Creativity vs Control: Where AI Fits in the Creative Toolbox
42:22 Share

The learning curve has always been this great obstacle to people entering the space. Resourcefulness is a far more powerful capability than resources. We don't know exactly when, we don't know exactly how, but we do know that the ultimate interface is like an inch away from our retina. My thesis for the next 10 years is going to be that taste will outperform skill. Being a creator in 2025 is tough.

But building for creators is even harder. Every creator is inherently creative, so their workflows are diverse, unpredictable, and highly personal. But they also want magical modern experiences. So few companies have been able to crack that code, although many are trying now that AI is opening the aperture.

Perhaps no better person understands this conundrum than our guest today, Scott Belsky. Scott was the founder of Behance, has now spent over a decade at Adobe after Behance was acquired, and is now their chief strategy officer, and has also advised companies like Pinterest and Atlassian. So how can AI enable creators? And where should that tool set end? What does an AI-native UI even look like? We're going from the prompt era to the controls era.

Plus, do people really want bespoke personalized experiences or do they desire familiarity? And what happens when something rare becomes ubiquitous? Listen in to find out. By the way, this podcast was recorded live in San Francisco during our fourth iteration of our A16Z Games Speedrun program. If you'd like to learn more about Speedrun, head on over to a16z.com slash games slash speedrun.

As a reminder, the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com slash disclosures.

Scott, I'm extremely excited for this conversation. You have worn so many hats over your career. Of course, you were the founder of Behance, which is the platform for creatives to discover and showcase content.

You sold that to Adobe, had a stint at Adobe. You then stepped into the venture and startup space where you invested in and advised many companies from Uber to Pinterest to Carta, Flexport, Rho. You found your way back to Adobe where you were chief product officer and now chief strategy officer. And in the midst of that, you've

written two books and you're a consummate student of creativity, technology, AI, frequent reader of implications. If you don't follow Scott's blog, it's a great read as well as a frequent commenter on Twitter. So with all of that, this is such a unique time and you are maybe one of the most interesting people to ask

questions about where we are headed in the future of creativity. And so maybe with that, I'll start with just at a high level, where do you think we are headed with AI and creative tools? How will it change the way that creators interact with the medium and think about creation?

Yeah, well, first of all, thanks for having me and thanks for the nice intro. I'm happy to like share perspectives and some lessons learned the hard way and whatever you want to cover. On the question about creative tools, listen, I mean, creativity and storytelling has been my passion forever. And I think that historically, I mean, I got Photoshop when I was 13 and it took me maybe a year to learn a quarter of it.

The uphill battle of learning how to leverage these tools, the learning curve has always been this great obstacle to people entering the space. And the skills required to use these tools are also a badge of honor for the people who have the skills.

And now you have this moment where two things are simultaneously happening. The floor is going down and suddenly more people can enter the world of storytelling, whether you're a salesperson or a data analyst or a marketing person, you can do your own thing as opposed to rely on others to tell your story.

And then at the other side of the spectrum, you have the ceiling going up, where suddenly if you used to be an imaging professional or an illustrator or a motion graphics artist, you're now also able to create in video, you're able to create thumbnails without having to go to a graphic designer to do it for you, and you're able to push your category to another level.

folks that are in the 3D space. I mean, the things that are happening now with 3D and animated 3D and 3D vectors. And I mean, it's just absolutely mind-blowing what it looks like to also raise the ceiling. You know, 10, 15 years ago, there was very little venture capital going into the creative space. There were only a few games in town, names in town, and it was extremely hard to break into those workflows. And now I think it's just

The world is going to be filled with humans trying to differentiate themselves through creative expression because they're competing with agents, right? And I think that that's sort of this broad awakening. So it's exciting. But the early era of AI and creativity was somewhat disappointing in my view.

It was very prompt-based. It was driven by these basic models. One core conviction I have is we're going from the prompt era to the controls era of AI-based creativity. And what I mean by that is you'll actually be able to generate assets with layers. Like you can think of Photoshop as layers. You can have a lot of levers to change things and get it exactly how it is in your mind's eye.

You'll have a lot more capabilities to reference certain ingredients, styles, structures, colors, color palettes, themes. And it's fun to see our design team is building this entirely new tool set for the controls era of AI. And a lot of you are probably thinking about that as well. And it's really exciting because everything goes out the window, like all the playbooks of the past.

this linear workflow and the daunting nature of it are all in the past. And then of course, like the agent-assisted flows are also a very core part of all that. Yeah.

Lots of different threads to pull and unpack there. But as someone who has been at the forefront of building creative tools and advising creative tooling company for so long, we have a lot of creative tool builders in here, a lot of creatives in here. Maybe on the tooling side, now with this AI wave, there are researchers and sort of technologists

some of which are creators themselves, some of which maybe aren't. But I'm curious, what are some of the misconceptions about building for creatives as someone who has done that their whole life? Well, I think that creatives, and I mean, we're all creative, but like creative professionals, they hold dear their ability to achieve what's exactly in their mind's eye.

And if you think about it, that's the antithesis of the notion of these generative AI models that essentially are trained on a discrete data set and always essentially output like the center of the distribution of the data that it was trained on. These are also creators who are oftentimes trying to stand out and do something different. And so in some ways they're frustrated with the average manifestation of their prompt, right? They want something different. Also, I think these creators want choice and models.

So one of the concerns I have about the companies that are model companies competing in this space is that if you're a model company, it's hard to offer all the models. One of our major decisions that we've made at Adobe over the last year is to actually become a platform agnostic model company. So we do build our own models in a way that's commercially safe and is important to a lot of our customers. But we're also going to support

all the major models and people are going to be able to choose which model they want for which action. And I imagine there will be probably thousands of models, some that specialize in animation or explosions or drone aerial shots or whatever the case might be. And customers, creatives, want to have their choice of what model and they want to be able to use it in their workflows. Creatives have

always complain about workflow, right? I mean, the whole notion of work and flow, like flow is the creative state where you can just do whatever's in your mind and work is what gets in the way of that. So if we can like remove more of the work from workflow, like that's the ultimate holy grail for most creatives. And I think that that means that a lot of these pieces need to be in the same tool oftentimes. Totally. Yeah.

And maybe building on that, where do you see Adobe in that evolution of how creatives interact now with AI? You know, a lot of folks think of some of these AI models as the camera. But of course, there's AI-assisted tools within some of the existing toolset. And so how do you think about where is the right place to expose AI in Adobe's current products? And are there any areas where you think a net new interface makes sense?

So we're doing both. We believe that all of our products, including the marketing products, not just the creative products, need to have AI and exactly the right time and place. So whether that means having an agent that critiques your design work,

or can critique your design work based on a certain persona that you're trying to target with your marketing. That's something that we're surfacing within products like Photoshop, as well as the marketing products. The ability to, of course, generative fill, generative expand, like all those sorts of things are integrated throughout the entire suite. But we're also building these net new native Gen AI products that are

really throwing out the playbook of the assumptions of layers and everything else, and really just starting from scratch. And we talked about publicly a project called Project Concept.

which is basically like mood boarding and concepting in the age of AI. So it can let you start with your work. It can let you source stuff from stock or wherever else. And then it lets you use, it has the Firefly models, but also has the Flux models. And you can just start to go crazy. And the cool part about it is that if you have

an image here and an image here, you can use a little eyedropper and take some ingredients and aspects of this image and then apply it to this image and then use that as a prompt for your next image. So it's really like a remix type of product as an example of a native Gen AI product that we're building. So we're looking at both. We're also trying to partner with a lot of folks in the ecosystem. I mean, workflows has always been

really important in terms of how creatives work. So historically, we've always worked with different sound products. We've always worked with different motion graphics products. And similarly, the products that are in the ecosystem, we want to like invite into the flow. I think that that ultimately makes our customers happy. Awesome. What creative problems do you think AI is unsuited to solve or help? Where does AI stop? This is like one of the greatest questions I think about all the time now.

because AI is only getting better. And yet, what makes creative content really effective is when creative content moves us in some way.

Now, what makes something move you? It's usually emotionally triggering. It's usually a story you didn't expect. It's usually something that has to do with the taste and the meaning that was injected into the content by the creator. And AI is not good at that. AI is good at doing things that have been done and applying things to formulas that have been...

declared as efficient formulas on the internet for telling stories. If you talk to great directors, I spend a lot of time these days really interested in the whole filmmaking and storytelling space, just personally, less Adobe, more personally. When I talk to these directors, they talk about how it's all about the timing, right? You can have one scene with actors, and if it's like one second too long, either the joke doesn't land or it's not emotional or it's not real.

And so, so much of that intuition is very much based on this director's personal view on the world. That is not something that I've seen to be replicable by AI.

And so I think that these tools need to enable humans to, again, get out of me more of the work part so they can do more of the flow part. And the flow part, if you could spend 10 times the amount of time you have every day on the flow part, you can do even more meaningful, more incredible stories. And if I were to sum up my declaration about my thesis for the next 10 years, it's going to be that taste will outperform skill. Because skill is going to be offloaded to compute. And what's going to differentiate everything you do in life? Taste.

Do you think there are any use cases where you could have 100% generated, fully personalized for Scott Belsky entertainment experiences where there's no human interfacing?

Well, first of all, let's start definitely for marketing purposes. So we have a product called Gen Studio now. You take a campaign that you made, right? And then you say, great, I want to mount this folder of 10,000 images. I want to employ this copy agent to make 10,000 slogans. And then I want to create this in 120 languages. And then you push the button. And then literally it creates like 160,000 assets.

that you can then drive across platforms and do A/B testing and automate this and whatever else. Like, that is the future. And when you think about all the production costs that is currently spent on stuff like that, it's just insanity, right? So there's no reason why that should be done manually. That should be done through AI.

And with the human labor you free up from that meticulous production work, you're going to come up with, instead of one campaign to test, you're going to come up with 100 campaign ideas. You're going to raise the ceiling of what's possible, right? Now, on the entertainment side, it's a little different. I happen to believe that part of the entertainment sensation is a shared experience. I can't talk about House of Cards with you if we saw a different House of Cards, right? And I will stop talking about it because I'll know that you saw a different one and it's not a shared experience.

I would anticipate a hybrid. Back in the Behance days, we always thought about building networks as like the 99 in one model. 90% of people consume, 9% sort of curate and 1% create.

And if you think about it, if you apply the same thing into this new world we're in, I imagine that 90% of people will just consume content and want to share experience. Maybe 9% of people will be like, oh my gosh, imagine if that episode ended this way and they might engage in the tools that you allow them to with character, IP, and whatever licensing we have to make a variation. Right.

or maybe the 1% are the people who do that and the 9% are the ones who curate and decide what's worth prime time for the 90% to see. So I think it's gonna be a similar model, but I just don't think that everyone will have a personalized movie. - Yeah, no, I agree completely. I think I could watch an endless supply of Star Wars content. - Streaming just in your eyeballs, right? - Right, but if I don't have a place and I don't have like podcasts, as much as I enjoy watching it, I also enjoy listening to podcasts and reading Twitter and Reddit, et cetera. And if I don't have that shared experience,

but you can sort of amplify that shared experience if I can remix the ending of Rogue One or you name it. I mean, one other thought experiment, which I'm sure people have thoughts about, but I've been thinking a lot about the history of what happens in periods of abundance for any sort of item. So when shoes become ubiquitous and commoditized, what do we end up doing? We end up seeking more scarce versions of shoes by buying higher price shoes that are branded.

And same thing for handbags and same thing for cars. Like when anything becomes commoditized, humans seek the scarce version and are willing to pay more.

And so what happens when content is commoditized? I mean, historically, it hasn't been commoditized. It was super expensive to make entertainment content. Now that it's not, will humans react in a similar way? Will we start to seek the more scarce, higher premium, higher taste version of the entertainment content or not? I think that's an interesting debate, but I personally would be bullish on the best storytellers in the world in a world where anyone can be a storyteller. Fascinating.

You talked about ad creatives and marketing use cases. From your purview at Adobe, you get to see what the entire media landscape or entire user base of Adobe is doing. What has been the general reactions across the spectrum towards AI? Like which use cases are really diving in, which are more hesitant and why?

Well, we're at a moment in time right now where there definitely are two factions on either side of the spectrum. On one side of the spectrum, there's a good swath of the creative professional community, especially illustrators and graphic designers that are extremely threatened by this technology, really worried about the implications, and have become basically anti. And I just think that you can never fight the advance of technology.

But what you can do is find ways to make it work for you. And we're involved with the conversation with that community. And I think that they'll start to tinker and recognize that there may be appropriate ways or ways that they're comfortable with leveraging it while also retaining their craft. On the other side, you have these kind of AI creative hackers who have basically discarded all their traditional tools and are like all in. And to them, it's like sacrilegious to use a product that is traditional. They just want to totally go pure AI.

And it's fun to see them work within those creative constraints because they are constraints. You know, it's really hard to do in AI. It's like really hard to make an ad and describe the text and get it right. It's so much easier just to literally take it into Photoshop and write the text, right? But you can try and you can spin your wheels.

There's that faction. But I don't know what the numbers are, but I would say like it's 10 and 10 and then 80 in the middle. I think the 80 in the middle are like, listen, I want to do the best work possible and I want to do the least mundane and repetitive work possible. And they're very much embracing these tools when it's appropriate. Hey there, it's Steph. If you're looking for another business podcast to add to your listening queue, then check out Masters in Business.

Each week, Masters in Business features wide-ranging, deep conversations with prominent investors, economists, and executives from all walks of life, like Greenlight Capital President David Einhorn and Morgan Stanley Investment Managing Partner Andrew Sliman. So join host and former Bloomberg Opinion columnist Barry Ritholtz to hear the stories behind some of the most successful names in finance. Follow Masters of Business on Apple Podcasts, Spotify, or wherever you're listening now.

I want to step back a little bit to Behance's founding experience. So it wasn't the traditional way that at least we think about starting companies. You bootstrapped for the first five years as you got the company off the ground and then later in the journey decided to raise venture funding. I'm curious, going back, would you have taken the same approach? How do you think of that in retrospect?

I think about this a lot only because I get asked this question and I'm like not sure if it was a best or worst practice. I mean, we nearly died five years bootstrapping. The fact that we didn't meant that we could raise money

Our first round was from Union Square Ventures leading five years into the business when we were breakeven. So we actually only had to sell a very small portion, relatively speaking. And then when we got acquired, that played a real factor because our acquisition, apples to apples, I was like, oh gosh, like it's actually the same outcome as if we took two more rounds of funding and that much more dilution and that much more market risk. So I definitely played a role, a beneficial role. At the same time,

There were some sideways lost years of Behance in that five years. There were a couple of years where we were not able to afford the talent we should have been hiring. We were not able to level up the team the way we should have. And I was, for some reason, very proud of being a bootstrapped company and having control over my own destiny and that sort of thing. It was a stupid ego thing because when my team started coming to me saying,

I literally can't do the best work of my life here because I don't have the resources to invest in this opportunity. When we started to grow and we couldn't scale, I felt like an idiot. I was like, oh my gosh, why am I depriving this team? And that's when we went and raised our round. So the one thing I would say is this, though. Resourcefulness is a far more powerful capability than resources.

Resources, when you have them, you throw them at problems like carbs. It feels good, but dissipates quickly. Whereas resourcefulness is like muscle. When you develop the capability to address challenges in the hard way, that sticks with you over time. And I think that capability developed during that five years was valuable to us.

Yeah, you mentioned the messy middle. And of course, you wrote a book on this. And so maybe unpacking that part of the Behance journey, when you faced the initial spark of excitement, and then a little bit of plateau, how were you able to keep the team motivated? How are you able to make the hard decisions about what features to deprioritize?

I'm sure there were probably, you know, push and pull signals to potentially pivot some that you had to navigate. What are some of the lessons for the folks in the room who maybe had an initial spark, but are now sort of questioning whether that's enough to call them to greatness?

The Messy Middle project was inspired by the fact that we don't talk about these questions. As a tech community, we typically focus on the romanticism of the start and then these like glorious finishes, whether they're gloriously great or bad. That's the soundbite. Now, of course, there is a lot of really messy stuff that happens in the middle that all of you are also probably in the middle of. When it comes to this question, though, of when you should quit or when you should stick with it,

To me and to my friends who've gone through this as well, it's all about this measure of conviction based on what you now know. When you start and you're starting from a place of ignorance, you're imagining this great product that solves this huge problem and everything goes up and to the right.

And that's a very easy period of the journey to have a ton of conviction in. And that's actually what helps us quit our jobs and unplug from the reward system of society and jump into starting a new business with tremendous amounts of risk. When you start to get shoulder to shoulder with customers, I think we all end up realizing to some degree that the thing we had in mind is at least 30 degrees off of product market fit of what the customers actually want.

Why? Well, oftentimes customers do things in very strange ways. Like they won't use an enterprise product, even if it makes their life better, because they won't get as much credit with their boss. And so they'll do it the old way so that they look like they're doing the work that merits the credit. It's like weird, silly things like this that actually are the difference between utilization and not utilization of a new product.

But no entrepreneur in the beginning has this belief that, oh, we have to stroke people's egos in the enterprise to get this product to be utilized. Like, we're all more rational and we're like, oh, yeah, if it's a better product, people will just use it. So there's a lot of things like that as this one silly example that we learn in the process of shoulder-to-shoulder-to-customer that help us lose conviction.

And as we're losing this conviction, it's always the question of, oh my goodness, what do we do? Or sometimes we actually gain conviction, but the product in our mind keeps changing based on what we're learning. So when you're at that moment where you're like, okay, should I pull the plug and try something different? I mean, there's, by the way, there's no pride in sticking with something that's not working. That's just silly. It's a waste of time. So I always ask founders like this conviction test. I'm like, all right, knowing everything you know right now,

Either do you believe you just have the wrong solution and you have a good sense of what it is, but you still have conviction in that long-term view that this industry will change as a result of what you're doing? Or have you literally just lost conviction in the problem in and of itself? It's a false premise or it's the wrong problem or it's the total wrong solution space. And I'm like, hey, if you've lost conviction based on what you've learned, quit. Try something different. Otherwise, stick with it because that's just par for the course.

So at the end of the year, you wrote some predictions about where you thought tech and AI were headed into 2025 and beyond. And I want to pull a few of those out and talk to you about them. One of them was a prediction towards more homegrown or DIY software, both on the consumer side and the enterprise side. So like consumer side, if family game night and create a game to play with your friends or your family or enterprise software, who's deciding that custom software is

built with AI tools is actually more efficient and more helpful than some of the third party tools. For the companies in the room, both consumer companies and A2B tooling companies, how do you think they should reorient themselves around that evolution? Well, it's interesting. Since I wrote that, I got a lot of feedback and my thinking has evolved a little bit.

I think right now all the rage is DIY software. Replit, cursor, whatever, like being able to go in and say, I want this. And then it like pumps out this kind of half-baked app, right, that you could use for certain purposes, but still requires a lot of final mile work to get it to really work. And that could be a threat of all of us building software if everyone can just build their own software based on the code that's on the internet or whatever.

I'm actually more interested these days in what I've started to call conformative software. And what I mean by that is it's software that actually gets to know you and meets you where you are. If you think about it, software has always been generalized as much as possible for everyone. If you want a big TAM, you make the lowest common denominator app that everyone can use.

And then you end up listening to requests from power users. You need to monetize. You need to add more features and capabilities. You need to make it enterprise grade. And before you know it, you've made it complicated. And then some new startup supplants you and everyone flocks to the simple product. And then the process repeats itself again and again and again. And that's what Silicon Valley is, right? I think it's interesting that this next generation of software may actually conform to each of us. So we may come on and the onboarding for you and me might be different.

the tool set might be different. The purpose that you come into the product with and I come into the product with may dictate an entirely different experience. And that could be an agent-driven experience, but it could also be a new variant of UI that is driven by AI, right? So the UI actually conforms to the needs of the user and everyone has their own sort of tailor-made bespoke experience of this one piece of software that people are using.

I think that's a much more likely future. And it's not like one SaaS company is going to enable that for people. Everyone's going to be able to do it because we're all going to have customer data to leverage for personalization. We're all going to have these agents and LLMs that are personalizing the experience. And I think designers are starting to crack these AI-driven personalized UIs right now. I think that's a huge part of the future. So I guess my thought for all of you is,

is since you're building products that are going to be thriving in that era, think about your different personas and think about what different accommodating or conforming experiences you could offer that might make different personas thrive in your tools. It's a real TAM expander. It's interesting to think about the idea of generative UI and back to the conversation we were having before about generative

creatives and what does that mean for a designer, a UI, UX designer, product designer in terms of having the guardrails around that within your app ecosystem? Yeah, it's going to be pretty wild because the UIs that we put out there now, oftentimes as design teams, we're like very pure about

But as for those of us that use Amazon a lot, it's like Amazon is not a beautiful UI. It's just like super damn effective. And so you can imagine certain products be open where there's like going to be a huge red button. There's going to be three things flashing. We'd be like, oh, my gosh, this is the ugliest thing I've ever seen. Except I know exactly how to use it.

And of course, voice is like one of the areas that a lot of people are. Yeah, I'm curious if you have any thoughts in terms of how voice may impact some of our consumer experiences in a more pronounced way over the next couple of years. Yeah, I'm excited about voice for the things that we know we want, but I am not excited about voice for any form of discovery in products that we need.

Most product experiences and e-commerce experiences are so discovery driven. So it's really hard to have discovery in voice, right? That's always been my concern there. I do think that voice is a very disruptive interface because it obfuscates choice away from the consumer, which some companies could use. I mean, if you can imagine if we're all going to be wearing these AI enabled headphones and you say, "I want a car." And it's like, "Great, I will have a car downstairs for you."

You're not even going to be able to choose between Waymo, Uber, or Lyft anymore. It's going to be a third-party marketplace of APIs competing to satisfy your request. So it effectively removes the brand power, which is a very disruptive thing.

That actually this sort of like ambient computing idea dovetails nicely into one of your other predictions for 2025 in an area that's near and dear to my heart, talking about AR. And you said that the legs for AR are now beginning to take form in some of the work that folks are

at Adobe are doing on the software side with Aero and Project Neo, and then of course on the headset side with Orion and Apple Vision Pro. And you predict that we're sort of on the precipice of a big wave towards AR over the next three years, much like what we're seeing in AI. And I'd love to hear you unpack that a little bit more. And then if there's any kind of use cases for AR, maybe also coupled with AI that you think are really compelling.

Well, it's one of those sometimes in your career, like patterns hit you hard. And for me, the pattern that's hitting me hard right now is that I remember in 2015, 2016, the conversation around AI and its relevance and how transformative it will be in 2015 and 2016 is very similar to the conversation right now about AR and immersive experiences, in my opinion.

We don't know exactly when, we don't know exactly how, but we do know that the ultimate interface is like an inch away from our retina. And it's just so clear that's going to be so transformative. And back then, people always debated, is it next year? Is it three years from now? A lot of the startups that I saw back in that time in the AI world failed because they were too early or they missed generative AI. So it's an interesting moment.

But I would argue that in five to seven years, AR immersive is going to have all the buzz of AI today. I believe that at that moment when that happens, I think that context-aware OSs, like location-based OSs where AI

The stuff that I need to know right now, based on the fact that I'm with you, that you're you, that I'm me, and I'm at a 16Z. Those are the context pieces that inform my interface right now, right? And it's fascinating to think that's the OS in some ways, is that context. And the context will probably be informed by AI. So I'm really excited. With my Adobe hat on, what we're trying to do is build the best in class 3D interface

creative tools, both for video as well as for imagery, because we believe that people are going to want to contribute in that space. And as soon as it becomes mainstream, like every company in the world is going to want to have a really high-end, photorealistic 3D environment. So I think it's going to be big, but...

It's a few years off for sure. What is maybe a contrarian view you have about AI that you're surprised is a contrarian view in the landscape of the conversation? I don't know how contrarian this is, but I think that most models will be commoditized, number one.

Just, I mean, what is it, ChatGPT 3.5 is now like 98% cheaper than it was when it launched and 4.0 is already like 90% cheaper. I mean, it's just, it's so dizzying. As a result of that, I believe that these model companies that are currently powering all these products are going to have to go up the stack to capture the value in the business.

And as a result of that, much like the flashlight app was a profitable application until Apple shipped the flashlight, right? I mean, I think that similarly, there's going to be this large swath of applications that are wrapper apps, whatever, that are going to be just absolutely blown to pieces. But even developer applications and products, I mean, they have to go up the stack.

And if you've raised billions of dollars to build these fundamental models that are ultimately commoditized, AWS did that. I mean, they went up the stack and started to build developer tools and new capabilities. And that was how they captured the margin of the business. So I do think that that's the case. Maybe that's not completely contrarian, but there are still tons of excitement about these model companies. And I just think they're going to be ubiquitous. I guess another thing I would just quickly say is also how much is going to happen on-device.

And if you think about it, the on-device models are also becoming more powerful. The distance between the frontier models in the cloud and the on-device models for a lot of categories of usage is actually narrowing. And so at some point, if 90-something percent of all the stuff that people want to do in your products is routed to a local model...

then you're not paying any more COGS. These models are not making any more money, the big ones. So I also think people are underestimating the speed of local model meeting the bar.

In your analogy of flashlight and iPhone, you are sitting at the iPhone company in this creative landscape. And so I'm curious for a room of folks building in and around the creative space, if you were then wearing your investor hat or builder hat, where do you think there are good fertile areas to build and where would you avoid incumbents?

Well, I think that the two areas that I would be most interested in, which are actually maybe similar to the answer before AI, but one is network effects. Now, network effects also means data in the world of AI applications. So having data of what your customer base, how they're using a product, like what they need, how to make it better,

But I also mean it from the standpoint of individuals. If you understand me better than any other third-party product, I'm going to stick with yours. If you understand me across platform in a way that other products can't because they're only one platform, you're going to have me as your customer. I'm a big proponent of this notion in the consumer world of preferences being the ultimate moat.

I think that products are going to want to know who we are. They're going to want to be accommodative, as we were talking about earlier. They're going to want to know my preferences. And if I have an application, for example, that knows that I'm a vegetarian, that doesn't like mushrooms, and I love having a tequila on the rocks with an orange slice, and I like tea after dinner and all these different things, and every reservation I make becomes a more personalized experience, I'm not going to take the time to get another app to learn that about me. I'm going to stick with the app that knows me.

So the data layer and having an advantage there and developing a network effect around that data to some extent as well, I think is number one. And then I think our number two is the interface layer game. So my other theory about a lot of our world is that it's ultimately a game of slap a hand and the interface at the top wins. And so everyone's always trying to have like the interface at the top. So OSs have a huge advantage here because that's the ultimate interface that will always be at the top of every other interface, right?

But within categories, if you're the interface that is at the top of the stack, you're just going to have all the pricing power. You're going to be able to decide the defaults when someone wants a car, all those sorts of things. So I do think that in order to succeed in this world, you have to be owning an interface. And that's another thing I look for when I'm looking at companies. Is there some interface that they're truly owning or are they just going to be like subjugated by another one?

Shifting in that direction, you're both investing and advising founders. What are some behaviors that you've detected pattern matched across the most successful founders who you know and who you've worked with? Real quickly, rapid fire here, but sending rapid updates, like test flight updates and that sort of thing. You know, Ben Silverman at Pinterest did that, Kayvon at Periscope, like every week.

always getting it out there. There's something about holding your feet to the fire and just constantly doing that as opposed to founders that obfuscated what was happening away from their community investors. Very big positive signal there. Another one that I've learned is I think every founder has to be a great storyteller, but they shouldn't be a promoter. And the biggest investing mistakes I've probably made is

We're not getting that red flag of someone who is making something look better than it was. Because when you're investing in someone, you want someone who faces the bad truth and speaks it to their team, to their board. Those are the people who expedite solutioning and those who try to make everything look better than it is and present data with a bow around it and stuff. Those are the folks who will continue to do that and will obfuscate the truth away from their own teams.

So I've become very allergic to the promotey stuff. And I try to look for very pragmatic storytellers. I try to be a leader who's both optimistic about the future, but very pragmatic and pessimistic about the present. I feel like I like also investing in people who are that way as well. Are there any apps as you think about AI that you're hoping to have your hands on tomorrow? What's the thing that you're not seeing in the market? One of the areas is definitely on health.

It's just amazing to me. We're going to look back at medicine until 2025. It's going to be the equivalent of like bloodletting. We're going to be like, wait a second. So you went into a doctor who gave you like five to 15 minutes and then you described your symptoms and he took a glance or she took a glance at the chart and then somehow in one minute decided with a very small data set of you and

some other random stuff in the person's head, what your situation is, that is just absolutely insanity. It's overdue. I mean, doctors should all be augmented by AI. And also our personal relationship with medicine should be a very forever context window type of experience. Another thing, by the way, is management.

The whole world promotes people that are great at what they're doing into something that they're not doing well. And most managers suffer. And most managers just want to focus more on the things that matter in production. And they want to just help people succeed. Managers should all be augmented by AI. And the idea of driving as a manager without AI will become sacrilegious.

Even though it's sacrilegious now to say that AI would manage us, it will be sacrilegious to have AI not manage us in the future, to some degree. And as a manager, it would be amazing to be informed of everything my people have done and what they're struggling with and where they are in their OKRs and everything magically summarized. I'd be more effective as a manager if I could have that.

I think the last thing I would just say is the whole agency world. A lot of the stack of production, like we were talking about earlier with this product, Gen Studio, there's just so much antiquated stuff that's still happening in the marketing and content production stuff where personally, I would just expect us all to have that in our pocket at some point.

You've written a lot about kind of the future of work. And that's the question on everyone's mind these days with AGI looming is like what happens? How do people fill their time? There's a lot of folks who talk about creative endeavors, taking up more of people's time, both professionally and personally. Is that how you see the wave shifting? Where do you think 20 years time spend our time?

I'll try something quickly on you, which is quickly share what I think a company will look like and then quickly share what I think everyone else will be doing. A company, I've called them cognicos, and the idea is a cognition-driven stack. Basically, most of the business and the functions of the business ultimately running off of inference, right, and core models that are optimized for these functions. So if you can imagine the future company is mostly compute,

And then it has nodes that are the new version of functions. So there's like a node called HR and a node called marketing and a node called legal. Those nodes are probably running software like the stuff you're going to be building or you're building already. And the people that are manning those nodes, so to speak, those are stewards.

Of the node, they're supposed to make sure that the node achieves OKRs and also follows the rules. Rules is going to be a big part of the AI world. Everything's going to be about rules. And so the stewards manage the OKRs and the rules for the node and the technology underneath the node.

and they're orchestration designers and orchestration engineers. And these orchestration designers and engineers are like basically taking all these different APIs and functional products of the node and they're like wiring them up and making sure it uses inference efficiently and whatever else. So I call them the Cognico 'cause it's almost like this cognition driven company that is a whole different stack and a whole different role of people, right?

Well, then the question is if those companies require far fewer people, which I do believe that they ultimately will, what are people going to be doing? And so here I will look to Japan. Like I've always felt like Japan is like a little glimpse of the future. Before the iPhone, if you wanted like futuristic phones and devices, you would go to Japan and it was always like the future, right? I think similarly today, when you go to Japan, they've been living in the future, right?

Everything's like a small micro business. When you go out to eat, it's like a place with 12 or eight tables and the bar, it's like 40 bars on one street and each one has six chairs. And it's just this deeply craft-oriented, storytelling, very human, non-scalable experience at scale to some degree.

And so I do wonder, because if humans are going to be craving the scarce non-scalable version of everything, right? Maybe humans will respond to that. Like maybe there'll be a boom, 100X of small businesses and it certainly won't have to do accounting anymore 'cause that will be done by AI so they can actually focus on the craft. And I think that that will be like really exciting and a fun life to live. Like I love going to Tokyo. - Awesome.

All right, that is all for today. If you did make it this far, first of all, thank you. We put a lot of thought into each of these episodes, whether it's guests, the calendar Tetris, the cycles with our amazing editor, Tommy, until the music is just right. So if you'd like what we've put together, consider dropping us a line at ratethispodcast.com slash A16Z. And let us know what your favorite episode is.