Latest / The Interline Podcast / Is AI Already Good Enough For Fashion?
Transcript
- speaker-0: Hey, welcome back to the Enterline Podcast. Is AI, quote unquote, good enough? That might just be the central question of the current moment. And if you go looking, you can find evidence to support both answers. I'll put our evidence first because it's the data I have closest to hand. We surveyed about a hundred of them earlier this year, and fashion professionals are broadly optimistic about AI. But in the here and now, They self report that they don't trust it with critical decisions, they don't have a good rubric for assessing its value, they fear that their companies are overestimating its capabilities versus what it can actually deliver, and they consequently haven't connected it to their wider tech estates, which, you know, good enough? Maybe. But the size of the market and the investment statistics that the brands themselves are disclosing to us suggest that the available applications clearly are good enough to invest in. Nine in ten professionals use AI every day at work in some capacity. Nearly 85% of companies have accelerated their adoption of AI in the last 12 months. And almost 70% expect their spending on it to increase as we go into 2027. And you can also just, you know, look at the huge landscape for AI tools for fashion that exist. Both as additional capabilities that have been bolted onto existing solutions and as pivots and as whole new platforms being built with AI as their foundations. There's a lot of all of those. We interviewed some of the executives behind all those types of tools in the same AI report that that survey data I just mentioned has taken from. And I think it's fair to say that they all see a pretty unprecedented market opportunity in front of them. And I think plenty of their existing customers and their new prospects see the same thing. Or at least they think they do. So I'm not about to debate the existence of a market for AI solutions in fashion today. We've actually kind of done that earlier in the summer when I interviewed Ara Carazian from Ramp about AI adoption and economics. And the short answer is that that market very clearly does exist. But good enough is subjective. If you've been listening to these weekly shows for a while, you'll be familiar with the idea of the jagged edge of AI. The principle that a single model or a family of models can at once outperform people in incredibly deep, complex domains like higher mathematics, but then periodically declare that it isn't connected to your Google Calendar when it is, and then you remind it and it goes, yeah, I totally am. Those are two very glib and extreme examples. So let me give you a couple of more nuanced and more practical ones. It can be simultaneously true that. AI is very good at generating images of fashion products, and also true that it might be terrible at understanding how those products are made or what they're intended to do. We see this discrepancy in our data. The fashion professionals who took part in our AI survey told us that they perceive AI to be the most mature in analytical use cases, i.e., planning, market analysis, competitive benchmarking, and so on, and in concept generation. And then all the way downstream in visualization, campaign photography, and product detail page gallery content. They then perceive AI to be the least mature the closer it gets to real product development and to specialist disciplines like pattern making. That spread of opinion seems right to me. And it's a reminder that there's really no single monolithic AI that you can reasonably grade on aggregate. And A platform that's good applying AI to one part of the product journey isn't necessarily going to be as good at applying it elsewhere. But put a pin in that for a second, because are we or are we not seeing plenty of generative AI companies testing new features that are clearly pretty far outside their established lanes? We are. If you listen to my interview with Weber Wong, the CEO of Flora, a few weeks ago, you'll heard him talking about why a generative workspace can and should be the place. To build features that cross over into the territories that have traditionally been occupied by product development tools and PLM. And if you do a quick canvas at the wider market, you'll see the same thing from other Gen AI companies in fashion, most of whom seem to be coalescing around the idea that you can go from generating images of products, the most mature use case per our survey data, to building tech packs, all within the same suite of software, and that latter one being one of the least. Mature use cases according to our data. Now, I don't know how far I buy that, really, but I do know that I shouldn't just dismiss the idea outright. So today, I'm bringing on the CTO of a company that's squarely in the middle of everything I just talked about. That company builds what they call an AI-powered design suite for fashion. That company surveyed their user base this September and they reported, and I quote, fashion has stopped debating whether AI is good enough. Amongst teams using it daily inside live collections, that question is settled. End quote. Which, you know, okay, we're gonna have to talk about that. They're also now building tech pack and sewing pattern capabilities as beta features, or beta for our American listeners. That company is the fabricant. You'll be familiar with them. And if you listen to our MailRoom episode recently, you'll also heard one of the co-founders tackling a question about starting a brand in the age of AI. But today I wanted to get more into the weeds. So my guest is Marco Marchesi, who is the fabricant's CTO. I figure one of the best ways to get another angle on whether AI is good enough and what that question actually means is to talk to someone building for it and building with it for fashion. So let's have that conversation. Okay, Marco Marchese, welcome to the Interline Podcast. speaker-1: Hello, thank you for inviting me. speaker-0: Pleasure. looking forward to this one. I've spoken to a few people from the fabricant in the past. You and I have never really directly spoken. So this is it's a conversation I've been looking forward to. And we might as well start by defining what you do there. So we run every one of these shows with two things up front. We ask the guest about what their day-to-day life, working life looks like, and we ask them to define something. I'm curious about your day because in the time I've been running the interline, so six and a half years. The fabricants been through a few different iterations, geises. and they've all been pegged to different technology waves. simplifying this a little bit, but you've been a digital fashion blockchain company at one point, which I think is the first the first iteration that most people know about. You were a 3D kind of consumer co-creation company at one point. You're now an AI tools company. As the CTO, you've presumably been pretty influential in each of those directions because they align so strongly with tech. So I want to understand what you do when you sit down at your laptop in the morning and how you determine which technology waves to go after. And also how your work right now differs from what it did in twenty eighteen or twenty twenty two. speaker-1: Okay, let me go in order. The first one is the iteration of the Fabricants. The first iteration actually I was just a spectator of that because I was actually a big fan of the Fabricants. It was when they started I think 2018, like doing 3D, 3D photorealistic animations. of digital garments. I was so impressed by that. I told my previous company, guys, this is the future. I want to do this. We should do that. And I said, no, cares about digital fashion. no more. will never happen. And then in 21, we actually had the opportunity to join the team. And the idea was for the second iteration to build a digital marketplace for garments, for digital garments and that's set to co-internation. And then from there we actually over-impose a bit of still doing the traditional 3D and the effects. But then we decided to pivot into the AI. This one because my background in AI, they was kind of my pioneer AI for creativity for almost 10 years. And we have a lot of support from Nvidia and Google and everyone was telling us guys you really need to pivot, you should really pivot into that because that's the right time, you have the right skill set and we decided to go into the AI thing. So it's almost three years and let's say that we are working on that. Going back to the second question about myself. So tech... I've been in tech and creativity for, let's say ages, without quantifying too much. it's a job that needs to be reinvented every single day. literally there is, yeah, we can predict a lot what's happening, what's gonna happen, but at the end of the day, a lot comes from what users want and on the fact that there is a lot of unpredictability. on the new tech, no one will ever predict how much AI will have impact into the creative industry in the last couple, two, three years, let's say. And this one I said, because when I started doing work for fashion using AI 10 years ago, yeah, we knew that nothing was ready for the industry, by the end of the day, say, yeah. something will come. I remember, I think it was eight years ago, during a panel, I predict to the crowd, said in 10 years, and I said eight years ago, okay. So in 10 years, we're gonna see movies completely AI-generated at the theaters. And I think I got to almost basically right, right? And in terms of timing. But I will have no idea that we will arrive at this level of craziness and quality. speaker-0: I think I think on that one you're I think on that one you're right on the technology capability wave. I I think maybe time will tell if you're right on the does anybody want to see an AI generated movie in a theater. I know. That would be Okay. speaker-1: A lot. I can tell you I'm the big supporter of AI, also the biggest critical on that, on the sales side. Because, and that one, because my, I don't know if you know about much about my past, but I used to produce music videos for 10 years. So I was on the other side of the field, as a creative side. And I totally see. the pain of giving up on the creative process to AI, not just because of the quality but also the storytelling, the whole pipeline. There are many things that we can debate on. speaker-0: And I think we we may we may get into some of those as we go. I think the interesting thing you said to me there, amongst amongst a few interesting things, is the idea that the CTO role is an constantly rewrite one that constantly needs to be rewritten. I think I think that's true and I think it's become especially true over time. I think if you were a CTO fifteen years ago, or t maybe maybe twenty, you would be like, the cloud is a very clear direction and we have a decade to execute on it's something along those lines. Like you you could see the way that things were going. You could see that desktop applications were on the way out and that the web was to become the distribution layer. I think a lot of that you had time to react. I think the difference now is you you don't have a lot of time to react to how quickly the role and how quickly technology is being rewritten because of AI, right? speaker-1: Yeah, but the risk of course, the treatise is about focus. Because the moment that you want, decide to react too quickly on everything, you're going to lose focus. And the focus will be on the product, on what you want to really build for users. So the big challenge for me is not anymore the technical part of that. let's say they're not the trivial code because... Honestly, I think people in general think that coding as something, or software in general, something really esoteric, where you need a lot of preparation, a lot of high skill set. But it's true for one part of things, especially when you want to do R &D and real innovative things. But 90 % of what code does is very trivial and repetitive. So it's something that I was expecting. I was really hoping, looking forward to AI to solve that out of my debt. So that part is solved. let's say solving the sense that is completely changing the way that we have to deal with. I mean, if I can compare even six months ago or two years ago when I started using AI agents to support my job, my day is completely different now. In some ways, way more relaxed because I know that I can solve things super easily, things that were trivial but still need time. And now I can solve very easily. At the same time, there are other things that they open, for example, the cybersecurity risk, the testing that you can never assume that something is true. You always need to. think about, okay, there is some percentage of hallucinations that probably I bring on top of everything. And so the job is different for sure. And the CTO, yes, I think is one of the most affected roles in that sense, because things are moving very fast. speaker-0: Yeah, I think I think that's right. and I think if I look at my own work, I don't use AI to write, I do use it for a lot of other stuff. I think my my working life looks very different to what it looked like six months ago, certainly a year ago, just in terms of where I'm doing the work. And that does bring me to the second question, the thing I wanted to ask you to define. So I interviewed Adriana, who's your CMO and one of the company co-founders on our first Mailroom episode recently. She fielded a listener question about what it means to launch a brand and what it means to be an AI native company. she pretty well covered that from the brand side of things. So if I want to design and sell clothing, what it means to be AI native there. The thing I want you to try and define is what it means to be AI native from a technology point of view, both in terms of what an AI native brand, so a brand starting right now, would have as their technology stack, and also how you as a technology provider Are changing the way you work and your setup to develop solutions. Because you know, to go back to our thing about the difference between AI and cloud, for example, I think everybody knows what the e-commerce tech stack is and was, even if they can't name all the individual parts. But there's so many different solutions that exist, huge diversity of AI models and things that they sit on top of right now. I think it's much harder at this minute in time for one brand to understand what another one might be doing technology-wise, and the same for software development companies to understand what the others are doing. So tell me what you think it means to be AI native as a brand. What does your tech stack look like? And what does it mean to be an AI native technology developer? speaker-1: Let's say I have a very different opinion on that. Mostly because I don't believe in the concept of AI native. I tried to explain. don't think the concept of AI native for me, it's self, it contains a hype element. means like, what does it mean? AI native means you bring some technological step as part of your background. But what exactly means for me is almost nothing because it like, I don't know, I kind of compare when people in the CV, they put like, I know how to use Word, I know how to use Excel, but they don't say exactly what, okay, but what you can do with that. And it's kind of, I have a very kind of critical thinking about that because I feel like when you are a brand, should know that the world is going into some directions, but not necessarily you need to make sure that you are native on that for be able to embrace and and get the most of that. And that's why I prefer to see ourselves as the people that they are helping the brands to embrace AI rather than, yeah, you need to be AI native to come and understand what we're doing. I think that AI native for me, the best replacement of say AI native is just you are a curious mind. So you are happy to try. different ways to be creative, but not necessarily you are someone that is adopting AI itself because people are using that. I don't want to be, I think for me it's more important that the brands comes with their own strong opinions and stronger identity of what they want to build in terms of as a storytelling, as a taste, a brand identity itself as in general. And then they ask, okay, what AI can do for us? And that's where we are happy to help with the tools that we have or building new tools, they're going to help us. Because I will say, I kind of always been, because I'm an age in the past, artist of all different nature, like VFX, graphic design, motion design, all things. And I always find very difficult for me when the artists are really attached to the tool. know, I really feel like the tool makes them creative and not like they are creative. So when you still think about AI native, I still have the same kind of opinion. I still like and I still think that it's not such a thing as AI native. I think it's more... Okay, you are curious and you see there is something new that needs to be used because it actually can give you benefit. that's it. But that's it. Nothing more than that. speaker-0: Fine. So I think I see things a little bit differently in that so being an AI native company to me, as a brand or as a technology developer, would mean your the way you work and the offer you have couldn't exist without the underpinning technology. That's what I consider to be native. And if the the the example that I would go to would be something like Facebook could not have existed without occurring at the time that it did and taking advantage of the the like the lamp stack, so Linux, Apache, MySQL, and so on. But that's that to me is what it means to be AI native. And I think there are e-comm native brands in the same way. So you have like the the the whole wave of direct-to-consumer. Yeah, a bunch of those really didn't pan out very well. But you had brands that could not have existed without the e-commerce stack. They they that was a precondition of their entire business model. I to me, that's what AI native means, is that there will be brands who their entire business model requires AI. AI is a precondition of it. And I you I that could mean AI from a creative point of view. It could mean they sell through AI only channels. It could mean a bunch of different things. It's not a very fixed form, but that would be my definition of AI native, I think. speaker-1: It's good. Fair enough. speaker-0: Yeah. Okay. and I think I don't think it I don't think that does anything for the creative side of stuff necessarily. I think you're right. I think being curious and having your own brand and applying the tools to it afterwards is is definitely a way to think about it. And if you're a pre existing brand, that's the right way to frame it, I would say. we talked about developing software. I want to quiz you a bit on how much of the fabricants' business you would actually describe as software at all. So over the course of the different eras the company's been through, you mentioned you work on the sidelines for the first one, but then involved in the ones thereafter. there's been one thing that's consistent to me as an outside observer, and that's that I've never been quite able to figure out how far you were a tech company and how far you were a service company. I certainly felt that way about a lot of companies when digital fashion was all the rage because and I'm not saying this is the fabricant necessarily, but there was a lot of kind of misrepresenting the digital fashion as a turnkey technology when it was actually a pretty bespoke service. AI is definitely not that. You can feel however you want about where you use AI, but you can't deny that the LLMs are real, image generation models are real, and so on. You can go use them. They are turnkey technologies. What's the balance of your business right now between building tools and software on top of models and on the other hand, offering training and hands-on work and that bespoke side of things? How does it shake out between software and service? speaker-1: The thing is that the moment that you are able to automate, accelerate the software development, the importance of software itself is less than before. And it's way more important to push and test the product that you're building on top of the software. Of course, software and product are two different things. So what I say is, and if you notice most of the companies they are in the AI field and they are using AI for their business, actually forward deploying their employees in order to accelerate the adoption of AI and their service. So basically I will say that everything is becoming a service in that sense. So every single one, OpenAI, Anthropic, Microsoft, they are all adopting the Palantir model for forward-to-pre-engineering because that's the way to really be able to introduce these new technologies to people. And at the same time, we have to do the same for brands, of course, because we don't want that brands spend time on trying to figure out something. where we can expertise the technical expertise and we have the opportunities to move quickly by customizing tools and making the tools they are tailored completely for their needs. So that's where we are service. I think we are, let me tell you, I think we are in the industry of listening a lot. it's like the fabricant, most of the things that we do, is listening, is like literally, it's not building software, it's not is testing, but is mostly about listening to people and understanding what they are, what are their pain on their, on what they are doing, on their daily fashion process where we have many different challenges, they are still needs to be solved. tedious task. tell you, just an example, we are working with the UAL for a sim pattern tool, automating pattern making. And one of the things that impresses me, I love to talk about that in general, is like we have, I notice how many clicks, button clicks, mouse clicks they do for doing their task when they do pattern making on digital tools. And for me that was, okay, that's where you need not just to build software, but you need to listen to them and so, listen to them and see them in action. And that's most of what we do. because the software part is becoming where you can run agents that can build all the result of hours and hours listening to people. But that part is still very human related because it's difficult to automate for sure. You may say it's service. Let's call it service in a very modern way. It's not like anymore. speaker-0: I think that's right. I I agree. I think that I think software service d that delineation is a very consultancy mindset. And you know, I spent some time in advisory, so that's that's probably a legacy of that. If we extrapolate from that then, and if we say, Okay, you're listening to somebody, you're listening to a brand, you're listening to a creative person, and you're coming away and building something that, to pick your example, reduces the number of clicks or that changes the way that they work. As a CTO. How are you working backwards from that to then figuring out what you need to build as a as a tool provider and what is already covered by the frontier commodity models that exist from the major AI labs? And tied to that, how do you think about keeping the stuff you're building ahead of the evolving frontier? of model capabilities, because it's certainly not a technology that's standing still. So you listen to somebody, you translate that listening into a solution that improves their day-to-day work. How much of that solution are you building and how much of that solution is architected on top of commodity models? speaker-1: If you have two choices, if you have a problem to solve, and you want to solve in the most trivial and standard way, you just tell the problem to the Halaland and it gives you the most trivial solution. Because he learned on billions, trillions, let's say, of examples and most of the knowledge that he made was on top of standard. solutions and that's how to make it just right. so if you want to, if you have, you need to build a, I don't know, the UI, the interface for a specific task, you just tell, okay, this is the task, you make a UI. It's going to make a UI that is perfectly tailored for that, but it's going to be the most standard one because it's the most obvious. Okay. This is not an off, of course, and that's where we build on top. So usually I can start from there if I want, if I want to go quickly. Five minutes, you got a prototype working and that's it. But then from there, there are many times where I say, okay, that's a good starting point, but it's wrong this, it's wrong that. And that's where I put my, let's say my product person, myself as a product person on. I started chatting with the team and see, okay, how designers work, how pattern makers work on that, because it's not going to be like that. And you realize, yes, it's not that standard way because designers have other ways, other conventions, how they want to deal with this interaction. And that's where we build on top of foundation model. Of course, and this is one thing, then of course there is all the harness that we put on top with our agentic system to make sure that what we build adhere to our standard as a platform. So quality, performance, things that are very specific of our platform. And that's where also we have a second kind of layer of let's say building on top of a National Model. The third one, it will be on data. Data are very specific on brand and most of what we do is try to create tools and a solution there, improve and learn from the brand data in a sealed way so each brand has their own knowledge and the tools learn from that. knowledge specifically to improve themselves. All this stuff happening on top of foundation. speaker-0: I think I think that's a good insight and I think that's correct. Cause I think we've been through a period of people in the very early days you had people derisively saying, okay, X solution is a wrapper around existing models. And no doubt that was true in some cases. It was a very thin interface layer. I actually think you're right. I think the process of verticalizing AI for particular industries or adjusting it for specific disciplines like design, light pattern making, for instance. Is a process that relies on understanding. and you will see, as you just mentioned, like the the harness in a lot of cases is what makes the difference more than the underlying model, because it's built with an understanding of the way that people actually work. So I'm I'm with you on this. I think I think that's true. one thing, little follow-on to that would be if the interface Is what steers adoption, if we want to say it like that, if the interface is what makes up the delta between a generic model and a tool that is useful for a particular industry or a particular role. How do you think about things if the interface goes away? How do you think about people using some of these tools through MCP? within an LLM that they already use or through a CLI or something along those lines. Because that seems to be a place that creative platforms are increasingly getting into. and I've had a couple of interesting chats with people about that, where you would say, okay, the interface is the main thing, but actually we're also making a play for the post-interface era. How do you think about that? speaker-1: Yeah, yeah, we are already working on a bunch of things related to APIs, integrations with tools where basically our layer is invisible, but still like is the layer that gives the quality that is needed, that the LLM cannot give. So, the simple is that. I think we need to... We don't care about the UI much. The UI is super important because there are people that really, first of all, they love our UI, but also because that's the way that we use as a playground to test things. So that's fine for us if at some point the user, they say, I don't want to, I just want to use my, I don't know, agent or let's say AI assistant, muse, even instinct, they can do. Whenever they want to. Actually, we have the mobile version now of the platform. It is a chat if you go there. It's like literally we, the moment that you users are able to insert the input, ask the our tool what they want to do it and get the result. That's fine. is not really, we are not really keen to be locked to a specific interface. The most important thing for us is providing the quality that our users want. That simple as that. And that's the focus on everything. speaker-0: That's a decent answer. Now, when I spoke to Adriana for the mailroom show that I mentioned earlier, she told me that the fabricant had, and I'll quote, left 3D to embrace AI. You referred to it as a pivot earlier, right? and the reasons that Adriana gave in that interview were that 3D had proven to be, at an industry level, not for the fabricant, but for brands, it proven to be, as she said, too expensive, too technical, and too slow for for the audience you wanted to court. Now I think I understand the commercial argument there. If you're aiming to build tools for everyone, then specialist solutions like the ones that exist under the digital product creation umbrella are not the way to do that. I also understand the technical argument. I think there's definitely no easy turnkey way to build an end-to-end digital pipeline and get a digital asset that serves as a single source of truth. There is no full and complete digital twin that exists. And the the biggest and best capitalized tech companies and brands are still trying to build that. So there's not much that are more boutiquey kind of company can do. What I don't think I understand is whether you believe the choice is really that binary between 3D and AI. So you've showcased some workflows recently where you put a 3D render of a jacket side by side with an AI image of the same thing to highlight the discrepancy of time and cost required to achieve a visual standard. That makes a ton of sense if your remit is purely visualization, right? If you're just trying to get to something that people can look at, then AI is unambiguously a faster, cheaper way to do that. I personally feel like there's a much deeper need for simulation than ever in near future pipelines because they need to sit either side of generative visualization. Visualization is not the whole purpose of it. So if it were me, I don't know if I would have the same rush to embrace AI and leave 3D behind. Maybe that's not what you're doing. Maybe it's just I'm keen to see what you feel about this. What role do you see AI and 3D playing in your tool set over the next couple of years? speaker-1: Yeah, mean, five years ago when I pitched to Adriana actually about myself, I told her, look, I'm going to see at some point we're going to have a camera app where you point to yourself or you point to someone and you're going to change what you see in real time in a way that literally you imagine that you see yourself, but not like you're not going to superimpose AR. when the reality affects. No, you actually, you're gonna change what you see by literally rotating things, regenerating the whole image from scratch. And that's something that if you follow the world models things, that's what happening. mean, the idea is not the AI, they are just generating images, frames or videos, but it's more... They generated the entire world in terms of physics rules, dynamics, all things that they make a simulation possible. So if you look at the long-term vision of the people working on AI, the idea is that image will be just a subset of this, will be just a render of the 2D representation or projection of what you see. but the whole thing, the whole world behind will be generated as well. So in that sense, there was the need of 3D in terms of 3D as things made by meshes and vertices and all this kind of usual terminology and elements of the traditional 3D technology where you can generate things that they are very, very accurate at the same level of a deterministic 3D. Of course, only need for a mesh of vertices is because you can manipulate them, right? Otherwise, it's just also because you can transport the information passed to someone else and pass to tools, they can render that information or manipulate themselves. There is not a real need of 3D itself. if people can render that, manipulate, doing whatever they want with the 3D input in a different way than just using Blender or Maya. It's literally a problem of tools. So, in the short term I will say no, this is not black and white. We need 3D, we still need 3D because 3D is still like the highest quality that you can get to manipulate. three-dimensional content, okay? But there are, as I saw, part of renders where, of course, you may debate, yeah, but AI is actually even better now. But everything where you need total control of every single pixel, you need a 3D. But this is the short term. I don't think that in five years, probably, we may need at the same level. It's very difficult to think considering the curve of progress that we are experiencing now. Of course, it doesn't mean that it's gonna happen at all costs because it may be that still like there are situations where for some reason 3D is still the best solution. Maybe because it's more affordable, it's cheaper to do. Could be the case because... You don't know if 3D, using world models, is starting to cost a lot too much to be sustainable. Of course, people say, I still believe that it's better to do 3D. Remember that at the end of the day, cost has always been a massive element of determining the success or not of technologies. I've always been like that. speaker-0: I think there's I think there's two from my perspective, I think there's two ways that plays out. The the second one, I think you're right. So I think there is a world where 3D becomes the more affordable and more efficient option. because I I see it now, not in world models or anything, but I see pipelines where people are like, let me generate a visual representation of this thing I want to make, then let me use a vision model to segment out the product that I want to change from that generation. Let me recolor it with AI. Let me all of this is just kind of brute forcing around a gap in the middle, which is an accurate digital representation of the product that would allow you more efficiently to do the things that you're describing and potentially at lower cost, particularly if you start to inject all the world model stuff around it. So I do agree with that one. I think you and I probably diverge on I'm I'm c I I don't find the idea of world models as they currently exist very persuasive because they seem to do a very good visual approximation of simulation without actually doing real simulation. Now, is there a horizon where the distinction between that stops mattering? Maybe. Yeah. If you if you can get a visual approximation of a simulation right 99% of the time, is that good enough? Yes, for most use cases. But It's not a simulation if it breaks down sometimes. so I I have an issue with the fundamental architecture of world models that I think is distinct from simulation. And barring a real architectural shift in how we approach them, I think that will always remain the case. Now, your five year horizon, do we get an architectural shift? Maybe. Yeah. You know, I think I think maybe that's reasonable. speaker-1: Yeah, no, that's exactly the thing. I will not believe on the short term of this for sure. But definitely it's something that we can expect. Because even simulation, even deterministic simulation 30 years ago were not the quality of what you can get now. Because computation is more accurate. There are many things that technically improve in a way that makes calculus way more accurate than before. And that makes simulation more accurate also in the deterministic world. So that's where I think I expect as well that more better architecture, they hallucinate less, they are kind of more understanding what is wrong, what is right. Of course, will massively change the scenario. speaker-0: I want to ask you about so it is tied to what we just said there. So we're both fully aligned that if you can realistically visually depict something earlier in the product lifecycle, so that visualization piece we talked about, you can get people aligned on that. Like I I f fully buy into that idea. I'm less clear on whether that alignment on a visualization of something actually translates into the full product life cycle running any faster. or any more efficiently. So I looked at a recent paper you put out, which talks about faster design to approval processes as a result of AI. Great. I I a hundred percent. It also talks about faster production as one of the primary gains for design and product teams. Explain to me what you're seeing and your experience as a technology builder that shows you that those two things are related to one another. That visualizing something faster means you can produce it faster. speaker-1: That's a good question because to be honest, of course, I'm not someone that is manufacturing things. So I kind of know from the report that Adriana produced from our users and manufacturers. I cannot, let's say I'm not going to argument on that fact, on the comparison because it will be not my... in-person experience. But I give you the example of what I've done in the past. And first of all, I will say yes, I agree with you in some level that not necessarily is going to accelerate. For example, I tell you one of the things that is super funny for me always is about when we discuss rendering time. over the decades because I started using Maya, think was more than 20 years ago, my first 3D thing. And you are able to render things like very nice, they will look cool at the time, and it takes quite a while to produce one image, rendering from there. And when you get every single upgrade that you get on machines, you actually realize that you're getting the same time. And say, why are you getting the same exactly time to render things while actually the machine, I don't know, is four times faster all the time? And things like that. And the reason is simple because I was trying to get four times better results than the previous generation. So every generation was trying harder to get more results, better results than before. So... Going back to this comparison, it's not necessary that getting AI gives you faster production. It's mindset that is going to give you better production time and faster iteration. Because the moment that you use AI but you spend more time maybe on experimenting, you're going to still get a lot more time than before. on steps where you may, maybe you were spending one day on doing something, now you spend one day on trying multiple versions of the same thing that you were doing just once before. This one applied to so many things. Another example that I had in past, I remember when I was a video director, I was also filming with film, right, with a super 8 or 60 millimeters as well. And I produced a video of 4 minutes using 12 minutes of film, while at the same time a friend of mine produced 4 hours of footage for the same length of a video. 4 hours, okay? Because it was digital. It spent days and days to edit that footage, to understand what it was in the footage, because the unlimited resources to get something, and that's the same happens with AI, needs to translate into a way to select and filter out what is good and is not. So, AI generation doesn't necessarily translate into faster iteration if you don't change your mindset of be very selective, prepare, what you want to achieve before and having a storytelling in mind very clear, having the tools that are able to help you on select and filter out what is noise, what is not. And that's something that I've seen in so many levels on that area. speaker-0: I think that's right. To play devil's advocate for a second, I think one of the primary criticisms people point at 3D and digital product creation is it went a long way, got a lot of investment, didn't meaningfully change the product calendar. and I think whether the same criterion ends up being applied to AI, I guess we'll see over time. your enterprise tier currently lists tech packs and sewing patterns as beta features, beta. Sorry, I should say. A lot of generative companies doing this now. This seems to be, there seems to be a market need here, or there seems to be like a market opportunity that people have identified. Tell me how you ended up here. So you start with visualization, as we mentioned. What is it about either the work you're doing with the brands and and and customers and educators and so on that you're working with that leads you to naturally say, okay, we've done this. It's logical for us to do that. Because it seems like a lot of tech companies are converging on this idea now that you can build generative workspaces. They do a great job of the early stage visualization, end stage visualization. So concepting, exploration, then replacing product photography or what have you. A lot of companies now are seem to be saying, well, let's do the middle as well. Let's do tech packs, let's do patent making. how did you land there? And what do you think needs to happen on that journey now? speaker-1: I mean, it's a feature request, so simple as that. Brands ask if it was possible. We actually pitch pattern making almost two years ago for this time, but we were not happy about the results, so we parked just to focus on other things as well. And then we resumed last year working together, when we started together with the UAL. on this project. The reason why is because people are asking. to be honest, the pattern making, I don't think is the simplest one. I think it's kind of trivial to say, yeah, we can do that. But when you go, I've seen a few examples, the quality is not there yet. I think the problem is the fact that Many companies are doing not necessarily needs means that they're actually doing right. actually I see that probably there are also needs of the big competitive on showing a richness of portfolio and offer. There's actually another big problem in general. And this one affect all the creative industry. Because if you look at that, Now everyone is offering all the high models, is offering all the possible tools. I think the idea, fundamental idea of most of the creative tools, creative suites is about we need to show as much as possible. The problem is the focus. The problem is that you, I see, of course I'm not like naming, but I see. Many tools they look very average, not necessarily in the quality, but how they really try to create a user interaction with the user. it looks like they are really listening too much to the user because they are really very generic, let's say vibe coded solution sometimes. speaker-0: True. Yeah, yeah, yeah. There's this we get we get pitched vibe coded sh pattern making platforms every day. That's not that's not even an exaggeration, I don't think. yeah and that's they're very clearly vibe coded because the UI looks exactly like every other claude designed UI. speaker-1: I mean, if you literally go to cloud or to GPT and you ask, can you generate a pattern for me? Can you generate a tech path for me? Of course you get an answer. So the idea to create a tool, of course it's trivial because it's like five minutes to start prototyping something. The massive problem is where you want to really make something that can be used in production. And that's a month and month of conversation with users because it's not even like, they give me the brief is done. No, I wish I wish it was like that. It's there is a lot, there is a lot to do. this is not technical. It's not even the most challenging thing that tech back probably. but there are one of, nuances, amount of details that you need to put in place, it's knowledge that you create only talking directly with brands. It's not like something that you can get a data set, get a paper and produce. speaker-0: Yeah. I think there's nothing there's nothing about running a generative workspace, even if it's specifically calibrated for fashion, that makes you qualified to do tech packs or makes you qualified to do pattern development without spending that time in industry that you've just described. there's no there's no straight shot from I have image generation tools to I have patent making tools. They're fundamentally different disciplines. and with my Cynix hat on again, I'm not naming names, but there are a fair amount of these companies that are venture capital backed or private equity backed and they will be looking to expand their user base, I guess is the simplest way I can I can frame it, is to say, okay, we've tapped out the marketing department and we've tapped out the merchandising and early stage creative design. Who else could we reach? so I think I think that's a big part of it. I'm more time here because I don't want to take you away from your work for too much longer. So I do want to product something else. The September paper that you put out, which we will put a link to in the show notes, you in there you say that amongst the fashion teams that were surveyed, so I think this was the work Adriana did, you mentioned, all of who all of whom are using AI every day in the development of real collections, it says that the question of whether AI is good or capable enough is quote unquote settled. Now, the survey that we ran earlier this summer says the opposite. So it was around a hundred brands. That tells us that ninety percent of fashion professionals do use AI every day at home and at work, but only about quarter of them trust its output enough to base important decisions on. And around half of them are concerned that there's a gap between what people assume AI can do and what it can actually do, which doesn't sound settled to me. I know these aren't apples to apples comparisons, but I wonder what are you finding that we're not? And I also want to know how you f personally feel about all this, because I come up against the like jagged edge of AI a lot in my own experiments and everyday use. And I pinball pretty strongly between AI is amazing and AI is useless and I've just had I've just forced quit codex because it's annoyed me that much that it can't do something fundamental and basic. What is it you're seeing that you think leaves the question of whether AI is good enough as settled, whereas what we're seeing is that it's not. speaker-1: Yeah, I mean, first of all, you're interviewing people that visibly they are not using our tools. Sure. No, I'm kidding. But of course, results that Adriana collects from our users, think, I don't want to, of course, sell what we do, but the idea is that probably... The way that we support our users make a big difference on how they perceive AI and how they perceive the utility of AI for what they need to do. think again, going back to the initial questions about what is software, what is service, it makes a massive difference. I think if you leave AI as a tool to someone that has not much knowledge about how AI should improve their life in general, that's very, I think it's very, very difficult to tell, improve the AI is gonna make a difference. And again, and you can see this in so many different industries and contexts. So this one, think is the most important thing that you're gonna make differentiate your result maybe from our result. Because of course I cannot, I don't know the users, the interview, how they use AI to understand. Because I know how users, how they use AI. And every single thing that they do wrong, any time they reach out, we help not to tell, yeah, you can do this and that. But really, it's a hand-to-hand thing where we try to tell. what AI is capable of, but also what AI is not capable of. And that's super important, more than what is capable. And yeah, that's kind of, I think the overall big differentiation that you see. And that there is the biggest challenge. That's why again, everyone needs to educate in order to. And that's where the problem is actually that, you know, the moment that everyone is trying to fear about AI killing everyone or AI getting all the jobs. All these things are not going to help, of course, because they are going to create this layer of contracts and tensions between technology and humans that honestly is useless. speaker-0: Fine. And I think I th I'm going to support you on that. So I think the big difference is between our general survey, which was across a bunch of different disciplines, a bunch of different use cases and so on, and included people who don't use and don't like AI in some cases. and there's a difference between that and what you've described, which is the harness makes a difference, the support makes a difference, the industry knowledge makes a big difference. And if you sit down and you interact with a tool that is Opinionated, and I mean that in a good way. I mean that in the literal software sense where it comes ready-made with ideas for how you should use it and it is intended to steer you in that process. I think that's you're gonna land with more trust and you're gonna land with a better opinion of that than if you're just presented with a blank slate and you're told that you can just just trust play around with it. And you mentioned MetaMuse before. I think the reason that's ended up being so popular, again, not not available in the UK, so I can't test it right now. But the reason that's so popular is not because it can do something that you couldn't already do with Codex or Hermes Agent or OpenClaw or what have you, but because it has a friendly face and because it comes with ideas. There's a literal ideas tab in there that is like, you should do this with this and this is what it's good at. And sh let's leave out the things that it's not good at. And that does seem like it makes the biggest difference for how people think about and trust things. Final question. I'm gonna go back to that same paper you wrote, which is there's a conclusion there that says that most brands are stuck in what you call phase one. and that does align pretty closely with our survey. Like the data does line up there. and we found that very few companies have a really good rubric for judging the return on their AI investment. And most of them haven't really connected it to their wider tech estates. It's it's it's what you would class with my analyst hat on as f the kind of fairly immature deployments. To bring us to a close, mm tell me what phase one is and tell me why you think companies are stuck there. and while you're at it, tell me what you think it'll look like for fashion to get to the next phase. speaker-1: I would say that what I think is most challenging is to try to educate the leadership teams, home brands, how to, or like the team that are managing the actual users of the AI. I kind of see a lot of anxiety on leadership teams sometimes to understand that. how to really leverage because you know, there is a lot of spread of word about, yeah, with AI you can improve this and that. It's all about numbers, right? Where, and of course, translate numbers into actions, into how, okay, what are the action, what are the really things that we can do to improve these numbers on our business? That's the whole thing because of course, yeah. Maybe they rushed into, okay, let's buy business license with cloud, let's buy business license for enterprise tools of the fabricant or any other one. And then they realized, okay, but we have to use it and how to make our work better. And I think it's again to live this... phase one and make sure that they know how to work properly with the tools. They need support and that's again where companies that are working on their tools, they should really work a lot more. I do believe, I'm talking for us of course again it's very important to do, but I kind of curious to understand where is the industry going for that in general for they are helping everyone because it seems like from discussion that I see that there are many brands that they are really like stuck into. Okay, we spend, I don't know, $10,000 this month on cloud tokens, but we didn't prove anything. And that's one why, because there are many reasons. Because first of all, nobody says how to use this stuff properly. I give you an example. Because it's a very honest thing. A feedback that I give internally to my team is don't produce document with cloud or any other tools. must not be reviewed by you, by yourself. Because... the moment that you give me Claude and anyone else, anything else like that, they're to produce very verbose document. That's the standard. And I can't say if it's the standard on purpose to spend more tokens, of course, but definitely there is, so I don't know if it's a strategy, of course, but definitely they are very verbose. So whatever you ask to produce. the document will be, I don't know, ten times larger than what the human auto will ever do. The problem is that someone needs to read. If you pass a document that someone has to read, the risk is they're gonna pass to my client agent to read that. So garbage in, garbage out means because if you don't review something that you produce, and don't review something that you produce, we're gonna... propagate and amplify errors, right? So this is something that I've seen, I kind of stopped a few times that happens in my company. And now we are way better on that, all of us, because it was also me kind of producing stuff sometimes where, you know, we are in rush, you need to produce something. By singing many things, I don't know, Victor, the Sea of Synthesia, post once about on LinkedIn, about that. It's a very, very common problem. So yes, you get the tools, but you believe that the tools are there to, I don't know, reduce your time. Actually you spend even more time because you need to review something that becoming incredibly larger. So that's phase one. That's where we have the tools, but still like you. We don't know how to use it in the way that we should. It's very common. The way is education. It's taking time. speaker-0: Yeah, I agree. And and I think we s we see that from the data. We see it from our own conversations with people as well. Is if the incentive or the mandate is to use AI, but it's not clear what you're using it for and where you measure the return, what you end up with is a lot of output, but not a lot of outcomes. and I think that's that's what you exactly what you're describing there as well. So I think we'll see some changes in this over time, if only because people start to enforce token budgets, if only because people start to get more restrictive about where the AI spend is measured and where the AI return is measured. I think that kind of scrutiny is is inevitable. I think that lends itself to then tools that are well calibrated for delivering outcomes within specific industries and specific disciplines. Cool. Marco, we've we've gone long, but I enjoyed this conversation. I think could have gone a bit longer, but I will maybe we'll have you back on some time and see how some of this stuff, particularly with world models and everything else, plays out over the next few years. For today, it's been great talking to you. Thanks very much for your time. speaker-1: Thank you. Thank you so much. Bye bye. speaker-0: And that's the end of my chat with Marco. As you can tell, he and I don't exactly see eye to eye on everything. I think that's probably true of me and a fair few C-suite members of AI companies, to be honest, even though I consider myself pretty optimistic about AI in general. Are you gonna hear some more of that come out over the next few weeks? I appreciated Marco being as game as he was to talk about a lot of different things with me today. When you listen to these shows, it maybe sounds easier than it is to be in the hot seat and to feel my questions, but we do try to go after some big topics, whatever subject we're covering. And it's better to do that with a guest who's open to being challenged. So I enjoyed today's conversation. Hopefully you did too. As I mentioned at the top of the hour, we're going to be coming back to some of these themes again before the end of the year. So keep an eye out for that. In the meantime, I'll be back next week with both another one-to-one interview and a fresh instalment of the edit. So come back on Tuesday to hear Grace and I debate the headlines from the last seven days, and come back on Thursday for the next installment of this series. For now though, thanks for listening, and I'll talk to you again really soon.