Ep 36: Adobe CPO Scott Belsky on How AI Will Transform Creative Workflows

Scott Belsky is the founder of Behance, a creative platform sold to Adobe in 2012. He’s since gone on to take on the role of Chief Product Officer at Adobe leading design for all products across Creative Cloud, Document Cloud and the Digital Experience business. This week on Unsupervised Learning Scott shares his thoughts on the future of creative tools with AI, a future where hyper-personalization wins and what humans will do when content is commoditized.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Jacob Effron?

  AI is transforming creative tooling today across images, video, music, and more. And today, we had the perfect guest to discuss this all, and Scott Belsky, the founder of Behance, prolific investor, and chief product officer and chief strategy officer at Adobe. Scott and I had a really interesting conversation that hit on when companies should train their own models and how Adobe thinks about this. We talked about startup opportunities in the creative tool space. We talked about what will matter to people as brands become increasingly hyper-personalized. And then we also hit on the future of Adobe AI and what's still left to be built for enterprises. I think folks are really going to enjoy this episode. Scott, thanks so much for coming on the podcast.

• 0:33Speaker 01

  Thanks for having me.

• 0:34Jacob Effron?

  I know we'll dig into a lot of things on the intersection of AI and creative tools and a lot of the exciting work you're doing at Adobe. I'd love to start with, it feels like every week, there's some piece of AI-generated content that just goes viral. And I'm wondering, of all the stuff you've seen, what's your favorite piece of AI-generated content that you feel like is providing a window into where we might be going in this space?

• 0:54Speaker 01

  There's obviously the mainstream poke-in-the-white-Gucci jacket or Dolce Gabbana jacket type stuff. But I'm particularly more interested in the way that creative professionals are unlocking new workflows. So, an example that went, I don't know if the word's viral, but amongst creative professionals, we released a product called Project Neo, which is basically a 3D illustration program where you can render any 3D structure. And then people realized that you could take that, make something, and then use it as a structure reference for Adobe Firefly, our image model,

• 1:35Speaker 01

  and actually give Firefly extreme precision in direction of how to generate something that you prompt. And we didn't actually expect this because this was a 3D product that was made for illustrators and had nothing to do with the Firefly image generation capabilities. But this workflow got unlocked, and then suddenly you see on social hundreds of people doing this to show their audience and flex, like, "Look what I discovered." And it just popped. And I think my big takeaway from that was that our obsession with the next best model is probably something that's a little passe

• 2:18Speaker 01

  and that we are now going to an era where it's not about the next best model, but it's really about the controls that one can apply on top of the model.

• 2:27Jacob Effron?

  Yeah, I love that, that like you'd obviously released kind of two very separate, you know, really interesting features in the Adobe products. And then, you know, folks were just playing around with it and found ways to combine it in ways that you might not have expected. But it's really helpful in kind of control and guiding your underlying image generation model. I mean, I think it gets into, you know, maybe before we dive in specifically, I'd love to just kind of, I think for our listeners sake, get an overview of kind of all, I feel like every, you know, every month you guys are shipping something new on the AI side. And so, you know, I think maybe it would just be helpful to get some, some quick context on those tools before we, before we dive in.

• 2:58Speaker 01

  Yeah, sure. Well, you know, at Adobe, we're very focused on the interfaces, the models, and then the data underneath. And so, you know, and I'll start with the models. The Firefly family is like a, is our, is our own, you know, homegrown set of generative models that are all trained on licensed content. And we have a, we have a compensation program for those that contribute the content to this pool. And we really aim to blanket indemnification of all of our customers, and we really care about how these models are trained. But on the interface level, we're trying to figure out ways of integrating these models into the Photoshop type products of the world.

• 3:42Speaker 01

  We're also increasingly making these models available with third parties into big companies that need to do like big, complicated, at-scale workflows. And instead of having to have individuals do these things piecemeal, like could they do them at scale automatically? And then also building a lot of these control capabilities, like structure reference, which I just mentioned, and style match and all that kind of stuff. But on the data side, we're very focused on unleashing custom models for customers. So imagine Nickelodeon training a version of Firefly on SpongeBob SquarePants, where they can really, without any concern, generate tons of stuff to ideate and explore new character development and storylines, right, et cetera. So

• 4:25Speaker 01

  that's, that's really been our focus. And then we have some partnerships on the LLM side, so we don't build our own. Those are partnerships that we surface, you know, through Acrobat, through our AI assistant, through the digital experience products, for marketing analytics, you know, inquiries, that kind of thing.

• 4:40Jacob Effron?

  I think one thing that's really compelling that you've done on the Firefly side is obviously, you know, been super focused on making sure your customers are legally protected and kind of, you know, training in compliant ways with data. And I remember when the models first came out and people were prompting Spider-Man into them, you were saying like, yes, this is a feature, not a bug that you're not going to get Spider-Man at. There's like all sorts of, you know, obviously legal issues with that. And so I think it's, you know, obviously a really, we hear from customers all the time. That's, that's kind of always top concern on their list. And I think you guys have done, yeah, have done a really good job with that. I guess like, you know, the example you mentioned earlier about, you know, folks kind of combining different features, I think is really interesting because it feels like we're still in such early innings of just like using these tools that have come out. And I imagine you see all sorts

• 5:22Jacob Effron?

  of different Adobe users using the things you've shipped. Curious, like, what are the folks that use them best do? Like, any things you've started to see on like the power users or like, you know, something that separates, you know, maybe folks that have just messed around with it, but folks that are using this like, you know, just as a routine part of their workflows?

• 5:38Speaker 01

  Well, one thing I've learned over the years, even in the Behance days,

• 5:44Speaker 01

  was that the greatest creative professionals really are very good at covering the full surface area of possibility to solve a problem. And they'll pursue little mistakes of the eye here and there and go down the track and see if there's something there. And if not, they'll go back and then they'll revert and try another path. And, you know, ultimately, when a, when a creative professional comes up with, you know, two or three solutions for their customer or client to choose from, you know, that's the end of the process, right? That probably went from 12 dwiddled down to three final potential solutions to a problem. But the constraint that they're always fighting against is time. Like, they just got so many cycles in their life. Like, they can't actually, you know, pursue too many things. They just don't have

• 6:28Speaker 01

  the time. And so to see folks use these tools to just explore far more surface area of possibility far more quickly is just a real consistent theme that we're seeing across the franchise. We launched this thing called Generative Recolor, an illustrator that allows people to take Hundreds, if not thousands of color palettes based on prompts and then apply them to their vector creations instantly. And then pick the ones that they think are the best color combinations like that that used to take days like you had to have groups of your people dedicated to that. And now they can be unlocked for higher order exploration of the actual vector objects themselves as opposed to just the color variations.

• 7:12Speaker 01

  So I think what we're seeing is people take advantage of this for unlocking more cycles of discovery. To find ultimately better solutions to problems that they didn't have the time to find previously.

• 7:22Jacob Effron?

  So we talked to a bunch of startups and I feel like they think a lot about this blank canvas problem where there was all these really powerful AI features. And their users, you know, kind of face this blank screen and they don't exactly know how to get started or how to begin using them. I'm sure that's something you guys have thought about a lot on the product side. You've obviously introduced a lot into your products. How do you think about user education and, and, uh, you know, it's, it's obviously in some ways a very new way of working with some of these tools.

• 7:44Speaker 01

  Well, I think it's a great question around, yeah, just because you can give someone some of these new superpowers doesn't mean that they know how to use them.

• 7:53Speaker 01

  I mean, when the camera was first invented, you know, I was kind of going back in history, trying to make, you know, pick up the rhymes of what we can learn. And there is a lot written about the artist who used to be hired by families to do family portraits. Who are deeply offended by the idea that someone could just capture a family portrait with clicking a button, you know? And, uh, and it was just, it was a disaster, right? In terms of the, the, the, the shift in the industry that the camera, uh, uh, sort of insinuated. And yet photography ended up becoming an art in and of itself. Why?

• 8:31Speaker 01

  Because you had to determine how you would capture the image, what lens you would use, which is sort of almost like what prompt do you use? You know, lighting, and then also, once you actually capture hundreds, if not thousands of photos. You have to choose which ones are best and that is taste. Which is in some ways like a skill in and of itself. That's more important than a lot of the rudimentary skills that are now being offloaded to compute. And so I would argue that taste is more important than ever before. And, uh, you know, and I, and I, and I, uh, when it comes to creative professionals. Trying to figure out how to use these tools, like our job is to make sure that they can flex their taste.

• 9:16Speaker 01

  So how can we do that with prompt augmentation? How can we do that with onboarding experiences? How can we allow people to take where it is in their mind's eye? And you know, the better vision you have, the better your outcome should be like that's our job.

• 9:29Jacob Effron?

  Yeah, it's interesting. I feel like taste comes up a lot, even even on the model side, right? I think a lot of what people credit the success of mid journey with is, you know, was very good taste on the on the on the kind of data side. And I'm curious, like, you know, maybe to that point, if you see, you know, a world in which we, you know, converge on kind of a single creative model? Or do we end up in a world with a bunch of different models with different aesthetics and different, you know? This is for for folks with this kind of style or taste? Or where do you think kind of the underlying model world heads?

• 9:56Speaker 01

  I think the models themselves, we will have thousands to choose from. I think a lot of them will be commodities. And I think a lot of them will just focus on very niche use cases, train on a very specific amount of data and tuned for very specific reasons. And I believe that tools, you know, like the ones that we make should allow people to choose different models for different use cases. You know, we talked about the importance of having commercially safe models that are trained on licensed data. However, if you're doing just a mood board. I mean, mood boarding is one of the most popular features on Behance. And no one has ever once complained to me that we allow people to make mood boards using other people's images. Like, that's how every creative process starts.

• 10:41Speaker 01

  So why shouldn't you be able to use models that are trained on other content for that purpose? So long as you're not using the content for commercial purposes, it's just like an ideation exercise. And so I want to, I want to start to lean into that. And I, and I would like us to, to work with the ecosystem of models that people might use for these use cases.

• 11:02Jacob Effron?

  Yeah, no, it's a really interesting point. Obviously, yeah, there's different, obviously aesthetic needs, but also different kind of, you know, compliance and data requirements for different parts of the creative process. And so it makes sense to have a bunch of these, a bunch of these models. I mean, do you think in the long term, like, is, is, you know, is having your own model something that's super important? Or was it like, obviously, at this moment in time, there wasn't necessarily a compliant one that you could go bring to an enterprise customer? Or like, you know, bring to a Disney and kind of guarantee them. So was it a moment in time issue to have to build your own model? Or do you think long term, it's always going to kind of make sense for Adobe to have their own state of the art one?

• 11:35Speaker 01

  Yeah, I think for the areas where we are world experts, we're going to be able to deliver ultimately a better end-to-end experience with our own models. So in these areas of media generation, it's not just getting the training data and building like a massive, you know, many parameter model. It's also about the fine-tune controls, the custom model capabilities, leveraging a lot of the data around how someone actually uses our tools to fine-tune the actual output of the model. You know, it's funny, when you look at a lot of people's prompts for these models online, people are specifying the lens that they're using, the lighting, you know, the contrast density, like all these, you know, these colors,

• 12:17Speaker 01

  sort of hints, tints and whatever. I mean, these are all in some ways like Lightroom dials and filters. We know that better than anyone else in the world. So like, we should be leveraging that data set as well to enrich these models and enrich the outputs. So I think that's like the litmus test for my team, is if it's something that we believe we can do better than anyone else in the world, it's our responsibility to do it. However, if it's something that is a commoditized, or some other company is going to definitely invest more and always be having something better than we have, like let's partner with them.

• 12:51Jacob Effron?

  No, it makes a ton of sense. And I mean, I feel like a lot of companies are thinking through now, like, do they build their own models? Do you kind of fine tune something that's open source? And, you know, I really like the idea that like, you know, there's, I'm sure a lot of what you're seeing in prompts, you can put some of it into models, and then some of it actually into just the features you already have in your product, like whatever the person is asking. And so it kind of creates a really interesting opportunity for you guys on the product side.

• 13:14Speaker 01

  And one other thing, Jacob, is that right now, it's sort of like, we're all focused on this command line, agent-driven experience, or prompt and get something. But the next generation of this is when you ask for something, and then not only use the model, give it back to you, but it gives you like custom UI to fine tune what you just got. And it's almost like UI on demand for the actual use cases that you might have at the moment. And then when you're done, it just like disappears. And, you know, that's another generation of what these models will enable that, you know, suggests to me, we have to be building our own to unlock that.

• 13:53Jacob Effron?

  Right. And so basically, rather than having like one standard UI like you're using, you know, Photoshop or Illustrator, and this is what it always looks like. It's like based on what you're trying to do, it can kind of spin up an exactly custom UI for the, you know, the long tail of many use cases of things that people are trying to do.

• 14:08Speaker 01

  That's right.

• 14:09Jacob Effron?

  That's really cool. Um, where, where are we on the journey to that?

• 14:13Speaker 01

  It's coming, you know, if there's a, we ought to, you know, we, we also have to make sure that this stuff is professional grade. And I think that's the tricky part. I mean, you see all these startups and everyone saying, oh, we have a video model, a video model, a video model, video model. There's like already half a dozen or probably a dozen at this point. You know, we are building our own video model. But when we focus on specific customer use cases for commercial grade media creation, the bar is really high. Like it can't look weird. Yeah. And, and I think that there's a, much like self-driving, I mean, we were talking about self-driving cars 15 years ago. And it took like kind of 15 years for us to actually be able to get into a Waymo in

• 14:53Speaker 01

  San Francisco and go around. And that's like kind of just happening today. So I actually think there's going to be a big gap there in some areas between the cool happy path demos and the actual pragmatic, everyday commercially viable, you know, quality use cases and some of this stuff.

• 15:13Jacob Effron?

  I mean, I find out all the video models, right? It's like the demos are great. And then you prompt it yourself and you're like, that didn't quite look like the demo. Actually, I'm curious, what, what do you even see as the milestone? Like, what do you think about as like, oh, that would really unlock a lot if videos could do, like, what are the milestones you think about?

• 15:26Speaker 01

  You know, I have a much more pragmatic view right now because I'm just really focused on the things that customers will really value. So if you're editing a movie in Premiere Pro and your scene is three seconds short of what you need it to be because of the song mapping or anything else for that matter, or just like the mood and the emotion and the, you know, the, the director's like, I just need like two seconds more. And you're like, I'm sorry, I don't have that. You know, if we could let you extend by a few seconds, would that change everything? You know, would that save going back on, on set and reshooting the whole scene and all that stuff? A hundred percent. So can we like double down our horsepower to unlock something like that?

• 16:10Speaker 01

  You know, as opposed to, um, you know, some generate aerial footage via B-roll or, you know, any sort of geography in the world for 30 seconds of high quality or 4K footage. You know, I think that some of those things might be a little further off.

• 16:26Jacob Effron?

  And I guess like, I mean, I'm curious how you think about just the broader, you know, AI for creative tools market. I thought we see a lot of companies in the space and, you know, obviously the people go endlessly. We'll talk about, you know, is, is AI something that just benefits incumbents and folks like yourself that have amazing workflow products already and can layer these models in, or are there opportunities for net new companies? And I'm curious, as you look at this like broader landscape, like what are the opportunities that you think are exciting from a, from a startup perspective?

• 16:50Speaker 01

  Well, you know, my bet, both internally and also as I look at the market for where I think there is real opportunity, especially in the creative and marketing space, is that, is, is the differentiation at the interface and the data level and the data. And then again, the models, it's like, unless you absolutely have to build your own, you have some advantages if you do, but otherwise I think they'll become further commoditized. And most of these startups are using third-party models, right? So they need to, so then the question I guess is like, really, where is this industry going? So I think that the creative and marketing space is going towards hyper personalized digital experiences at scale. Like all of the websites we go to will speak to us as if

• 17:31Speaker 01

  they know who we are. The emails we get will be hyper personalized to us. They will have a video that's made featuring the stuff that we bought, you know, and all that kind of stuff. And to do that, what do you need? You need the workflows. Uh, firstly, you need the customer data. And then you need, um, and you're, Adobe has like a customer data platform that we do for a lot of big brands. You need the workflows and marketers use to not only deploy, but also measure and optimize campaigns. You need the creative stack connected to this because you need to make these templates with guardrails and, you know, have sort of brand check AI that makes sure that you're not leveraging copy that's inappropriate. And, you know, it takes one asset being really bad to like screw up a brand.

• 18:15Speaker 01

  And so there's a lot of like fine tuned stuff in that workflow that makes me, and it's really not models. It's like interface workflow type stuff and data that I think are real key differentiators. So when I see these startups pop up and it's like, oh, you know, we do that. I'm like, well, how do you really do that? You know, how do you have all these different pieces? And so that's why I, you know, I, I'm a little, you know, I, I push a little bit in prod, you know, when I see these companies making some of these promises. But I do think there was a ton of opportunity in spaces that, um, that, you know, were extraordinarily cumbersome before. A lot of the ways small businesses work and manage themselves.

• 18:59Speaker 01

  You know, I, I've started calling it the, uh,

• 19:03Speaker 01

  the sort of, you know, this, this era of small businesses, but like, operating as huge businesses. And I'm not sure exactly how to like, how to come up with an acronym.

• 19:13Jacob Effron?

  They have a marketing team of a hundred people because of some of the, the capabilities these tools enable.

• 19:17Speaker 01

  Yeah, they have the equivalence, right? And so that's an area where I do think startups could really innovate because that's not something big companies focus on is like helping small businesses operate as huge businesses. I also think, you know, tackling spaces like law or government or all these places where there's just such ridiculous, um, old school antiquated process.

• 19:41Jacob Effron?

  I know you've talked to us before, like this kind of move to hyper personalization and everything. And I mean, maybe it's just, it's kind of a zoom out broad question, but as we think about like hyper personalized media in the future, I'm like torn between two things. One is like, um, obviously, yes, everyone would like to have, uh, content that's tailored exactly to, you know, what they're interested in, see themselves in stories and whatnot. There's lots of places your mind can go with that. On the other hand, there's obviously this like shared social experience of like, we all consume very similar content. Like where does that all go on the media side? And how do you think about that balance?

• 20:10Speaker 01

  I think it's a great question because I always think about where the pendulum will swing and also where it will swing back as a result of it. Right? So every brands are about to flood the zone. We are going to get so much hyper personalized content delivered to us every, everywhere we go and every interaction. You know, the next generation under us, they're going to wonder how websites used to ask us, you know, gender or shoe size. Like, how could they not know that about me? That's crazy. And you know, that's like the era that we're entering. And, uh, and, and content will become free to generate and very, you know, create variations and acculturize and translate and all that stuff. So that's going to make us totally inundated. And as a result, what are humans going to do?

• 20:55Speaker 01

  Well, every time something becomes commoditized, we end up seeking more sort of scarcity and meaning in that category. Shoes are commoditized, and so we wear more expensive shoes that flex our cultural sensibility, right? Cars are commoditized, and so we buy luxury cars. Recently, I've been thinking about what that means for luxury software. You know, I've also been thinking about what it means for the role of humans and creativity.

• 21:22Jacob Effron?

  I actually have like artisanal, handcrafted software, like, you know, this wasn't prepared by AI coders. It was really made by like, you know.

• 21:28Speaker 01

  Maybe, you know, like when superhuman, you know, manually onboards you to their products, you know? It's it's kind of like this is a luxury software experience. I mean, we're basically an interface on top of Gmail for most people, but you're paying for it and it's great. And it's a flex that you value your time more than the average person, you know? It's kind of a good example of an early use case. So I think we're going to, I think we're all going to crave scarcity, meaning and craft in the digital experiences that we consume. And I believe that will actually reboot the role of humans in content creation. And I agree with you by the way, that we're not going to want entertainment to feature us in it. We're going to want to have a shared experience that we can talk about with our friends. I don't think that's going to go anywhere.

• 22:12Jacob Effron?

  What about music? I mean, I feel like we've seen all this incredible, you know, AI-generated music over the last two, three months. And I feel there are some folks who have a vision of, you know, hey, maybe we just go entirely AI-generated music. And other folks see this as kind of like collaboration tools with with artists. Like, where do you think kind of that whole space goes?

• 22:28Speaker 01

  Well, one of the things that I think is missing is that a lot of these music startups are so focused on the role of AI. And they're not remembering that the reason we listen to a song over and over is oftentimes because we know the story behind it. You know, oh, this is one of Taylor's breakups and everyone's wondering who it was with. And like that sort of allure, that folklore ends up driving the value of the experience. So why aren't we allowing people to not only generate music with AI, which lowers the floor, lets more people participate as music creators, but also tell a story that is very human behind it so that you resonate with it.

• 23:10Speaker 01

  You know, if I tell you this is the song that this is the, this is the, this is the moment in my life when, I don't know, like something happened to me and it was very profound. Again, you're going to be like, oh, I want to listen to it. I want to hear it. Now I'm going to think about the words, like what the meaning is. And that engenders like this relationship between the artist and the, and the, and the consumer that I think a lot of these AI tools are still missing.

• 23:35Jacob Effron?

  There's obviously a lot of analogues to, to, you know, the other parts of, of, uh, you know, creativity and artistry as well. And, you know, I guess I'm curious, you know, what are the capabilities or other milestones across, you know, generative image, generative video, uh, that would be meaningful to you today as you talk to end customers, you know? I'm sure, uh, you get all sorts of requests, like what, what can't we do today? That is, that would be like incredibly helpful to them.

• 23:58Speaker 01

  The, the power of generative AI at the very base is about like glorified help, helping you figure out anything, answering your question, searching the web for you, all that stuff. One ladder up on that is like doing things for you. So instead of having to learn tools in a product like Photoshop, the AI just kind of does it for you and you can kind of hop, skip and jump through the creative process with a command line experience and, and get things done for you. But I think like the next year after that, that a lot of our customers are starting to ask about and we're starting to work on is about the proactive suggestions of things you didn't know you should be doing or trying, right? So if you're trying to solve a problem and suddenly the, the agent sort of suggests

• 24:40Speaker 01

  you should explore this way or explore that way, or that wouldn't perform well in that geo or make a variation of this for this version, you know, that's like a superpower. And, uh, and I think that that's kind of where we're going.

• 24:55Speaker 01

  Part of, and then the other thing I think all of our customers want is just increased speed and quality. So a lot of the unlocks right now are really just around speed, latency issues.

• 25:05Jacob Effron?

  As you guys have been building these features, obviously you could suggest things throughout the process. I'm sure there's inference costs you have to balance. Like, to what extent have inference costs played into this? Like, are there things you would do if the models, you know, obviously they get cheaper by the day, but if they were, you know, 100, 1,000 times cheaper, are there different things you would do? Or is cost not really a consideration, given the value you're able to drive on the products?

• 25:24Speaker 01

  Well, cost is definitely a consideration because it allows us to do things faster and, you know, and there's always a parallel track of innovation around how do we reduce the costs and efficiencies and increase efficiencies of our models, which is healthy for the business and as well as the customer experience. But, you know, when I go to the design team and, you know, and talking to the product teams around what we should unlock for the customer, I'm never having them work within constraints of cost. I want them to unlock like whatever's in their minds eye. And if there's enough value associated with it, I mean, our business model is an economy that we control around generative credits. And so if I can allow

• 26:06Speaker 01

  you to do something that is extremely intensive with using up generative credits, but if it's worth it to you because it saves you days of work, then great, you're going to be willing to pay for it and we should deliver that. And we shouldn't like not just because it's expensive for you, you know, as the customer. So, you know, we're not, I don't think it's constraining our innovation, but we are definitely also innovating on the cost side.

• 26:30Jacob Effron?

  Makes sense. And how have you thought about pricing? I mean, I feel like so many folks are still figuring out, do we price separately for these or rate limit, you know, what you can do and, you know, charge more if you want to do a lot of calls? Like, how have you guys thought about pricing for these AI features?

• 26:44Speaker 01

  Yeah. So what we've done is we came out with a very basic amount of generative credits that people are entitled to with their existing plans that they have with us. And then we've seen a subset of customers break through that. And we just made, you know, new plans for them to upgrade to that give them, you know, more generative credits. And it's as simple as that. And there's kind of no risk in this approach because as we unlock new capabilities, whether it be generative video or other capabilities that are more intensive, we can just change the generative credit, you know, amount that something costs our servers to do for you.

• 27:25Speaker 01

  And but our mission is to keep that as as cheap as possible because we want it to be accessible and we want it to be a part of everyone's workflow. And, you know, I think we're very close to a time where everything that anyone makes has some degree of generative AI in it.

• 27:40Jacob Effron?

  Totally. And it makes sense. I mean, with pricing directly, you know, in a way that's directly correlated with your cost. Obviously, you know, it kind of just aligns everything nicely and other new features or video or whatever it is fits in nicely to that way of doing things. I mean, I we always like to end our interviews with kind of a quick fire round where we get your thoughts on on a few common questions we ask guests. So maybe to kick things off, you know, I'm curious. What's been the biggest surprise in building AI features for Adobe? Maybe like one thing you thought would work that didn't or one thing that that worked, but you thought it wouldn't. Yeah, love to start there.

• 28:12Speaker 01

  And I'll go back to, you know, my friend, my friend Dave Marins, a classic product saying that the devil's in the defaults. That is so true. When we put generative fill, you know, in Photoshop in this new default generative, you know, generation bar that people saw as soon as they came into the product, it totally unlocked utilization in a way we could have never forecasted. So now the question is, how does AI become a default part of how people use every product in any way that would actually give them superpowers? But it's a, you know, it's a different playbook from our typical, oh, let's debut a new feature and hide it in a panel in the third tab, you know, that kind of thing.

• 28:57Speaker 01

  It's a, it's a different ballgame.

• 28:59Jacob Effron?

  What's something you've changed your mind on, like since the start of the, of the, of the post chat GPT AI craze?

• 29:05Speaker 01

  I think I definitely have changed my mind on, will there be a few models that win? You know, now I'm realizing, no, I actually think there are going to be thousands of models. I think that a lot of stuff will end up happening on the edge. The, the, the increase of capability of these models is happening so fast and so rapidly that they're displacing each other every day, you know, with like new things. But the thing is, is that as this, as this happens, the curve that most use cases that people have falls below the frontier of what these models are capable of. And that means that we're going to be a little less choosy on maybe which model does things and we're going to be starting to focus on costs.

• 29:49Speaker 01

  Uh, as we deliver these capabilities to our customers and especially when we have these, you know, local or open source models that we're leveraging as well. So I just didn't expect that. I thought that, gosh, there'd be like one company or, you know, three major players that are going to be the next Googles of the world. Uh, maybe that's not true.

• 30:07Jacob Effron?

  Do you worry at all about like a frontier model player like OpenAI or Pika getting so far ahead on video that like, you know, they do, they, they transform something that captures a bunch of people before it's kind of an open source?

• 30:18Speaker 01

  I mean, first of all, you were working with folks like Pika and Sora, you know, in our, uh, you know, in our, in our products. So as long as people still need an interface to tell a story and they need to leverage data to customize it. And they need to have all the tools to make it look like whatever's in their mind's eye. Like we're, we're winning when every model gets better. It's like, these are platforms to us. And every time Apple comes out with a better chip, our products get better. Every time these models get better, like our products should get better. Um, we're also building models that unlock things that we think only we could do. And I don't know if those companies would be able to do that because they don't have that like full stack and image. But, um, Yeah, so you know, I think that they're going to be part of the ecosystem and in many ways I'm rooting for them.

• 31:00Jacob Effron?

  What startup outside of like, you know, in AI startup outside of the AI creative space, like are you most excited about right now?

• 31:05Speaker 01

  Gosh, you know, it's hard to, uh, hard to say. I mean, I guess one I'll pick is this company called Cobalt, which is a company that's leveraging AI for identifying mineral deposits. And, you know, this is a

• 31:19Jacob Effron?

  Was that, was that a personal passion before you met the company or?

• 31:23Speaker 01

  No, only when I met the company, um, did I realize like how antiquated the whole world is of these miners. They, their R and D departments are dilapidated. They don't even allocate any cash towards that stuff anymore. You know, when a mine comes online, it's $10 billion, you know, of value and it takes billions of dollars to even build a mine. And it's just so massive and it's so inefficient. And then you think about the application of AI, if it can increase the identification of mineral deposits by, you know, four to 10x in terms of, um, both speed and reliability, but also reduced cost. Then you kind of have a global changing, you know, company, right?

• 32:07Speaker 01

  I mean, these are the minerals are going to power batteries and EVs and, you know, it's like, this is a big deal. So I just, I think it's a great application of AI.

• 32:16Jacob Effron?

  Yeah, it really touches everything. I guess, you know, I mean, obviously I feel like you have one of the most interesting jobs in the world right now. You know, sitting on top of all the stuff at Adobe, um, you know, if you weren't building an Adobe and you were kind of, uh, back to, to the entrepreneur world, um, you know, what, yeah, and not mineral deposits, but where else, where, where else would you go build?

• 32:32Speaker 01

  Well, I think one area I've been really focused on and, uh, you know, thinking a lot about is around personalization. You know, every, every website we go to is going to be a conversation, you know, that we have with a brand. And, uh, in every physical space we go to. We're still strangers and these agents still treat us like strangers. And there might be memory in between sessions, but like all your long tail preferences that make you who you are, when you actually are remembered at a restaurant and they know your favorite drink, or when you, you know, have a personal shopper you've worked with for years at a store or something like that, it's really magical. How can AI unlock that for everyone?

• 33:14Speaker 01

  I remember when, you know, when Garrett and Uber and Travis were first kind of pitching the idea of Uber, it was like everyone's personal driver. And it was a sensation of something that no one, you know, very few people were entitled to suddenly being available to everyone. And so what's the, what's the version of that for hospitality and for feeling known wherever you go is very interesting to me.

• 33:38Jacob Effron?

  Yeah, I love that. I mean, I think so much of, I feel like what we've touched on is this theme of just, you know, stuff that was, you know, obviously not scalable and only big companies had, or you know, maybe the finest restaurants or whatever, you know, now you can, you can scale across the board. And I think to your point, there's a whole set of businesses and that might not have had access to these before ever thought they could do this. And that's a great opportunity for, for startups.

• 33:59Speaker 01

  Yeah, I think so. So I'm excited to see folks, there are a bunch of folks building that space, you know, see what happens.

• 34:05Jacob Effron?

  Well, Scott, this has been a fascinating conversation. I want to make sure to leave the last word to you. I'm sure folks will want to pull on all sorts of threads from our conversation. I know you, you have a bunch of places you can point people, but where, where is, is best for folks to learn more about you and the stuff you're doing at Adobe?

• 34:18Speaker 01

  My side hustle these days, that is both a accountability mechanism as well as a way to get me to connect the dots, share my thoughts with my own teams at Adobe, but also all the founders I work with and people that I respect is this monthly newsletter called Implications. And it's just implications.

• 34:34Jacob Effron?

  It's very good. Folks should sign up.

• 34:36Speaker 01

  It is, it is a really, I mean, gosh, what a grueling exercise. It's usually like at 30,000 feet when I'm just painfully trying to connect the dots of everything I'm learning in the rooms that I'm in and the conversations I'm having. And I try to be provocative, direct, and, you know, I'm not necessarily known for my brevity in this newsletter, so fair warning. But, but that's implications is, you know, is where I'm pushing myself these days intellectually.

• 35:01Jacob Effron?

  I love it. Do you use any AI in the writing of it?

• 35:03Speaker 01

  I don't. But I, you know, I, I try, I try to scrape my own intelligence for this one.

• 35:10Jacob Effron?

  Well, Scott, thank you so much. This has been an awesome conversation. I really, really appreciate coming on. Of course, Jacob. Anytime. Thanks for having me. That was unsupervised learning. Thanks again to Scott for a fascinating conversation. Definitely stay tuned. Next week, we'll have another amazing AI leader coming on the podcast. And if you enjoy the episode, please make sure to share with a friend.