Latest / The Interline Podcast / Fashion Week, On-Device Models, AI-Native Product Development, And Doomerism
Transcript
- speaker-0: So welcome to the edit from The Interline, the show where we run a quick fair analysis on the pick of the most important fashion and beauty technology stories from the last seven days. I'm Grace, the social editor, and I'm joined by Ben, the editor-in-chief, and together we have around 25 minutes to give you our analysis on the stories that really matter. speaker-1: Hey Grace, good to see you again. I'm gonna apologise for both of us in advance. It's an AI heavy week this week, but you go where the headlines lead you. speaker-0: Yeah, exactly. It is, but they all are slightly different in a way. So I still think it's gonna be super interesting. speaker-1: I think every conversation is an AI heavy conversation, particularly this week. So without further ado, let's do it. Hit me. speaker-0: So the first story is about Google's push into New York Fashion Week. So Google partnered with two emerging designers. One is called Jane Wade and the other is called Sergio Hudson. They both used AI Google's AI Flow creative suite, but they created slightly different tools. So Jane Wade created firstly a visualization tool. So it helped her to think about the styling of her runway looks prior to doing the show, and also it helped her to make kind of colour and fabric decisions. She also worked with Google to create a kind of simpler version of Adobe's Illustrator. So this is gonna help her team do tasks quicker, and it also means that they don't have to have Adobe skills necessarily. Sergio Hudson worked with Google to kind of create a more spatial visualization tool. So he used it to create a visualization of what his set design would be for his runway. And this meant that he could visualize it before committing to kind of the costly build of the runway. Now, Google has said that they've intentionally wanted to partner with creatives to develop these tools. Because they really wanna champion the creativity within all of it. But do you think this is a different strategy from Google? Do you think it's gonna have a big impact on the industry? What do you make of this first story? speaker-1: So Google have so many different AI products that I struggle to reconcile them and to understand which ones are which. I think people within Google struggle to reconcile these things as well. ⁓ So the visual generation side of AI is one of the two kind of areas that are AI survey, that the respondents told us that they perceive AI to be the most mature. So generating images in general. The other is the kind of analytical and market trend ⁓ analysis side of things. Now, generating visuals, people your mind tends to instantly go to coming up with concepts for new products. That's what most people mean by generating images in the creative stage. Like the same way that historically you would open a mirror board or what have you, and you'd pull together some material swatches and reference images and things. Being able to do that kind of pre-visualization is what I think most people think of as being the Image and concept generation side of fashion. Then you have it all downstream. So you have all of the product detail page photography, campaign photography, and so on, which there's a big generative push there. They're all basically the same thing, which is generating visual representations of something that doesn't physically exist in place of either you can't have that thing because it's not here yet, or it's impractical or costly to Build it, prototype it, and ⁓ and photograph it. This story is interesting to me because you can take the same principle there and you can just apply it to a range of other different use cases. So previs for show looks ⁓ has always been, I think, a pretty intense creative process, but one that also involves a lot of like non-value add tasks, literally moving clothing around, layering it, experimenting with it, having models in one place. I think this is one of those areas where. The unit economic side of it makes a bunch of sense, the same way that it does in product detail page photography. If it's faster, cheaper, more flexible, and more agile and more creative to be able to do some of that previs stuff for a show with AI, I think it's it's logical and it's a it's a decent fit. Then the other side of things is if you want to do architectural visualization, which is what runway builds, the actually building out a runway in an environment and stuff around it technically is. Historically, you would have done that in 3D CAD, or you would have done it with some 2D sketches and some artistic license. Again, that's something that you can do faster, easier, cheaper with AI. And in a way that I don't think is taking away anyone's job. It's not taking away any creative potential. And that's true of both of these use cases, really. You have teams or individuals who previously would have spent a lot of time doing something manually, can now visualize it, not virtualize it, but visualize it in a way that gets them closer to the end result and that cuts time and that cuts cost. So from a actual kind of does AI make sense to fit into fashion weeks and into runway shows, yeah, it seems like it does. Seems like a pretty clear fit. Is this a product for Google? Is this something you're actually going to go and like productize and bring to market and give it a proper name? And do anything beyond this kind of halo experimentation. ⁓ who knows? It seems like something that's destined to fall into Google's pile of we have this AI idea and we tried it and isn't it cool? And it's somewhere on this kind of scattered assortment of of products that we have. But you don't need to partner with Google to do this. If you're putting together your own runway show, no matter how big of a designer you are, you can do exactly the same things here with other off-the-shelf models. You don't need a formal partnership for it. But It's when when you put this one across my desk, I was initially a bit like okay, AI involved in Fashion Week. Here we go. I wonder what that's gonna look like. Actually, deeply practical, deeply useful. ⁓ I just don't think it's necessarily gonna turn into a formal product. speaker-0: The next story is about a development from Daydream, so the AI shopping discovery application. They've just built two new features using the new iOS 27, so Apple's new iOS. Essentially, the first one uses Apple's image context capabilities. So that means that any image that is in the camera roll on your iPhone, Daydream can analyse that image and then make the image shoppable. So for instance, if I had a picture of myself wearing an outfit, or if I'd screenshot a picture from Pinterest or from Instagram that I wanted to use as outfit inspiration, Daydream will essentially analyse each piece of the outfit and then you can click on it and then you'll be able to shop it. And also I think Daydream is also recommending different pieces that could also go with the outfit. On top of this, they've also just built in this Siri system for the app. So now a user doesn't even have to open the application on their phone. They can simply grab their phone, use Siri, and ask Daydream to find a new product for them. So I think the example they gave was someone needed a new Blazor for a board meeting. They could just ask Siri and then Daydream would find it for them. I also think Daydream is now incorporating shopping passports. So I know we spoke about this a couple of weeks ago with e com ID shopping passports. So I wanted to get your take on this. Do you think it's a novel, new experience and a new feature of the application, or do you think it's a good insight into how we're going to be shopping in the not too distant future? speaker-1: I do have some thoughts, but I'm keen to get your perspective on this one. two questions. One, have you updated to the latest iOS? And two, is this something that you think you would use? speaker-0: I haven't updated the latest iOS yet. I will be doing it, but I think I would use something like this, but only because it's becoming more and more spoken about and agentic shopping is getting bitter bigger and bigger. So it does seem like it has value. But I don't think I would have been an early adopter of something like this. speaker-1: Okay, fine. So I've also updated to the latest iOS, although weirdly I'm on the wait list for Siri AI, just professional curiosity more than anything. So I can't test this myself for a bunch of reasons. Now, what you're describing with the camera roll stuff is ironically, based on what we just talked about, another product that Google had, and I think they've probably since forgotten that they had, which was building the wardrobe ⁓ quantification stuff from your Google Photos library. Saying, ⁓ look, these are the outfits that you already have, these are the outfits that you own, which runs on the same fundamental thing, identifying garments and pieces in pictures and then surfacing that to ⁓ to the end user. Now, the history of smartphones, not to get way off topic into general tech, is very much Android is a grab bag of features and places that Google and its kind of OEM partners and people just try stuff. And some of it works, some of it doesn't. It doesn't tend to break into the zeitgeist the same way that it does when Apple does things. And ⁓ iOS by contrast and Apple devices by contrast are the ones that attract the most attention. ⁓ they're the ones that have the the largest footprint in the way that we think about technology. I know people who are Android users get endlessly frustrated by this because it's it's a case of saying, well, what do you mean? This this has existed on Android for a long time. It probably hasn't existed on Android for a long time in a way that breaks into the public consciousness. Which is what I think you have Daydream trying to get ahead of here. Now, there's there's two bits to it. ⁓ access to stuff on your camera roll and using ⁓ the Apple's own image context capabilities. Great. You're you're tapping into something that exists and is broadly available and adds value to your application. The second part of it, though, is the most interesting part to me, which is that. A lot of this is on device. So a lot of this, ⁓ the Siri AI ⁓ side of things is done locally. Not all of it. There is a opaque process where Apple hands off things from on device to what they call private cloud compute. But if you think that going forward, so if I remember correctly, this is everything from an iPhone 15 Pro and upwards, is able to run the ⁓ the more complex on-device models. If you think of that as the the future target for these kinds of applications that leverage new stuff that your phone can do. This reminds me of the early days of the App Store as much as anything. ⁓ which is to say that felt very much I'm old enough to say that felt very much like an experimental wild frontier where if you had an early iPhone, you had that app where you could pretend that you were drinking a beer because it used the ⁓ the gyroscope to simulate that side of things. Seem completely useless, but or then before you know it, the App Store completely up-ended ⁓ application distribution, social media, ⁓ Apple's entire service revenues and everything else. This I think is a good and prescient use case for that. It's Daydream, I think, getting ahead of what these kinds of on-device capabilities that I think have a chance just because it's Apple and because of the ubiquity and because of the way it breaks into the public discourse. They have chance of becoming the next kind of app store, potentially, if AI really takes off the way that we think it will. ⁓ and if it does, I think you'll see more people building on top of these capabilities within iOS rather than doing it just in fragmented other applications. speaker-0: So, the next story we're talking about AI coming into the product development process. So, Raspberry AI has announced that they've developed a new system of AI agents that are essentially unifying every stage of product development. So, design, marketing, merchandising, all the way to e-commerce and actually selling the product. Now, this new pathway is Said to speed up the product development process nearly two to five times. And big brands like Skims and Aloe Yoga and Oscar Delorenta are already using this. Raspberry AI has obviously digitized parts of the product development process in the past, but now AI is really unifying all of those fragmented teams and elements. So I wanted to really get your take on this generally. And also ask if you think that this is gonna be the start of mass adoption of on demand processes. speaker-1: Okay. So first things first. I I don't know anybody at Raspberry. Never been able to successfully establish a line of communication with the company. If anyone from Raspberry is listening to this, we'd like to chat to you. We'd like to learn a little bit about how this works. What we have here is a pretty widespread trend at the moment, which is companies that got their start in generative image workflows. Now I don't I don't I don't necessarily mean kind of Campaign photography or even product detail photography, just generative workflows and workspaces that have a primarily or exclusively visual output. ⁓ there's and there's quite a lot of those. Raspberry is by no means the only one. There's at least five or six that I can think of off the top of my head. ⁓ on my ⁓ got a weekly interview show, I interview the CEO of Flora, Reb Weberwong. Again, not fashion specific, although ⁓ they have grown more in the way of fashion capabilities, but a big node-based generative workspace. Now, There's a lot of those, they've all been playing on the visual representation end of things. I think they're all starting to realize that when you sell to product-centric industries, so industries that make physical things and the images need to correspond to the c characteristics of those physical things, images only get you so far. ⁓ images get you to the point of if you're doing it for Initial concepting purposes, they get you to a point of, okay, cool. This was this was faster, this was this was more effective than traditional mood boarding. If you're using them to sell in to internal audiences for ⁓ and trying to secure ⁓ adoption, build assortments, ⁓ line review, that kind of stuff. Again, great. ⁓ if you're trying to use it downstream for marketing and ⁓ advertising and content creation, again, wonderful, ⁓ very well suited to that. Our data from the AI report tells us that the closer AI gets to real production. Which is what they are talking about here. When you talk about product development, when you talk about merchandising, pattern making, tech pack, or technical specification generation, you're talking about real goods. You're talking about turning images into real product. Now, the market, based on our survey, perceives that to be one of the weakest areas of AI. So I think there's clearly a gap here that companies are going after and they're recognizing that. The capabilities they have only get them so far, and that they need to build new capabilities. I don't think those capabilities are easy to build. Again, I have not seen, I've not seen the solution here. ⁓ so I can't say one way or the other. I will say that if you're if you're used to playing in the image space, the quickest analog I could use would be like Photoshop or or ⁓ or Illustrator or InDesign, you know, somewhere. an application or suite that's focused on like visual creation. That's not the same thing as making a product. And if we think about fashion workflows, for example, Adobe Illustrator is where sketches get made, ⁓ Photoshop is where th final images get retouched and polished and so on. The real work is not done in those suites. The real work is done in separate, very big solutions that have been developed over the course of decades, your PLM, your ERP, your supply chain management, and so on. I don't know if it's a bit of hubris. I don't know if it's gonna work, but like I I every AI company is converging on the same idea, which is how far do we go? Yeah, we we generate images for fashion. Why can't we do the rest of it? ⁓ how hard can it be to do the rest of it? I think the answer is quite hard. ⁓ and I'm gonna reserve judgment until I've had a chance to sit down and spend some more time looking at these capabilities. ⁓ again, Raspberry by no means the only ones doing it. ⁓ but it's it's very clear that this is something the industry is lacking and that the technology vendors are trying to build. I just don't know if it's as easy as people present it as being. speaker-0: It's funny you say that this is becoming a trend in the industry because Lectra has also just announced that they've launched Apogee. So Apogee is a cloud-based platform that uses Agentic AI to address all of the complex product development workflows. And essentially its aim is to centralize all of the critical product data, to bring together designers, manufacturers, and suppliers. All into one digital workspace. So key features of Apogee include concept generation from prompts, and also from sketches or images, and also real-time collaborative editing, and also prototyping through 3D simulations and photorealistic renderings. So I'm sure you have similar thoughts with this one as well, but I wanted to get your take on lecture. lecturer's approach to this and if you think it's any different to the previous story. speaker-1: Okay, so it is it is different, is the simplest answer I can give you. ⁓ I was at the press launch for Apogee in Paris ⁓ at the at the start of this week on Tuesday. So I do have some additional insights that are not contained in the press release. ⁓ I might also have some other additional insights that will reveal themselves over time as well. it is fundamentally different. And it's fundamentally different for the reasons that I just described in the in the previous one. So if you if you as a company and your product starts with the image. And you then need to walk backwards, or depending on I suppose backwards or forwards, depending on where you're generating the images, to actually creating real products. Real products made of materials, patent pieces, sewing operations, and so on. I cannot emphasize enough how fundamentally different that is from image generation. Just because an image superficially recognized. represents or resembles ⁓ the output of a real product development process doesn't mean that it instinctively contains any understanding of that product development process. Lectra as a company ⁓ has existed for more than fifty years. ⁓ they've I think ev most people know them for their hardware rather than their software, although they're I think they're a software business first, then hardware, then software has become a bigger component of their operations. That hardware is all in manufacturing. ⁓ so that hardware is in spreading, plotting, cutting, marker making, fabric estimation, all core product development and production work, right? Which means that something like Apogee starts the opposite way. It starts from real pattern development, real cutting, real production, and it walks into the visual representations of that and the agentic side of that. The other bet, the opposite bet from the the companies we just talked about, is that you can go from the image backwards into the complex ⁓ product development side of things. History, based on the journey that 3D and digital product creation have been on, would suggest that starting with the foundation and building upwards is easier, more effective, more complete and more robust than starting with the visual representation and trying to reverse engineer it from there. I'm not saying history will necessarily play out the same way. Again, I have not sat down and experienced any of these solutions. But my bet at this point in time would be that you will get a more complete, useful, and accurate version of agentic or AI native product development from a solution in a company that's previously done a lot of work in product development rather than from a company that's previously done a lot of work in. Image generation. I again no guarantees. I don't know if things are gonna necessarily play out the same way, but I do see these things as fundamentally different. So even if everyone seems to be converging on this idea of agentic product development, generative tech packs, ⁓ generative specifications, ⁓ generative pattern development, all this kind of stuff, the different ways into that are I think gonna mean that it plays out very differently for for those companies over time. We will be here to observe. I think you'll hear some more about that in the future. speaker-0: So the final story or the final topic that we're gonna end on is kind of looking at AI in a bit more of a scary light. So it seems that more and more there are people who are leaving the big AI companies and then coming out with all of these statements that AI is going to actually kill us in the near future. So one of the most recent examples is from Jacob Coxton. He actually went viral last week when he quit his job as an AI researcher at Anthropic, and then he made a viral post talking about all of the dangers about AI. This comes just after his ex-boss, so Dario Amodi, recently made a call out through an open letter for all of the AI developments to slow down. And this is something that Sam Altman at OpenAI agrees with, and even Elon Musk at X. AI as well. They've both agreed that the proposal to really slow down and regulate AI is very, very important. So I wanted to get your take on this. Do you think we should all be really worried about this? And also do you think it's gonna impact fashion in any way as well? speaker-1: Yeah. So this one is super weird. ⁓ it was the this was the front page story on the BBC when I was in Paris earlier this week. and even though I was looking at the French version of the BBC, it was ahead of Celine Dion's return concert in Paris, which ironically was at the the hotel that I was staying at. now we're not a general tech publication, so I'm gonna I'm gonna stop short of having a real opinion on ⁓ AI development in general and the social side of things. I I don't think it's gonna kill us all. I think ⁓ the the mechanism by which that would happen is incredibly convoluted and ⁓ and unlikely. What I think is more likely is that we see more AI incited or AI related disruption to infrastructure. ⁓ now how Big that infrastructure is and how essential it is to ⁓ either a particular country's operations or to the fashion industry's operations. It's like it remains to be seen. But you it doesn't require a big leap of faith to go from okay, you can have an agent swarm from an unreleased model going through testing and evals that breaks into an ⁓ an alternative AI lab systems. It's not it's a short hop from that to the same. Swarms and the same ideas ⁓ can break into alternative systems. And that seems like a far more likely outcome for me. Now, ⁓ alternative systems can mean anything. Alternative systems can mean businesses, they can mean critical civilian infrastructure and things like that. Now, ⁓ the week before last I would flew from France to ⁓ Germany. ⁓ so I was ⁓ chairing the the Clow summit in the in the EU. I was lucky I was flying from France to Germany rather than flying from the UK to Germany because I think my flight would have been grounded because of the air traffic control disruption which happened here in the UK. I'm threading two things together here, but I think a far more likely outcome is that you have un ungovernable AI that causes that kind of interruption. Things like I and that that's you know, that's sci-fi and scary and dystopian in its own right, is saying, okay. All flights are grounded because we're fighting off an AI botnet. ⁓ but it's not impossible to conceive of. In fact, I think that's something that ⁓ Dari Amade specifically cited was the idea that the whole internet could be taken over in six to twelve months with ⁓ a persistent botnet. Probably too much. But it's not a far leap from that to ⁓ disrupting shipping lanes to disrupting ⁓ the energy grid in ⁓ a particular region. And if that region is relevant to a fashion brand either as a consumption market or a production market, if that shipping lane is relevant, then all of a sudden it becomes a ⁓ a big fashion concern. And this sounds alarmist, but the example I would go to for this would be ⁓ Stuxnet. So if anyone has never ⁓ there's a good Alex Gibney documentary about it. There's a couple of good books about it. Stuxnet was the cyber attack that the US staged on Iran what a couple of a decade or more ago. Which used very simply used used software to disrupt the programmable logic controllers in the centrifuges that were apparently being used to refine uranium ⁓ for either nuclear energy or nuclear weapons development. that's an example of how cyber Attacks can manifest themselves quite quickly in the physical world and can cause actual damage to ⁓ a a piece of infrastructure or a piece of scientific development or so on. ⁓ so I'm not ducking and covering and expecting ⁓ AI to release novel bioweapons on us anytime in the near future, but I also wouldn't be shocked if we see more in the way of ⁓ crap, we can't source with th this particular shipping line is closed or this country's energy for it is is under attack. Not just within my lifetime, but within the next like five years or so. That seems like something that is feasible and ⁓ and possible. And that's scarier to me than the sci-fi stuff because it feels at least plausible and realistic. I don't know. ⁓ final thing, how about you? Are you expecting AI to kill us all, or are you ⁓ a bit more pragmatic about it? speaker-0: I really wanna believe it won't, but I think it is quite alarming when all of these big people are saying it. So I but I don't think there's any way to truly know the truth of what all their intentions are behind the scene, but hope hopefully not. speaker-1: Very true. I think these people do have vested interests in emphasizing the capabilities of their solutions. But for our business to business audience listening to this podcast for fashion enterprises, ⁓ the big AI labs are trying to sell tools to them. And I don't think you sell tools to enterprises very well if you sell them on the basis that they might go and attack somebody else's ⁓ and an another enterprise or somebody else's infrastructure. So I I do get the argument that maybe this is a bit grandstanding on the part of the AI labs. But I it's almost counter to what they're actually trying to do, which is to IPO and to become big big business software providers. So there's a kernel of truth in this. I just think it's incredibly weird, again, that this was a top headline on ⁓ on the BBC. And I think that's more representative of how intense and distilled the AI discourse has become than it is of anything else. speaker-0: Yeah. Well that's all the AI stories for this week. speaker-1: Let's try and let's try and mix in some nine AI next week, but we we go where the headlines take us. Grace, good to talk to you as always. I will chat to you again next time.