The Edit is our weekly show, where Social Editor Grace Robinson quizzes Editor-in-chief Ben Hanson on five of the most significant fashion and technology stories from the past seven days.
This edition covers the US government’s export-control ban on Anthropic’s Fable 5, fresh data on AI search and AI spend, Epic Games’ use of generative AI and interoperable skins, StockX’s move from gatekeeper to live-shopping platform, and what “beauty tech” actually means.
This show is available in video format, embedded above, as well as audio, on Apple Podcasts and Spotify.
For existing audio listeners of The Interline Podcast, our prestige interview show now releases on Thursdays, and will be upgraded to video soon. Subscribers to our mailing list will receive notifications for both shows.
NB. The transcript below has been lightly edited.
Grace Robinson: Welcome to The Edit from The Interline — the show where we run quick-fire analysis on our 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. Today we have less than twenty-five minutes to give our analysis on the stories that we think really matter.
Ben Hanson: Hey, Grace. Good to be here again — second time doing this. You’ll notice we now have a motion graphics intro. We don’t have theme music yet. We are working on it; it’s surprisingly hard to pick. We somehow managed to land on one for the interview show ages ago that we’ve been able to stick with for a long time. I’ve been agonising over this one, and we’re going to get it sorted. But for now, no jingle. From next week, jingle.
To get in the spirit of the whole thing — hit me. Let’s go.
Fable 5 — frontier AI and the export-control problem
Grace Robinson: The first story I wanted to talk to you about is more of a general technology story. Anthropic, the company behind Claude, released Fable 5 on the 9th June — but just three days later, the US government took it down. This was the most powerful artificial intelligence the company had released yet, and it was capable of very complex projects. Obviously this is a more general technology story, but I was curious: is it going to affect fashion technology?
Ben Hanson: Yes, absolutely — and in two ways.
The first is that everybody likes to be on top of the state of the art when it comes to AI models. Everyone’s familiar with the fact that if you’re building something LLM-centric, generative-image-model or diffusion-centric — if you’re building on a foundation of AI — it’s become a bit of a cliché that at some point you’re going to have to stop and start the work again, because the capabilities of the frontier models are going to change. Fable 5 represented a significant step, capability-wise. It was the user-friendly, safe, approved — ideally — version of Mythos, which Anthropic had previously talked about as a model that was too dangerous to release.
It’s interesting from a state-of-the-art point of view. The especially interesting bit is how the US government went about banning it — or rather, encouraging Anthropic to take it away from users — and the mechanisms they used to do it. They’ve used export controls. The reason it’s not available to anybody at the moment is that the US government stipulated it couldn’t be used or interacted with by any non-US citizen, including anybody who worked within Anthropic — effectively making it impossible for it to be available.
Now, there’s a world where that holds, and what ends up happening is that the frontier AI models are only available to US citizens. There’s a world where this happens over and over again, where frontier models get released and they’re consigned the same way cryptography was in the past — message cryptography was classed as munitions at one point, for export. They get flagged in the same way. If you’re building on AI models made by American companies, hosted in America, under the jurisdiction of the American government, and you’re not a US company and not solely employing US citizens, this suddenly opens up a horizon where the models everybody is building on — whether that’s Nano Banana, Nano Banana Pro, Gemini, ChatGPT, GPT image 1.5, any of these things — can go away. And they can go away in a way that represents a potentially uneven playing field and uneven distribution.
The US is supposed to be trying to compete with China on AI. This potentially opens the floodgates. Long story short: if you’re building on frontier models, yes, the capabilities change quickly — but the availability might suddenly shift from underneath you. If something like Nano Banana went away, every generative workspace that fashion uses is suddenly unavailable to its customers and to brands. It’s a huge deal.
Adobe and Ramp — two reasons for AI optimism
Grace Robinson: Staying on the topic of AI — it probably won’t surprise you that each week, when we’re looking at the stories, there’s a handful of negative ones we could pick from about people disliking AI. That’s exactly why the next two stood out to me, because they were actually positive.
Adobe reported that more and more people are using AI search, and it’s yielding positive results for travel websites and retailer websites. Specifically, Adobe reported that visitors coming from AI sources converted 54% better than those from non-AI sources — and that’s nearly doubled in the last year in terms of how many people are using it.
Adobe also found that 80% of consumers who use AI while shopping say they now rely on it more than they used to, and 79% say it makes them feel more confident about their purchase. On top of this, I saw data from the Ramp AI Index talking about how, as more US businesses adopt AI, it’s actually costing less than they initially thought.
These are two positive AI stories, in my opinion — but what’s your take?
Ben Hanson: I’ll play a bit of devil’s advocate. I think you’re generally right, and I sometimes fall on the AI-sceptical side of things myself, just from a value-return and spend point of view. Both of these are interesting from that perspective. I’ll do them in reverse order.
The Ramp AI Index is especially interesting because it’s updated very regularly, and it tracks adoption, spend and a few other variables specifically for US enterprises.
A lot of people have written recently about the fact that a lot of AI workflows, AI tools and AI business models are predicated on the token cost — the cost of inference — being subsidised. So if, right now, you do all your e-commerce PDP images using a generative image model with a fixed token cost, you know roughly what each image is going to cost you. There’s been a very real, prevailing understanding that that token cost is artificially held down, artificially depressed, and that what it would actually cost you — if the model providers passed the direct cost on to you — would be much, much higher. So the unit economics of replacing traditional photography with generative photography make sense right now, from a pure commercial point of view. Maybe they wouldn’t if the direct costs were passed on.
Now, the Ramp AI Index doesn’t change that calculus, but what it does show is that the median US company is spending a lot less on AI than you’d think — which means they’re a lot less exposed to token-price increases than you’d suspect. The top 1% are spending an extraordinary amount — something like $7,000 per employee per month on AI tokens. The median, though, is something like $11 a month per employee. You can inflate that 5x, 10x, 15x, 20x and it still doesn’t completely invalidate the business model.
So if you fall in that median bracket and you’re thinking, “Well, I’m spending $50 a month or so on AI tokens for my e-commerce employees” — it might go up, but your exposure is a lot lower than the top 1% would suggest. And the headlines do talk about the top 1% a lot.
The Adobe data suggests two things. It suggests that people who come from AI sources to retail storefronts are readier to spend; that’s what we mean by conversion. You can spin that two ways. This is what I meant by devil’s advocate.
You can say, yes, the traffic you get from AI is higher-intent, higher-quality, higher-value. That’s what this seems to indicate — and it’s also what Google and others tell publishers, that you’ll get better readers. What this doesn’t reckon with is volume. So, yes, your visitors may be 54% more likely to convert — but if you’re getting half as many visitors, it doesn’t matter; the net effect is the same. What you need to do to make this data make sense is to step up volume as well.
So if you’re an e-commerce company saying, “Right, AI is bringing me good, high-intent, high-value traffic” — if it’s bringing you significantly less of it, you need to make sure it’s compensating for the drop in overall volume.
Epic Games — generative concept art and interoperable skins
Grace Robinson: Next, I want to talk about two stories I saw this week. Epic Games gave more visibility on how they’re using generative AI to create concept art for their games, and the digital skins in games like Fortnite. Obviously this had a negative response from some Fortnite users, but I know The Interline has worked with Epic in the past on a report about 3D tools in fashion, so I wanted your take.
On top of this, I saw that Epic is finally figuring out a way for users to take digital skins and use them in other digital ecosystems and other games. This is something digital fashion has been talking about for years, but I think it’s actually one of the only really good use cases for digital fashion. So I wanted your take on that as well.
Ben Hanson: I think it’s useful for us to benchmark this. Do you play video games, Grace? I have, my entire life, and I continue to do so.
Grace Robinson: I used to play The Sims, but that’s it.
Ben Hanson: Fair enough — good to know. So, yes, we’ve done work with Epic in the past, and I’m personally very interested in the video game space, but also the real-time 3D space. It’s always been a passion project of mine. That was the project we did with them last year; The Real-Time Roadmap was fascinating, personally and professionally.
I’ll do these in the order you presented them.
Any behind-the-scenes look into anybody’s creative workflow that has even the merest suggestion of generative AI leads to a lot of backlash. People don’t like the idea that properties they engage with — the live-service, forever games like Fortnite and Minecraft, the sort of things you stake your identity to — have generative AI behind the scenes. I think people feel a sense of betrayal when they see it, particularly younger people.
If you actually go through the trouble of watching the videos of the artistic process here — Epic talk about using their own in-house, packaged-up generative tools, which are called GenMedia Bridge — it’s the best possible version of this I think there could be. You have people drawing things by hand. You have artists saying, “I’ve got to the stage where I have this idea and I just want to refine it to present it,” or “I want to quickly experiment with some iterations.” And that’s when they bring in their GenMedia tools to do it.
For me, if this is something people object to, then there is never going to be an acceptable use of AI in creative workflows from their point of view. And I happen to believe that, in this case, it’s a loud minority, probably, rather than the majority. Now, I do sympathise with the fact that generative AI tools, particularly for artistic purposes, are in a lot of cases trained on unlicensed creative output that people were not compensated for — not talking about Epic’s tools here, but generally. I understand why people are justifiably angry about that.
Equally, this is exactly what the creative workflow will look like in fashion. Every artist has their own red lines. Every artist has their own ideas. Every artist has their own thing they want to do. If you’re visualising your own creative ideas and expressing them through drawing, and then using AI to get them quickly to a stage where you can present them to other people and iterate, to me that seems like the most everyday, prosaic — and also perfect — use case for AI. So it’s difficult for me when people object to that, because I don’t see how you then have a sensible conversation about AI in creative contexts.
The other story, very quickly: we’ve written before that people like to talk about skins in Fortnite and similar games as the frontier of digital fashion. Every time you say, “Well, digital fashion doesn’t work, because it’s not something people want, it’s not interoperable,” people will say, “Well, look at the amount of money spent on skins in Fortnite.” If you want to look like Sabrina Carpenter, go for it — that’s your avenue of self-expression. Great.
Now, it is absolutely true, and we’ve talked about it before, that the commercial incentives usually aren’t there for people to take skins from one game, one universe, one IP, one engine into somewhere else — because the people who develop FIFA or 2K would much rather you bought the skins and kept them there, then bought them again the next year. What Epic are doing here, if I understand it correctly, is that Unreal Engine 6 — the upcoming big point release of UE — will let other games that also use UE6 bring in people’s wardrobes of skins and other items.
It raises a host of stylistic questions. It raises a host of commercial questions. Do Epic want a cut of that? I imagine they probably do, but time will tell. But that’s a shot in the arm for the idea that you can take skins and have them be interoperable and extensible, as long as you’re building on the same technology foundations. That’s actually quite a step in the right direction.
StockX — from gatekeeper to live-shopping platform
Grace Robinson: The next story is about the online secondary marketplace, StockX. They’ve been in the news for the past couple of weeks. This week they announced they’ve opened their first permanent store in Soho — interesting because it’s acting more as a third space than a generic retail store.
On top of this, last week StockX announced they’ve introduced live shopping, and this is the first time their users will get the chance to interact with each other on StockX. On the surface, this just looks like a business levelling up its capabilities, but I was wondering if you think there’s a different kind of platform play going on here.
Ben Hanson: Well, you’ve put the words in my mouth there. If I understand the StockX business model correctly — and I’ve never sold anything through it personally, so this is just professional curiosity — the selling is largely done for you. StockX do the authentication, the listing and everything else, and as a result they become the gatekeepers, the key-holders, the arbiters of authenticity and quality, among other things. When you buy from StockX, you’re not buying direct from a seller. The live shopping move seems to be their watershed moment, and there’s some great analysis from Business of Fashion on this from last week.
It is them saying, “We don’t need to be that intermediary all the time. We’re comfortable if sellers connect directly with buyers, and if the trading happens through StockX — but StockX isn’t directly responsible for all the heavy lifting in the middle.” That does make it a very different platform play. It puts it closer to traditional P2P models like eBay — which I’m sure is a comparison they don’t want. But it’s also probably necessary to crack the livestream side of things they want to do, because the alternative is that you say, “Sellers, send us everything, we do all the authentication” — and then we, StockX, have to stand up an internal team, internal studios and everything else to start doing live shopping.
Live shopping is something you and I talked about last week. This way, they can say, “We’ve opened a physical space, and there’s a content studio in it” — which is true. The physical space in New York is now permanent, and it now has a content studio in it. So they’re becoming a livestreaming platform for sellers to sell directly to buyers, as well as morphing into a different type of intermediary between sellers and buyers for the secondary market.
I think this is part of the bigger picture you and I have talked about before: the secondary market has to offer the same kinds of experiences the firsthand market offers in order to start eating more of that pie. If the secondary market is always the lesser cousin from an interaction and experience point of view, it’s not going to scale to the level people want it to. So I see this as part of that. They’re both smart moves, but I very much see it as: you have to become a different type of company — a much more technology-first company — if you want to eat more of the share the firsthand market currently has.
Beauty tech — Charlotte Tilbury, Deepscent, and a malleable definition
Grace Robinson: The final stories venture more into the beauty space, but they still really stood out to me this week. The first was from Fast Company, about Charlotte Tilbury. The beauty mogul was explaining how they’ve always been a beauty technology company from the start — in terms of the formulas they use, but also the digital customer experiences they employ. I found the term “beauty technology” ill-defined here, so I wanted to ask you about that.
On top of this, I saw an interesting piece that Deepscent Inc. had unveiled an AI-powered digital fragrance content platform — which, again, feels like a tricky label to define. My question is, in brief: across beauty, cosmetics and fragrance, what’s actually happening with technology, and how does it differ from what’s going on in fashion?
Ben Hanson: Yes, and I’m glad you asked. The Interline has slow-rolled into covering beauty more over the last couple of years, and in the process we’ve discovered that “beauty tech” is a malleable definition. Your guess is as good as mine as to what “AI-powered digital fragrance content platform” means. The press release from Deepscent does a reasonable job of explaining it, and I’ll get into that in a moment.
Charlotte Tilbury staking a claim to beauty technology is interesting, because you’ll also have device companies saying the same thing. There’s a beauty tech company based here in Cheshire, in the UK, that does a huge amount of LED masks and similar in-home, technology-first devices that they class as beauty technology. You’ll have people who run clinics saying that’s what represents beauty technology. You’ll have companies like Perfect Corp saying that being able to try on cosmetics virtually is what constitutes beauty technology. Or you’ll have companies like L’Oréal — I believe they did this — with print-your-own foundation, a formulation blend at home from different canisters, and that constitutes beauty technology.
Now, that means it’s broad in the same way fashion technology is, which is a good thing. It also means the way we look at this is that beauty has a different technology opportunity to fashion. You can take tech and use it in product creation, in design and formulation. You can use it across optimising your supply chain, doing the same kind of generative AI imagery we’ve talked about for selling. All of that is on the table for beauty the same way it is for fashion.
What Charlotte Tilbury talks about in that piece — which is written directly by her — is that beauty tech is a unique way of getting the product to the consumer. Wherever the consumer has demand and the beauty brand doesn’t have supply in the form the consumer wants it, they’ll use technology to bridge that gap. I think that’s a good working definition, because you can’t really do technology devices in fashion — it isn’t a category that exists. I can sell you some wearable glasses with AI in them; that just about counts. I could sell you a pen or something. Those are essentially the only devices I can sell you. You could argue phones and so on are wearable technology.
Beauty can sell you hardware. It can sell you technology. In fact, that’s probably a bigger fixture of the way a lot of fragrance and cosmetics companies work than the actual products are these days — that’s where they’re increasingly focusing. That’s an exciting category, because it makes beauty companies much closer to tech companies, in that sense, than fashion companies are.
The Deepscent platform was interesting because it’s deployed on-premise. In that case, it’s fragrances for hospitality and retail — buying up ingredients and combining them at the edge, as it were, live, to create a fragrance profile for your store, your hotel, your bar or restaurant. That’s something beauty can do that fashion can’t. I can’t, in a cost-effective way, sew a t-shirt down the road from you and give it to you in a retail store. I could do some garment decoration and printing, but that’s not really available to me. I can do it with fragrance. I can do it with foundation and formulations. I can do it with other categories. They can live in your house. They can live in the store down the street. They can live in a clinic. There are so many steps of the supply chain that technology can live at for beauty that just aren’t available to fashion. And that, for us, is why beauty tech is interesting. If it were just the same as fashion tech, we probably wouldn’t have started to cover it. We like it because it represents a different category from what fashion can do.
Ben Hanson: Grace, I enjoyed talking these through with you. I’m looking forward to next week.
