Scott Belsky: How Startups (And Incumbents) Can get Ahead of the AI Curve

0:04Welcome to Generative Now. This is a podcast where we talk to the builders who are creating the world's most exciting AI companies and products. We'll get their perspectives on how AI will impact the world we all live in today, right now, and in the future. I am your host, Michael Magnano. I'm a partner at Lightspeed, and we are a global venture capital firm backing bold founders with big ideas across the enterprise, consumer, fintech, and health sectors. And on this episode of Generative Now, we've got something special. Before we launched this podcast, we actually launched an in-person meetup, bringing together lots of people who were building AI products in cities like New York, Los Angeles, San Francisco, Paris, and a bunch more coming soon.

0:49And it was actually part of the inspiration for why we wanted to start this podcast. We wanted to be able to reach many, many more people and let them hear the stories of these builders, not just the people that were in these cities. And so a few months ago, we had an awesome event with Scott Belsky. Scott is Adobe's chief strategy officer and executive vice president of design and emerging products. Adobe is obviously the company that for many decades now has been leading the charge on empowering creators and the generation of media, now including through AI and their product Firefly. So Scott's had a front row seat to everything that's been happening with generative AI over the past year.

1:28He's also been a prolific angel investor, having been one of the earliest investors in some of the greatest companies of this generation, including Uber, Pinterest, Carta, and many more. And in fact, he was one of the first investors in the company I co-founded, Anchor, for which I am forever grateful. Scott was a fantastic advisor to Anchor on things like product, scaling, go-to-market, and even selling our company to Spotify. I learned so much from him throughout the process. And so I'm confident you're going to learn a ton from him here in this conversation. So take a listen. So very excited to welcome Scott.

2:03Scott, please join us.

2:11All right. Thanks for coming, Scott. Oh, thanks for having me. Well, let's talk about AI, generative AI. I mean, that's what we're all here for. And like I said, I feel like you've had a front row seat. For me, sort of the aha moment, I think everyone had different kind of aha moments for AI, but for me, it was maybe nine or 10 months ago when Dolly 2 launched. That was where, you know, sort of the light bulb went off for me. But what about for you? Like, what was the aha moment for you and maybe for Adobe as well? Well, I think that, I mean, one of the things that brought me, I was at Adobe, as you know, and I left for a brief period and then I came back, you know, as chief product officer five and a half years ago.

2:47One of the things that brought me back was was this idea of creativity for all. And I was like, oh my gosh, what are humans going to be doing in the future when a lot of stuff is just achieved through compute? And creativity really needs to become the next productivity. That's how people are going to stand out at work and in school and whatever. And when I came back to Adobe, I realized that the creative world is just full of friction. You know, it's really hard to get in. Like most people don't have the skills. Creative confidence probably peaks in kindergarten when everything we make, our teachers will put a star on it or our parents will put on refrigerators.

3:20And then we quickly realized that there are these things called critics and we don't have most of the skills that we would want. And then suddenly we kind of like go away from creativity, most of us. I always thought that was a shame. It's like, why does creativity, why does creative confidence go down as opposed to up over time? And then of course, there's also like the future of digital experiences and like what's next and 3D and immersive and whatever. So that's what brought me there. And then over that time, I just realized that the friction is there because it is hard. Like these tools are hard.

3:47And if you think about creativity as a box, with the floor kind of pretty high, it's hard to get in that box because of the skills and the capabilities and the expense of the tools, whatever. And the ceiling of the box is what you're capable of, given you only have so much time in a day. The objective was to make the box bigger, which, by the way, increases the TAM of Adobe and is good for the company as well. And there's been a million ways we've tried to do that. We've tried to improve the first-mile experiences, tried to create new products and web products and all these things, and they'd all make little dents here and there.

4:17But none of them were truly transformative. And then you see Dolly and you see some of these new capabilities that we were also brewing in our lab. And there was a light bulb moment for us where we were like, wait a second. Simultaneously, this lowers the floor of the box. Suddenly, people can just get in the box by just prompting something. You can start with something and just start to speak in natural language and make it into something in your mind's eye. That's an amazing way to lower the floor of the box. and you raise the ceiling of the box because suddenly people who are illustrators who never knew how to animate could just use generative AI to make an illustration animated.

4:53And you start to go down the rabbit hole of 3D and immersive. It used to be privy to people who knew physics and math. And now suddenly everyone can do it with prompts and with other capabilities that are made possible by algorithms. And so it was one of those moments where the ceiling goes up, the floor goes down, the box gets bigger. And that's where we were like, okay, we have to be all in. Right. Right. And did that sort of give birth to Firefly or did that come later? Yeah. So Firefly, you know, initially the first kind of white paper that I got was in 2019 from someone on my team that was supposed to be doing something completely different.

5:27But he's really smart and he kind of said, like, this is a big deal. We got to invest in this. And I remember being like, that's really interesting and also feels like really far away. So let's not fully invest in that yet, but let's like, you know, keep it on the back burner and feed it a little bit. And it was over that time that some of these other milestones in the industry happened. Crucially, some of them were like open source. But I think that the tide went up, so to speak. And it elevated the potential of all these different companies, including us. So when does that, 2019, way before anyone is really thinking about this stuff, or before most people are thinking about this stuff, when does that turn into Firefly, which is now not only just a series of models, but it's a product, right?

6:11I mean, it's really a product. That's right. Well, you know, like every company should make a very, very serious decision about what they're going to make themselves and what they're going to outsource. Right. And at the time, like, you know, I wasn't sure. I was like, are we going to be able to be market leaders in Generva by doing it internally? Do we just need a partner? And it became more nuanced. We realized pretty quickly that for LLMs, you know, these are massive, massive models and super expensive and constantly optimized. like we shouldn't necessarily build that ourselves to enable like AI assistance and that kind of thing.

6:43But when it came to imaging and video and 3D, as I did the inventory of the organization and the research organization and all the patents and everything else became pretty clear, like, this is something that we can do the best in the world. And also what became very clear is when I went to customers, they actually were super focused on whether they could even use generative AI, because they were concerned about how these models were trained. And so that's when we sort of said, But okay, we have the wherewithal to do this in a special way. And also, we need to do this in a different way than the industry.

7:13And that is actually what created the momentum to get Firefly into production. So maybe taking a step, I want to get to the right stuff. Maybe taking a step back before we get there. For those who don't know, maybe give an overview of what Firefly is today and what it's going to become. What are the features? What does it do? And where do you see it going in the future? Sure. So Firefly is a family of generative AI models that has started with the classic text to image. but then we also did text-to-text style or image style. So basically, like, if you want to make the word hello, but you want to make it out of sake bottles or you want to make it out of palm trees or whatever, like that sort of capability.

7:51And then we started to realize, you know, that there are a lot of, like, startups that are doing text-to-image prompts and there's Dolly and whatever else, but what actually people want to do for commercial use is they want to be able to prompt on a layer-by-layer basis and they want to be able to use the tooling of products like Photoshop to do much more with it. And so that was a realization to start bringing all these capabilities into our products. And so I think that we launched a product called Generative Fill just recently, which just absolutely blew our minds in terms of the reception, only because I think it was integrated into Photoshop and people, it just became like a meme in and of itself based on the way we integrated it.

8:28But I also learned or reinforced a product principle of mine, which you and I have talked about in many years past, which is, number one, the devil's in the defaults, right? In Photoshop, when you open it now, there's a contextual menu that's everywhere you go, and we put generative fill in it by default. And then number two is that when you can use this technology to reduce friction and make people feel successful more quickly, which is my definition for product-led growth, then people start to use it and spread it. They come back, they tell their friends. Exactly, which is exactly what we did with the integration of generative fill.

8:59Yeah, it's really been magical just seeing what people are making. with. And I've tried. I mean, it's incredible. Firefly, I feel like, is a great case study for what happens when really innovative, established players like Adobe go all in on the technology. And one of the things you and I have obviously talked a lot about, because we talk a lot about not only technology, but startups. When incumbents and established players move into a space, it really can cause a lot of disruption for startups. How should startups think about approaching AI when established players like Adobe hold so much distribution, so much data, and as a startup, you're basically starting from zero?

9:37Yeah, that's a great question. So let me put on my other hat here. And listen, I think that a lot of startups fail because they're founded by founders that have passion for a solution, as opposed to empathy with the people suffering the problem. And the reason is, is because you have so much passion for a solution that you just like see it and you keep going and going. And then you realize two years later that you're like 30 degrees off of product market fit. It's because you lack the empathy of what the customer's actually going through. What I say that is like, you know, enterprise customers, they are worried about what their boss thinks of them.

10:10They want to look good in the organization and they want to do more with less energy because they're lazy and, you know, whatever. And for consumer, like consumers, ego analytics, we've talked about this in the past, like, you know, consumers want to feel good about themselves quickly through a consumer product. So that's all empathy driven, right? That's not just passion for a solution, that's empathy driven. So I think that step number one is understanding the actual empathy with the frictions and the problems you're trying to solve. And not just saying this needs to exist, I'm passionate about it, let's build it.

10:38I think you have to have a data advantage, like you really have to have a moat. And so I would be focused on building a data set that an incumbent doesn't have, right? And I also think that on the interface, it's dangerous to compete based on interface innovation because it can be easily replicated. Stories, perfect example. Threads, you know, whatever, right?

11:01However, there's also, there's a disruptive interface approach too, which is you have such a orthogonal view of like how this world's going to evolve that you do it in such a way that like an organization like Adobe or someone else would have to like do some results backwards to be able to try to, you know, address that. And it's hard and big companies are slow. Yep. Boggles my mind every day. Well, Adobe is obviously not the only big company going all in on AI. I think even just yesterday, we saw news about Apple has Apple GPT or something coming soon. Obviously, we don't know what's going to happen with Apple.

11:33But for me, when I think about what they could do, it feels like it could be in the area of personalization. And this is something you've written and talked a lot about. You've talked about the personalization wave. Tell us a little bit about that and how it intersects with AI. Yeah, and I'm sorry to talk about this because I feel like there aren't enough startups tackling this. I don't think any big companies are truly thinking about this in the right way. And I'll give you my pitch real quickly. I think that in 10 years, you know, hopefully sooner, anyone like, you know, our kids and at least my kids' generation, 10 years from now, they'll be like, wait, you went to an e-commerce website and it asked you if you were a boy or a girl?

12:11It asked you whether you were, what, shoe size? Like, why would it not know? why would every commerce experience be generalized? Why would every media company have to deduce based on my watching preferences over a course of a year to serve me better stuff? Why isn't my preferences profile following me around? Why isn't every single digital experience made personalized for me? Even media is totally flipped. Now you still have a small group of people programming generalized content for the masses when you should have the masses programming content generalized for each or personalized for each of us.

12:47So I feel like the media model needs to be inverted. I think the commerce model, I think every website will be different. I think the immersive experiences that we have tomorrow through things like Vision Pro and all this other stuff that's about to hit, you know, everything I think the world is supposed is about to become hyper, hyper personalized. Let's dig into the media side of that for a bit. Everything you just described makes total sense to me, especially like the commerce, the experiences. When I think about media and I think about personalization of content discovery, right? You flip through TikTok and it's perfectly tuned, right?

13:19Then you take it a step further and it's like, oh, wow, content can not only be programmed for me, it can be created for me in real time. I see a good outcome of that. And I also see a bad outcome of that built on today's incentives, right? Around maximizing ad revenue. Like what does that end up looking like? And is there an opportunity for AI to personalize in a good way that doesn't lead to that sort of dystopian future. I think that's right. Well, firstly, I think that, I mean, one of my cardinal rules in technology is that we secretly long for the way things once were, but just with more scale and efficiency.

13:53So a couple of hundred years ago, if not more recently, we were all living in small towns where we were greeted by name everywhere we went. You know, we knew people on the street, the butcher knew our favorite cut, you know, everyone knew our kids' names, and that wasn't creepy, that was wonderful. And then we kind of went into this industrial revolution and the age in which we live. And then suddenly, like, we're all anonymous, you know? And companies need to, like, sniff information about us in weird ways to understand who we are. But I do believe that we actually do want to be welcomed by name.

14:24You know, if I go into Topping Rose for the third time and Major D is like, welcome back, Scott. Do you want your favorite drink? I'm like, this place is amazing, right? So why can't we use technology to sort of enable that but in a way that we're comfortable with? So I do believe that personalization has that opportunity. But to your point, we're also about to be like totally bombarded and inundated with content that is like strangely personalized for us. And every brand is going to flood the zone because they can. I mean, if you want to make a thousand SEO posts tomorrow around three keywords, you can basically do that in an hour.

14:56And have all the benefits of Google, which is why Google's algorithms are so outdated at this point in terms of what's relevant. So there's going to be disruption there. And so that's why part of me is also like, you know what's going to come back? meaning and soulfulness. Yeah, it's so true. There's obviously been so much discussion over the past several years around the media and fake news, but it almost seems like we might soon get back to a place where the most trusted or the most longstanding institutions are going to be trusted again, right? It's going to be like, oh, the New York Times said this.

15:31That means something, right? I think that's true. I mean, we all know we're kind of entering an era where we can no longer believe our eyes. Right. And I like to say the trust but verify is now going to become verify then trust. And the question is, how will we verify before we trust something? We're advocates of something called the Content Authenticity Initiative, which is an open source framework that we started working on in my team five years ago for deep fakes. And now it's very relevant for the generative AI era. But it basically adds content credentials to media so you can see the provenance of it.

16:02But I think that we're going to have to, you're right, brands will matter more. because we'll need to know if we can trust. So let's get into some of that. Let's talk about this sort of regulatory environment because it feels like we're on the verge of some big shift, right? You mentioned earlier at the top of the discussion that one of the potentials of generative AI is to bring the ceiling up, bring the bottom down, bring the sides out wider, enable more people to be created. Do you consider what is happening now is generative AI is enabling people to be creators or is it enabling people to be, as I think I've heard you say on certain podcasts, creative directors?

16:41And what's the difference? Yeah. First of all, I believe that great creatives want creative control. Yeah. And so I think that we're going to move past the prompt era. Like right now we're in this like you prompt and you get a dog, whatever, with a tree behind it and you're sort of, you can't control exactly where the tree is or what the tree looks like or the relationship between the dog and the cat in the tree and all that stuff. I mean, that's not going to cut it in the creative professional world, right? So I do believe that every creative has a set of skills and does work, but is also a creative director of themselves.

17:15And then oftentimes they also have creative direction that's augmenting the work that they're doing. So I think that the people that are closer to the floor will really just be creative directors. So if you're a social media marketer and you're trying to think and act in real time as a brand, Remember that moment when the lights went out during the Super Bowl a few years ago? And that was a really key moment, I think, for the world of marketing. Because within 30 seconds, Oreo came out with this campaign, You Can Still Dunk in the Dark. And I was like, holy shit, that was so amazing. How did someone in the organization or the agency that represents Oreo, how are they, A, empowered, how do they, B, have the source of truth brand assets, and C, who gave them approval within 30 seconds to do this?

18:00And of course, it was probably someone just going rogue and being instinctual, which is great. But every brand is going to need to do that in this modern era. And how do you enable that? I mean, generative AI is kind of the only solution, if you think about it, for that. And so people on the social media marketing level are going to have to be able to be great creative directors and just execute. Using source of truth brand assets and using train-your-own-model variations like Firefly, we're going to make versions of Firefly for brands based on their own brand collateral to enable that. So I think that's part of it.

18:30But I do believe that in the ceiling level, you're still going to want granular tools. Sort of on this topic, I feel like we may have just had AI's Napster moment. A couple of days ago, I don't know if you saw the South Park AI. Did you watch this? No, but I heard about it. I mean, it's pretty much perfect. It is a perfect replica. And I think on Stratechery, you said to Ben, hey, sometime in the near future, this is going to happen. And it just happened. Yep. What happens next? Yeah. So, I mean, Ben Thompson and I were talking about the era of like unauthorized sequels. And the fact that anyone, I mean, there's so many implications for this.

19:08Like, you know, when you train on any actor's likeness and voice, then you can basically write a screenplay and have the actors do everything you want them to do. And suddenly everyone's making unauthorized sequels of everything. And it's just like this Pandora's box opens. It's an IP nightmare. And at the same time, it's like super creative. So how do you reconcile this? I think there has to, so that's why I think that's a great analogy. Like it's either the Napster era, which means that it has to like all be shut down, or it's the Uber, Airbnb era, which is that everyone just has to adjust. And this kind of has to end up being okay, even though we're uncomfortable with it at first.

19:48And remember, like in the early days of Uber, every town, every country had its own legislation, its own fighting. and same with Airbnb. Yeah. I think it's going to be more of the Napster approach. I actually do think that, because the only, and I'm not a lawyer, but one other thing I've learned about IP law is it's very focused on the economic detriment to whoever's IP you're violating. And so if I'm inspired by your style and I put in my mood board and I make something that's sort of like it, that's typically fine. But if I literally use your voice and likeness and you're an actor or a voice actor or whatever, and then I put something out there that people use at the expense of buying your material or paying you to do your job, that's traditionally not fine.

20:33I can see how the latter example gets shut down. The former sounds complicated, right? So if you're inspired by me, do I get to participate in the commercialization of that in any way? And how does that happen? Yeah, I mean, part of me is inspired by your former company, Spotify, where they build an attribution model and a compensation model that at first no artist was happy with, but became better than the alternative. And I wonder, in my mind, I'm thinking, well, why don't we let Behance members who have these amazing portfolios of really unique styles train a model and then make that model licensable by other people as they make prompts and generative AI tools and have them monetize even when they're sleeping?

21:17You know, it's so funny because many decades ago, there was a royalties infrastructure built for music where songwriters are now able to get paid no matter how big or small their contribution. It's a completely antiquated system. But in some way, it almost feels like a model for what might come next with AI. I think that's interesting. And I think the other thing to think about, I always go back to that other product principle I've talked about before, which is at least in the first 30 seconds of our product experiences, but also especially the enterprise, everyone's pretty lazy, vain, and selfish.

21:47Yep. That's kind of the truth for products. And so when you if you made it easier to use a great style in a licensed fashion, as opposed to going to some rogue genre eye tool and using something and maybe having liability and not having it be smooth and clean, like people may actually just pay, like they do with music, because it's easier and it's legal. And it's like straightforward, especially for enterprise use for commercial use. So I think there's an opportunity to build a marketplace. Let's get into the enterprise bit a little. Large language models in the workplace. How do you think this is going to transform the way we work, whether it be organizational design?

22:26I know you wrote recently about collapsing the talent stack or improving communication and streamlining meetings. Where do you think it's going to have the biggest impact? Well, it's going to change everything in some ways. But I also think that that doesn't mean it's necessarily that they're all good businesses to invest in, right? these LLM, and Microsoft is integrating it into every point, right? I know Salesforce is now, and like others are just, you should be able to use any product in the world with natural language. And if you look at what a assistant within a product can do, I sort of see it as a pyramid.