you so you know what they are? Should we organize the analytics for you differently based on what they are? Can we start to give you insights into what the video is because we know the part, like, can see within the transcription and and kind of contextually understand what things are.

All of that's brand new. All of that's really exciting. A lot of it is stuff that you could have an expert do for you before, and most people didn't. And so this is like a big democratization of a lot of these tools in tech. And so I think that's really exciting.

And I think that's what we're gonna continue to see is the places where you can just turn this stuff on by default and you get the benefits. It's gonna work. By the way, the part of AI that isn't new, which is, like, recommendation engines and all this understanding. This was in all the biggest tech companies. Right? It just wasn't democratized. Like, that's how photos work on your phone with the Apple. That's how Google photos works. That's how Gmail works. That's how the recommendation engines work.

It's just now we all can use it. I feel like that's what's happening. And then I'll - I'll keep it on video, but I think some of the most surprising stuff for me has been the quality of the generative b-roll. So, like, OpenAI, Sora, and a few of those things, like, it's pretty - It makes sense. Once you see it, it makes sense.

But I remember seeing Midjourney being like, "Man, it's gonna be a long time before video is really there." And then it was, like, 6 months later, okay, I wouldn't know if this is real or not. And so I think that's - that's probably the most surprising.

Adam: Just the pace, yeah, just the pace of it. I think you mentioned something that's a good segue in terms of, like, hype versus real, which is like the - and we can keep it kind of within the video space. When you, like, you know, think about that statement of, like, you know, and and the pain that we've all gone through here, it became like a meme of, like, "Yeah. Cool demo. But, like, what -"

Chris: Yeah.

Adam: "You know, when is this, you know, live or, you know, Google keeps doing the thing where they, like, you know, the demo sweet, but it's not really available yet." And then they're just this combination of, like, an incredible piece that in actual, like, we all are also, like, you know, I think getting spoiled in a certain degree where, like, we're all, now that it's happened so fast, we're, like, expecting it to keep happening so fast.

Within, like, the video space, what do you think is, like, further out than, you know, as kind of more hypey and further out than maybe you'd think? And and what do you think is maybe closer than than you'd think. Right? Like, so you mentioned the b-roll thing, Chris, where, like, you were surprised, like, "This is - that's real. Like, that's here."

And you didn't expect it to be, but maybe some different examples of, like, more in the business context. The sort of thing or, like, "I don't really know what to do with that" necessarily besides, like, the cool thing. Like, yeah. Yeah.

Chris: I think there's a lot. I think a lot of it is that. A lot of it's stuff that's really fun to play with, but, like, the truth is, you know, I mean, Sora - we'll stay on Sora. It's like, you can use Sora. You can make something really cool, or, actually, it's not open yet. I don't think, but you'll be able to make something really cool. And then what do you do? Where do you put it? Yeah. You know, which video is it in?

You know, and then it's like, you actually go back to another level of problem. Which is just like, how do I know when I should make a video? Should I be on camera? You know, Hassan's over here talking about, like, you can actually type it in and it's gonna do it for you. Right?

And I think these are all different ways to solve this problem, which is this foundational problem of people being afraid to be on camera, being, you know, fear of public speaking is real. And I think that's like there's this human cultural thing that stops it, right, basically.

So I think - I think we are for the - someone who wants to do the work, you can up the production quality of the videos that you're making with AI dramatically today. And you have to do the - when I say "do the work", there's, like, tools from Adobe that are gonna improve what a speaker sounds like. There's music tools. We have some in Wistia, but you can go use Sonic and create your own custom music. You can go find the b-roll. You can - you can use the AI avatar. You can clone your voice. You can make a great thing today, but it's still a ton of work to get in. Right? Yeah.

And I think one of the things that's hard, which we've seen, is the same thing as true with generative text, is the demo's incredible but there still needs to be a human deeply involved, like, understanding, like, the context. And you'd think some of this stuff would be easier, and it's still not.

And so I think that's the interesting thing in this moment. And by the way, I also think if we think AI is kind of like the printing press, like, if we think it's gonna change everything. And I do think it is like that. I think that the changes we're seeing are enormous.

What should you do? I think you gotta - if you were the person who made books before the printing press, you should go and figure out how to, like, use that damn printing press. You know what I mean? Like, you should go in and be like, you know, story structure, you know, when something should be a chapter, you know, things people care about. Like, be the first person in there, understand it, use it, figure it out, actually - back to our time - before you have to overcome this challenge of you have to keep trying things even though everyone's overpromising. And - Yeah. A lot - or not all - but most are overpromising, under-delivering. Yeah. So it's actually tiring. It's actually hard work, but I think that it will pay off because if you find the right things, they can really, you know, you can - you can be one of the first people in the new world who understands how these things are gonna fit together.

Adam: Yeah. Yeah. Yeah.

Hassan: I, I agree. To love that, I do think there's - I think there's a lot of hype around, like, solving the entire problem. And the reality is is that where AI will really, like, allow these new models, will really help is an existing workload, making them more efficient. Right? And I think that's the magic moment.

I think the magic moment isn't screaming "we added AI", it's actually someone experiencing something and then not even realizing. They're like, "Oh, wait. It just did that." And, like, that's the magic moment. I feel like should go to, which is, like, underpromising, and it just does it. Right? So it's like, yeah, like, like Chris said, like, before maybe a manual effort, now you had a human in the loop that would do it, now it's, like, way faster and just did it. And it's, like, it's, like, magic. Right?

I think that's true - it's true for knowledge, it's also true for video. I think there's a lot of companies like, "Oh, yeah. We're gonna - you're never gonna have to do any work for video anymore." It's like, no. I think you'll still have to do a lot of work for video to get a good video out.

But you'll be able to get a good video out faster, better with the right tools. And I think that's also true for code generation. I think there's been this, like, big, like, "You're not gonna need software engineers anymore. And you're just gonna be able to speak natural language." It's like, well, Chris and I already do that.

We just - I'll just be like, "Hey, engineers, can we make this thing?" And and, you know, they'll - and, you know, there's already - I'm already using natural language to ask for us to build something. And they have a lot of context, and they're super smart, they're smarter than most, like, they're - they're more intelligent.

Chris: I mean, I'm on the - you know, I'm on the side of, like, we're still much smarter.

Hassan: You're smarter. I think that humans are still smarter than - Yeah. Like, even the most, like, in these large language models. And so it's like, already telling them, like, there's probably like a 20% loss of, like, what you had in your head versus that.

And so instead of this, like, future where it's, like, there's no more software engineers anymore. And, like, you can just use natural language to create software. It's like, okay, actually, there will be software engineers, and software engineers will be way more productive because they'll have all this really great tooling that lets them focus on these, like, hard complicated fun problems.

And so, like, I think that's the same thing with video editing. Like, you're not gonna replace people creating great videos. They're just gonna empower them to create better videos.

Chris: Well, I was trying to set up a budget going to this other thing, but Hassan brought it up - I think it's worth saying, which is like on the code generation side is really interesting. I agree with you too. Like, the way I look at it is, like, if you have a 10x engineer and they understand how to use AI coding agents, they become the 100x or the 1000x. Right?

So you're gonna have a huge bifurcation, actually, probably what's gonna happen is you're gonna have way less people who are less skilled in actually being engineers day-to-day, they're gonna choose other professions, I would imagine. They'll still maybe be able to make things. But, like, you're also gonna have people probably who become the 10x, 100x, 1000x person. And, because a lot of that's, like, how you understanding how you build problems, and it's gonna shift the bottlenecks.

Right? So, like, for most people, the bottlenecks have been engineering. It's like what they believe - it's like, "Okay. If we just have more engineers, we're gonna get to this place faster."

But I think it's gonna switch a lot of pressure actually to product management and to other areas. And if you have someone who's, like, world-class at it, their impact's gonna be magnified really dramatically. So I think it's like, again, super interesting because it's changing so fast.

And also, you know, we could - I mean, I'm sure you've both seen this, but there's like, the predictions that we're gonna run out of energy, basically, because whatever it is - like, 19% of the US's energy is going to data centers this year. It's something massive like that. Yeah. And then it's like, well, if you need to go 2 more orders of magnitude, you're basically screwed.

And so we're just gonna run into, like, it's possible we run into real world implications to slow us down. And so what does that mean? Is - I've been asking myself that question. What does that mean if, like, we're gonna continue to advance at a fast clip, but it's not actually gonna be sustainable? And I think it's really gonna - you have to wrap your mind around.

Hassan: Yeah. I think we're gonna reach compute and energy limitations. And so I think - I think there'll be, like, even for models like Sora, like, they're incredible. One piece of it is, like, "Okay, what do you actually use it for?" The other piece is, like, democratizing it and making it cheap enough for someone to actually use - if we're actually further away than we might think.

Adam: Yeah. We've had - I've had some interesting conversations about that recently with, like, there's this phrase that got thrown around - "AI compute", right, where there's, like, this race, underlying race, like, outside of all the demos and, like, the stuff people get talking about, there's this, like, massive race, like, "We need to have more compute than anyone else in order to have the best model," generally.

And so there's, like, that race happening. And then there's also the whole chip thing, right, of, like, chips that are really fast, but super energy-intensive compared to maybe they're a little slower, they're a little - they're not as fast, but they're way cheaper, like, on the energy efficiency side.

And the term that got thrown at me - someone who was way smarter than me on this stuff was talking about it of, like, "What's the, like, performance per watt per dollar?" was, like, the breakdown of, like, how to get, like, where it actually - when you really kind of get down to it, like, you have to think about efficiency and these sorts of things.

But to pull it back, actually, to video stuff and more generally about what you were both talking about. And this idea of, like, the incumbent gets brought up a lot in AI, right, of, like, it accelerating, you know, the big players who are already doing this stuff. And now they're just way better at it. Or, which I think ties into what you were both saying about, like, an engineer who's already an engineer is like the incumbent here. Right?

Like, they're already being engineers. They're already good engineers. Now they're just better engineers, right, rather than it being someone who, like, wasn't an engineer before now competing with them.

Within the context of your two businesses, where do you think about - like, so Hassan, you're obviously more startup-y, like, kind of on the knocking at the door kind of a thing. What - how do you think about, like, the word "incumbent" as someone building this platform? Like, who do you think of? Maybe above you or, like, further along that you need to do better than or catch up to or whatever.

And then, Chris, I think, you know, as a non-VC-backed company, I think you maybe have more leeway, right, to, like, not - not necessarily that that's your ambition, but maybe what you wanna do with it. But, like, when you think about, you know, maybe people that are bigger than you in competing there, how do you think about, I guess, and we'll try to keep it in video as much as possible, this incumbent setup of these AI tools that everyone's using and everyone's kind of benefiting from?

Chris: I mean, I can go on this one. I think - I think that the whole question is, like, how much does AI change what the foundational product can be - what, like, the customer problem you can solve. And I think the more that AI changes the customer problem that you can solve, the more the incumbent is at risk.

And if it doesn't change it, it enhances it in some way, then I think incumbents have maybe arguably an advantage, but they have a disadvantage which is, like, they probably have a good business. And so they may not move as quickly. Right? We know this. Like, organizations get bigger and they often move more slowly. And so you have to actively fight against that.

And so I think - that's the way I look at it. Like, there are a few players that are bigger than us. We look at this as like, "Okay, can we - are there problems? I mean, we don't know." Right? Like, that's a lot of what we're talking about. We don't know how well we can deliver these some of these solutions.

Are there problems that we can solve dramatically better that literally change what level of the customer problem that we can solve? And if we can and we're doing that and an incumbent that's larger than us is not, we're gonna be able to take over areas that they're not getting into or that they can't move fast enough on.

If they move crazy fast and they have a ton of distribution, they're gonna be able to dictate a lot more of what happens in the market. But a lot of it has to do with how much it shifts. And I think I'm gonna speak for Hassan, but, like, the stuff that you're doing is really interesting because it's like, "Okay. You wanna help someone make a video by typing instead of recording themselves." That is very different. And there's gonna be a lot of different products and services that exist to help people with that. And so if you can actually get a thing that's high enough quality there, that's going to disrupt - make massive incumbents - they're not - they're gonna have to remake their whole economics, everything that they do.

Adam: Yeah. They just weren't doing it at all before. Like, it just - it's not even a problem that they were even thinking about -

Chris: Yeah. For it's like, if you have to disrupt your business to take advantage of the AI, you're in trouble. But if you don't have to, then maybe - Yeah.

Adam: Yeah. Enhancements versus, like, brand new takes on it. Yeah. That's interesting. Yeah.

Hassan: I agree. I think Chris said it well. Right? I think that even for us, I mean, we're building video models to help some what you might consider because some incumbents in the video space is actually use our models to, to take advantage of AI.

But I actually see it play out a few different ways. I mean, you know, we've seen some that are using our models to actually increase usage. So they're actually saying, "Okay, this is actually lowering the bar for people using our platform, and allows us to really allow people to create way more videos at scale." And now, you know, we're removing that barrier that people feel whenever they're trying to record.

And then, I think there's others that, you know, like Chris said, might have so much revenue or strategy dedicated to the older way to do things that they struggle a little bit more because, if it disrupts their business, then it's a harder decision to make.

So I think from our perspective, I mean, we're a little bit different since we are an AI research company. Our job is to essentially build models that disrupt the way that you do things today. Like, we're building models that essentially are trying to remove the limitation of time, scale, and skills. So it's a little bit different for us.

Adam: Yeah. I think it brings up an interesting - for me and just in terms of your two businesses also, and I'll be interested to hear how you're each approaching it, in terms of "buy versus build."

And I think within video specifically, right, where Chris, as you - as I misquoted and you correct me on, you already have millions of customers. You had millions of customers before AI stuff came around, and making decisions about buy versus build on the features that you already have.

And I think that AI has been, like, this really kind of - I know, we're all built on top of or most people are built on top of foundational models that, you know, billions of dollars went into training in one way or the other.

Hassan, that might be more proprietary. I'll be interested to hear your take on it. But so, Chris, you're kind of clearly in a buy versus build position of making those decisions internally.

And Hassan, by virtue of being a developer-first product, right, you're counting on people like Chris saying, "Buy instead of build," or "Buy and build less." Right?

So I'd be interested to hear, again, Chris, maybe any decisions you made around that. And Hassan, how your conversations have been going.

Chris: Yeah. I mean, we looked at this and we're like, "There's a huge amount of investment going into the space. There's a lot of people competing for many of the similar prizes." And so from our perspective, we get to play with them, and we can buy. And we could switch between them often in many cases.

I mean, it reminds me of, like, the CDN market when we got started with Wistia, which is like, you know, content delivery networks for, like, you know, that this is like the the if you could have a sustained advantage as an infrastructure provider and some of those products, you can have unbelievable growth, right, because you capture the biggest customers. They continue to expand. They live with you. You get more big customers. Like, that's an amazing place to be, but it's a hard place to be because you have to have some kind of sustained advantage.

And if the product ends up being the same, then it's a lot harder. And it ends up, you know, you end up fighting on cost versus fighting on innovation. Like, it's all about the cheapest thing wins because it's gonna be in people's cost of goods.

So looking at the market through that lens, we look and say, "Okay. There's a bunch of different infrastructure providers. We should test them, figure out which ones are really good. If there are ones that are, like, highly unique, we'll use them." But we made a pretty explicit decision - we're not going to build like, from the ground up a bunch of AI models. We're gonna live at the application layer and we're gonna integrate them in and try to find those moments where they really deliver a lot of value.

And, yeah, so, I mean, that's why, I mean, it's just - that's how I think about it. I think that's the way I looked at the space. I think it's also - we have the analogous space, which is LLMs in general, and I think that's shocking how much prices have come down. And how good the competition has gotten there. Right? Like, it's unbelievable. Just how good Claude is compared to ChatGPT compared to, like, the meta AI models that are open source. So that's - it's crazy how good they all are and how cheap they are.

Is great for the world, I think.

Adam: Yeah, I think -

Chris: Many more options.

Adam: Tough place to be on the infrastructure. Yeah. Yeah. And I think - and just to double click there for a second, because I think it's worth going a little deeper on the video side, because this has been a story in video for a while, it feels like where, like you mentioned, CDN, and all of these video apps in general, have always had, like, whatever, infrastructure underneath the service.

And unless you're building video apps, like you've never heard of any of these companies. Like, I know when we first got into streaming, like, doing streaming stuff and on the Stonks side, like, I had never heard of any of these companies. Like, And because if you're not in video, you wouldn't know that some of these big apps are built on top of it.

Is it just brutal? Like, Chris, you've been building video for a long time. How much do you guys move around on products that are, like, when you think about your stack, from an infrastructure perspective?

Chris: We do. I mean, if someone comes out with a better thing that is faster, better, faster, cheaper, we switch. And I feel bad sometimes because it's like, you know, it's like, we had a great relationship. And often, the best companies keep innovating. And so they find a way of driving the price down and they keep innovating. And then there isn't a reason to leave.

But it's - it's - you need, like, a culture that's set up for that. You need a business model that's set up for that, you know, this type of thing. Like, you get to more scale and you go back to your customers and you drop their price. Like, you have to actively proactively do that, I think, in the infrastructure world.

If you're in something that's hyper-competitive and your goal is to be the biggest with the most scale, and your price is so low that other people can't actually get to you, then - but that takes a very - if you're unsure of what your strategy is, your business model is tough.

So for us - we learned to actually kind of early because the first versions of Wistia were more basic video hosting that that could happen to us - that people would come to us and say, "Well, what's the price per gig? And what's the price for the for the stuff?"

And we happen to figure out, "Oh, wait. Like, we can build features in our player. We can build..." For example, this is such a - this sounds like the simplest thing in the world, but when we did it, it didn't exist, was just "You can capture an email to be able to watch a video." So on the player itself, it says "Put an email to click play."

I mean, I remember we made it up and we did it. No one else was doing it. Marketers loved it. We then integrated it to all the major email service providers, eventually marketing automation providers. And people could see - "I can get leads from a video." And so it was very concrete, and they didn't think about the gigs that were delivered. They didn't think about the number of videos. They thought about the number of leads.

And so over time, as we've added more functionality, a lot of it is like we're helping the marketer solve a marketing problem that they happen to use video for. So it's very different than, like, "We are just an infrastructure company."

Adam: Yeah. Moving out, making a decision to move out of the infrastructure layer.

Chris: Yeah. And it's funny. Even in our case, we have, like, some infrastructure customers. But the way that it's worked with them is, like, because they're on our APIs, we're able to give them video hosting extremely cheaply because of our scale. And it's not - we don't look at that as the core business. That's not the main thing we're talking about all the time, but we can give them an incredible service. And for us, it's really good because it pushes us to have more scale. Does that make sense?

Adam: Yeah.

Chris: And then that allows us to invest more into these differentiated, like, application level things, but we were big enough, I think, when that was happening, that we're able to do it.

Adam: Yeah. Yeah. So Hassan, would you describe yourselves as an infrastructure tool?

Hassan: Yeah, I would. I mean, we are an infrastructure platform. We specialize in building and training these models and deploying them for developers to leverage and build applications on top of. So in that sense, yes, we are an infrastructure platform, an AI infrastructure platform that happens to specialize in video currently.

But the idea is to allow developers to take our models, take our APIs, and build really cool applications leveraging generative AI. So from that perspective, we are an infrastructure tool. We provide that foundational model for people to build on top of.

Hassan: Yeah. I mean, I think I think we're - I think we're more so a platform provider. I think, you know, for us, like, the way I see it is is that we're taking a couple of different bets. One is that really good, high quality state-of-the-art replica or avatar models are hard to build and expensive to build, and not every company needs to build them. It's currently a sort of competitive moat for certain companies to have, like, the best in class model.

And we think that actually, it's our job to help commoditize access to the state-of-the-art models, and that there is a lot of research and science that is still to be done on creating really, really high quality models for digital twins or replicas. Right?

And so what we see it is like, you know, we're we're actually the one that's trying to commoditize access to these models and the models still have a lot, a lot of room to grow. You know, right now, most models that you see control the lower half of your face. There's an opportunity to control your entire face. There's an opportunity to do head pose control. There's an opportunity to do all sorts of stuff, right?

And that will continue to be true. And I think for platform providers, model providers, whether it's the large language models of the world, whether it's the video models of the world, I think there has to be a strategy to continue to adapt and build the next game-changing thing.

And I think on the video side, there's a massive opportunity. And there's a lot of really awesome opportunity to create, like, magical moments with some of these models.

So, I mean, the way we see it is we're taking the bet that not everyone needs to build these models. And right now, there's probably too many companies that are trying to build a foundational model. They're trying to build it and, like, actually getting state-of-the-art models is something that takes more time, more energy, you know, specific expertise.

And we're betting that actually not everyone needs to build the models. Like, they can just come to us. We'll give you the models. And yeah, and over time, I think that, given our scale, like, we'll probably be way more cost efficient at it, not just from volume perspective, but just from efficiency gains in how we built the models.

So, yeah, I think we are a platform provider. We power a number of customers all on the vertical spectrum, to create with these really awesome replicas of people, text-to-video, real-time stuff. But there's still a massive road map, along the way to create entirely new experiences within video.

Adam: Yeah. You use this word "bet" that I made a note of before that I thought was interesting - the question of good bets versus bad bets and your views both internally. Like, obviously, there's bets that you're making. I mentioned earlier, like, the bet, the "company moment" type of thing.

Chris said that you didn't think it was all the way there yet or hadn't really - obviously, the pressure to build it, but not like, "We're gonna die." What are some bets that you're seeing internally and externally that you see and you're like, you think are good bets kind of within the AI movement?

And then a little bit spicier - where are you seeing bad bets? But, like, when, if ever, have you seen people - have you avoided something that you think either isn't playing out already or you just don't see playing out at all in the future?

That's not - you mentioned - you didn't word it this way, but everyone building their own model, like, you're basically making the bet that, you know, you're gonna come to a centralized place, and you're saying someone who's doing it themselves is probably a bad bet because you're probably not gonna do it as well as the person that just does that themselves. Like, people trying to verticalize.

Hassan: Yeah.

Adam: So yeah. Go deeper on that kind of, you know, this idea of good bets versus bad bets in AI with as much of a tilt towards video as you can give.

Hassan: Yeah. So I think generally, what I've seen as like a bad bet is trying to build businesses around the limitations of models today because inadvertently, you know, we've seen this with the large language models, especially, but it's also true in video - it's like, "Oh, building businesses around just like, hey, like, this thing's not doing X today. Let me just build around that." And then naturally the model starts doing that. Right?

So, like, sort of betting against the model development is what I would say - not a great bet. But what I think - yeah, I think what is what I believe is wrong, but it's like, "Hey, it doesn't make sense for every single company to have a foundational model" because I think the companies that are focusing on building the model that are exclusively, like, entire research teams, like, "This is what we're doing. You go much deeper. You can go much deeper."

So, you know, even for text-to-video, you can go really deep in really getting fidelity to be amazing, improving expression control, voice. And those are things that you can do if you're focused and you go deep, rather than like, "Hey, we just want a, we just want an exclusive model."

And so, yeah, I think the bet that every single company needs their own foundational model, everyone needs their own video model, everyone needs, like, a Stable Diffusion model - I don't think that actually makes sense because it's the same way that, you know, not everyone needs to host their own servers. Like, AWS and GCP and Azure - those are perfectly great. They've made that bet. Like, those are good bets. You can see it.

Like, not everyone needs to host their own solutions because the AWS of the world do it better, faster, cheaper, at a much larger scale. And I think that there are parallels in AI where the model providers and model builders will do it better, faster, cheaper, way more advanced. They'll be able to move much faster on it. So, yeah, that that's my take on it.

Adam: Yeah. Yeah. Chris, good bets, bad bets?

Chris: Well, first of all, I agree with what Hassan said around betting against the models getting better. I think that's a pretty big mistake. I think good bets are ones that there's asymmetric upside.

So, I just said the most obvious thing ever, but I do think it's true as, like, a lot of the - what AI is gonna let you do, I see and think about today is, like, you're gonna - you can do things dramatically faster. We could do things that you could not do without an expert, otherwise.

So I think the way you should look at it is like, "Well, how can I build things and test them dramatically faster? How can I try these things that, like, you know, maybe to scale it would be really hard, but to do the one-off thing could prove that I can create something that's really gonna be different and valuable for customers?"

And I think it's all about looking for those places where you can use AI as a tool to make basically better, faster bets. And if you do that, you're not gonna be - you think you're really costing in the better time.

And so it actually can be pretty asymmetric in terms of it being an upside. The other thing I would say to think a lot about is, "Think a lot about the sensitivity of your business model to these types of bets." And so what I mean by that is I think it - really understanding if you're using AI in any sort of - like, what are the other options in the market? Are there other things on par with this one? Just like you're making an infrastructure decision.

So you wanna actually take a pretty long-term view and imagine if things do work, what happens? So if things do work, and, you know, and I've picked this partner and assuming you're not generating your own models, and it scales. What kind of position am I gonna be in? What does that mean for the price of what I charge to my customer?

And, like, I would just really encourage you to pay attention to that really early and try to understand the sensitivities because in some cases, you know, if you're - it sounds like what you guys are trying to do at Tavus is, like, be this infrastructure thing and drop the prices and get more scale and do that whole thing.

Like, that is that is a good bet for someone who wants to use you, right? Is like, "I want to believe that as I get to scale, you can bring the price down. And as you get to scale, you can bring the price down that together, we end up with this like symbiotic relationship." And that means I could - I can think about how aggressive to be with my customer today on price or how some free something should be or otherwise.

And so I would just pay a lot of attention to that because I think in infrastructure decisions in general, that's where I've seen things go awry - is not really thinking enough about which parts of your business model are being, like, most sensitive to these infrastructure partners?

Adam: Which - which I feel like ties back to some of the big, you know, the really flashy demo, and maybe you rush into an infrastructure decision or whatever, and it's a startup, and it doesn't work out or whatever the company that you're building on top of doesn't make it or doesn't deliver in the way they promise. Like, those are - I'm trying to think if we've had that happen and maybe either of you have had that happen, but that's an ugly situation, right? Where if you're building on top of somebody that - Yeah.

Chris: You do not - you don't want that. Yeah. Yeah. Which I do think is going to - it puts more pressure on all these solutions because, like, people are thinking about it like that. And so it's this weird conundrum.

So it's like, your potential as an infrastructure company, your potential largest customers could have the most scrutiny. But if you get them and you get enough of them, it decreases the risk of the business. Yeah. So it's - it's very challenging.

Adam: Yeah.

Chris: Like, things - it's..

Adam: Well, you want the logo. I mean, I know from the startup side of things, I'm sure Hassan and Tavus, I'm sure you guys think about this. I'm sure Wistia has too, where it's like, you know, if some huge customer already uses you and you're talking to another huge customer, like, "Look, they trust us. Makes it -"

Hassan: Yeah. Exactly. Yeah.

Adam: The logos on the website are there for a reason. Well, cool. I mean, we're getting towards the end of this. I wanted to talk - I mean, we've been talking about the future a lot and how to approach it and how you guys are thinking about it.

But I'd love to get some takes on within the video space - you fast forward, you know, 5 years - where you see things heading, like, and where you think things will look different, things that will look the same.

I think Hassan, you know, obviously, you're pushing and obviously believe in this future where maybe more replicas. Like, there's maybe more video, but maybe not, I don't know, maybe it's live, maybe it's not. So you can expand on that.

And then, Chris, I think - yeah. You've obviously been in the game here for a second. So you've seen some progressions and maybe some things that did or didn't work out. So I'd love to hear your take on where you think we're going from here.

Hassan: Yeah. I think so, you know, we started out with this thesis that, like, how do we solve the attention economy problem, which is that there's a ton of information and content at our fingertips that we're actively trying to make a decision of, like, what is actually relevant to us, right? And personalization is key there.

I mean, there's a reason we're addicted to TikTok. It's because we scroll on TikTok and we're like, "That seems relevant to me." But we actually see, I think over the course of the next, like, 5 or 10 years, it'll be very clear that recommendation algorithms were cool, but the future is actually that every piece of content that you consume is personalized just for you. Right?

And we're already starting to see that to some degree. You know, you go on YouTube and it maybe gives you segments that were more relevant to you. I think there will be a future where every single video you potentially watch is really personalized just for you. Whether that's in real-time or anything.

So I think that's something that, over the course of the next 5, 10 years, we'll definitely see is a higher degree of personalization in the content that you actually consume. And, yeah, so that would be our bet for the future.

Hassan: Yeah. I mean, I think I think we're - I think we're more so a platform provider. I think, you know, for us, like, the way I see it is is that we're taking a couple of different bets. One is that really good, high quality state-of-the-art replica or avatar models are hard to build and expensive to build, and not every company needs to build them. It's currently a sort of competitive moat for certain companies to have, like, the best in class model.

And we think that actually, it's our job to help commoditize access to the state-of-the-art models, and that there is a lot of research and science that is still to be done on creating really, really high quality models for digital twins or replicas. Right?

And so what we see it is like, you know, we're we're actually the one that's trying to commoditize access to these models and the models still have a lot, a lot of room to grow. You know, right now, most models that you see control the lower half of your face. There's an opportunity to control your entire face. There's an opportunity to do head pose control. There's an opportunity to do all sorts of stuff, right?

And that will continue to be true. And I think for platform providers, model providers, whether it's the large language models of the world, whether it's the video models of the world, I think there has to be a strategy to continue to adapt and build the next game-changing thing.

And I think on the video side, there's a massive opportunity. And there's a lot of really awesome opportunity to create, like, magical moments with some of these models.

So, I mean, the way we see it is we're taking the bet that not everyone needs to build these models. And right now, there's probably too many companies that are trying to build a foundational model. They're trying to build it and, like, actually getting state-of-the-art models is something that takes more time, more energy, you know, specific expertise.

And we're betting that actually not everyone needs to build the models. Like, they can just come to us. We'll give you the models. And yeah, and over time, I think that, given our scale, like, we'll probably be way more cost efficient at it, not just from volume perspective, but just from efficiency gains in how we built the models.

So, yeah, I think we are a platform provider. We power a number of customers all on the vertical spectrum, to create with these really awesome replicas of people, text-to-video, real-time stuff. But there's still a massive road map, along the way to create entirely new experiences within video.

Adam: Yeah. You use this word "bet" that I made a note of before that I thought was interesting - the question of good bets versus bad bets and your views both internally. Like, obviously, there's bets that you're making. I mentioned earlier, like, the bet, the "company moment" type of thing.

Chris said that you didn't think it was all the way there yet or hadn't really - obviously, the pressure to build it, but not like, "We're gonna die." What are some bets that you're seeing internally and externally that you see and you're like, you think are good bets kind of within the AI movement?

And then a little bit spicier - where are you seeing bad bets? But, like, when, if ever, have you seen people - have you avoided something that you think either isn't playing out already or you just don't see playing out at all in the future?

That's not - you mentioned - you didn't word it this way, but everyone building their own model, like, you're basically making the bet that, you know, you're gonna come to a centralized place, and you're saying someone who's doing it themselves is probably a bad bet because you're probably not gonna do it as well as the person that just does that themselves. Like, people trying to verticalize.

Hassan: Yeah.

Adam: So yeah. Go deeper on that kind of, you know, this idea of good bets versus bad bets in AI with as much of a tilt towards video as you can give.

Hassan: Yeah. So I think generally, what I've seen as like a bad bet is trying to build businesses around the limitations of models today because inadvertently, you know, we've seen this with the large language models, especially, but it's also true in video - it's like, "Oh, building businesses around just like, hey, like, this thing's not doing X today. Let me just build around that." And then naturally the model starts doing that. Right?

So, like, sort of betting against the model development is what I would say - not a great bet. But what I think - yeah, I think what is what I believe is wrong, but it's like, "Hey, it doesn't make sense for every single company to have a foundational model" because I think the companies that are focusing on building the model that are exclusively, like, entire research teams, like, "This is what we're doing. You go much deeper. You can go much deeper."

So, you know, even for text-to-video, you can go really deep in really getting fidelity to be amazing, improving expression control, voice. And those are things that you can do if you're focused and you go deep, rather than like, "Hey, we just want a, we just want an exclusive model."

And so, yeah, I think the bet that every single company needs their own foundational model, everyone needs their own video model, everyone needs, like, a Stable Diffusion model - I don't think that actually makes sense because it's the same way that, you know, not everyone needs to host their own servers. Like, AWS and GCP and Azure - those are perfectly great. They've made that bet. Like, those are good bets. You can see it.

Like, not everyone needs to host their own solutions because the AWS of the world do it better, faster, cheaper, at a much larger scale. And I think that there are parallels in AI where the model providers and model builders will do it better, faster, cheaper, way more advanced. They'll be able to move much faster on it. So, yeah, that that's my take on it.

Adam: Yeah. Yeah. Chris, good bets, bad bets?

Chris: Well, first of all, I agree with what Hassan said around betting against the models getting better. I think that's a pretty big mistake. I think good bets are ones that there's asymmetric upside.

So, I just said the most obvious thing ever, but I do think it's true as, like, a lot of the - what AI is gonna let you do, I see and think about today is, like, you're gonna - you can do things dramatically faster. We could do things that you could not do without an expert, otherwise.

So I think the way you should look at it is like, "Well, how can I build things and test them dramatically faster? How can I try these things that, like, you know, maybe to scale it would be really hard, but to do the one-off thing could prove that I can create something that's really gonna be different and valuable for customers?"

And I think it's all about looking for those places where you can use AI as a tool to make basically better, faster bets. And if you do that, you're not gonna be - you think you're really costing in the better time.

And so it actually can be pretty asymmetric in terms of it being an upside. The other thing I would say to think a lot about is, "Think a lot about the sensitivity of your business model to these types of bets." And so what I mean by that is I think it - really understanding if you're using AI in any sort of - like, what are the other options in the market? Are there other things on par with this one? Just like you're making an infrastructure decision.

So you wanna actually take a pretty long-term view and imagine if things do work, what happens? So if things do work, and, you know, and I've picked this partner and assuming you're not generating your own models, and it scales. What kind of position am I gonna be in? What does that mean for the price of what I charge to my customer?

And, like, I would just really encourage you to pay attention to that really early and try to understand the sensitivities because in some cases, you know, if you're - it sounds like what you guys are trying to do at Tavus is, like, be this infrastructure thing and drop the prices and get more scale and do that whole thing.

Like, that is that is a good bet for someone who wants to use you, right? Is like, "I want to believe that as I get to scale, you can bring the price down. And as you get to scale, you can bring the price down that together, we end up with this like symbiotic relationship." And that means I could - I can think about how aggressive to be with my customer today on price or how some free something should be or otherwise.

And so I would just pay a lot of attention to that because I think in infrastructure decisions in general, that's where I've seen things go awry - is not really thinking enough about which parts of your business model are being, like, most sensitive to these infrastructure partners?

Adam: Which - which I feel like ties back to some of the big, you know, the really flashy demo, and maybe you rush into an infrastructure decision or whatever, and it's a startup, and it doesn't work out or whatever the company that you're building on top of doesn't make it or doesn't deliver in the way they promise. Like, those are - I'm trying to think if we've had that happen and maybe either of you have had that happen, but that's an ugly situation, right? Where if you're building on top of somebody that - Yeah.

Chris: You do not - you don't want that. Yeah. Yeah. Which I do think is going to - it puts more pressure on all these solutions because, like, people are thinking about it like that. And so it's this weird conundrum.

So it's like, your potential as an infrastructure company, your potential largest customers could have the most scrutiny. But if you get them and you get enough of them, it decreases the risk of the business. Yeah. So it's - it's very challenging.

Adam: Yeah.

Chris: Like, things - it's..

Adam: Well, you want the logo. I mean, I know from the startup side of things, I'm sure Hassan and Tavus, I'm sure you guys think about this. I'm sure Wistia has too, where it's like, you know, if some huge customer already uses you and you're talking to another huge customer, like, "Look, they trust us. Makes it -"

Hassan: Yeah. Exactly. Yeah.

Adam: The logos on the website are there for a reason. Well, cool. I mean, we're getting towards the end of this. I wanted to talk - I mean, we've been talking about the future a lot and how to approach it and how you guys are thinking about it.

But I'd love to get some takes on within the video space - you fast forward, you know, 5 years - where you see things heading, like, and where you think things will look different, things that will look the same.

I think Hassan, you know, obviously, you're pushing and obviously believe in this future where maybe more replicas. Like, there's maybe more video, but maybe not, I don't know, maybe it's live, maybe it's not. So you can expand on that.

And then, Chris, I think - yeah. You've obviously been in the game here for a second. So you've seen some progressions and maybe some things that did or didn't work out. So I'd love to hear your take on where you think we're going from here.

Hassan: Yeah. I think so, you know, we started out with this thesis that, like, how do we solve the attention economy problem, which is that there's a ton of information and content at our fingertips that we're actively trying to make a decision of, like, what is actually relevant to us, right? And personalization is key there.

I mean, there's a reason we're addicted to TikTok. It's because we scroll on TikTok and we're like, "That seems relevant to me." But we actually see, I think over the course of the next, like, 5 or 10 years, it'll be very clear that recommendation algorithms were cool, but the future is actually that every piece of content that you consume is personalized just for you. Right?

And we're already starting to see that to some degree. You know, you go on YouTube and it maybe gives you segments that were more relevant to you. I think there will be a future where every single video you potentially watch is really personalized just for you. Whether that's in real-time or anything.

So I think that's something that, over the course of the next 5, 10 years, we'll definitely see is a higher degree of personalization in the content that you actually consume. And, yeah, so that would be our bet for the future.

Chris: I think 5 years in the future, what we're gonna see is just more democratization of video. Average quality of video going up, I think trust is gonna go down. And I think that, when you start to understand the power of where AI is going, you have to enter into a war where you can no longer trust what you see. And we've dealt with this before. This is like when Photoshop came along and people started photoshopping every ad and every billboard.

I remember the conversation was like, just because someone looks like they have a six-pack doesn't mean that they do. You know, it was like stuff like that. And it's just like these unrealistic expectations and you train yourself to be advertising isn't real. Right?

I think the same thing is gonna happen here, and it's gonna be like, there's gonna be this big thing of "I don't know if this is real." And there's a lot of cases where it doesn't matter. Like, training videos doesn't matter. Like, the bot on the site helping you doesn't matter. Certain educational and their hyper-specific things that are like B2B training doesn't matter.

But, like, working with a company and making sure and you're gonna make a big investment, does the trust matter? Yeah. You're trying to meet someone in person, does the trust matter? Yeah.

So I think it's going to democratize. I think we're gonna see higher quality. I think we're gonna see lower trust. And I think we're gonna have to find trust in other ways. So I think as that happens, I think we're gonna see more of an emphasis on in-person events, live events, and content where which is extremely hard to fake, where you have a real sense that there's an actual real person there.

And, I think that, I mean, I actually have a lot of time. I should probably stall. I could just go.

Adam: No. It's good. It gives you another one. Keep it going.

Chris: Another thing I would say, and I this I can make it video specific, but I think it's more general than that. I'll make it B2B specific. How about that?

Adam: Nice. Perfect.

Chris: There are so many products and tools that can add so much value, and people don't use them. They hook it up. "Oh, this thing's gonna reduce my churn. This thing's gonna help in onboarding." And then there's so much human work involved in actually getting the value out of it. My bet is that a lot of that stuff, we're gonna get real leverage from the products that we use. And we're gonna have the products are gonna do way more than what they do today. And it's gonna be like you get a product and it came with an employee.

And I think that will actually potentially be, I think it's really exciting because a lot of that work, you know, but let's say it's onboarding employees or it's like booking calls to do customer success calls or it's like, whatever. A lot of that work is not that fun to do. There's fun really fun parts of those jobs that that's not the fun part. And, also, it's not where most of the value is created. The value is created in, like, creating the amazing culture, creating the amazing experience for the customer, you know, doing something that they would never get otherwise, pushing to build a feature that's never been built.

And I think we'll end up in a world where that is what we spend more of our time on. And I think that's, like, really exciting. And I think we'll get much better ROI out of the products that we use.

Adam: Yeah. Interesting. This is, like, the idea of, like, the "it just works" thing. That's, like, the Apple thing. Right?

Like, you know, that you just, you know, it just works. And, like, so many products, and I've had this experience a million times. Like, we're gonna use it in all these different ways, and then there's that lift. There's this big hunk to get over in terms of, like, actually getting the value out of it. So, like, the rise of, like, "it just works" for stuff that's much more complicated, right, whether it's like, yes.

Hassaan: It's a magic moment. Right? Like, you want it to feel like magic.

Adam: For sure. And then that idea too of, and there's a larger conversation to be had there, but, like, the video because video is so powerful today in building trust. I've been such a believer in that, like, for the last few years of, like, kind of being in the space.

Obviously, Chris, you've been on it for, you know, almost a decade now. Hassaan, like, it's a huge part of your business. But this idea of it of maybe approaching a peak of video trust, and then we're gonna have to,

Chris: If that trust goes down and having to, like, you know, deal with the world. I think video, I think if you believe that the person you're seeing or interacting with is in charge of what you're getting from them, I think you'll still build trust with video. So, like, Adam, if you're making stuff and I know you and I did try to with you, be it in person, be it a live event, do something like this where someone really believes it's, like, really you and they get to know you. What you're looking for is to understand that with that individual person, you have trust.

And then you, you're gonna have more freedom and flexibility. People know it's just like if I make stuff with Adobe Premiere today, there's basically been AI in it already, and I'm at making all these edits to what I'm saying. It's not exactly what was recorded. But you know me and you trust me and you trust the message as I'm saying.

Like, so what I guess what I'm saying is I think default trust is gonna get a drop. But I actually think that, like, trust with specific individuals might actually get higher.

Adam: Yeah.

Chris: Because we need that to navigate the world.

Adam: It bifurcates again the same way you would do with the engineers. Yeah.

Hassaan: Yeah. I mean, I agree. I think, and those are the use cases where it doesn't matter. Right? Like Chris mentioned this already, where, like, you don't really care.

You're like, whatever, I, it's if there's a two-way value that you're getting, it's like, "Oh, I got the content that I would want anyways, whether it's generated or not, I don't care." But then there will be, like, experiences where you're like, you want the other person. You want to be able to trust the other person. You want that level of, like, human connection.

And for those, I think that what's gonna happen is that regulation will actually be behind. We're always behind regulation. But at some point, I think there will need to be more thoughts around the ethics and implications of, like, generated content. I think that, like, watermarking, will be a cat and mouse game that will be really important to focus on, for consent will be an incredibly important thing. Disclosure will be an incredibly important thing.

And those are all individually hard problems to solve, especially without there being regulatory oversight that's actually good. And so especially, like, we're approaching an election. Right? Like, you will have to be a little bit more careful of, like, what is content that you can't trust and can't trust. Because the mechanisms aren't actually in place yet.

It's actually incredibly difficult to detect whether a video has been tampered with if you're using best in class models. And I think with that, it's really important for model builders and AI builders to be thoughtful about how they're building. Like, is trust implicit? Is transparency implicit? Is consent implicit? And something that we certainly think a lot about, but I think, like, there's a little bit of fast and loose game being played in AI right now.

A lot of, a lot of model, you know, development, and I think that there needs to be more thoughtfulness because otherwise, it causes general distrust of being higher in the street.