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 widening split between general-purpose and custom-built technology, from Thinking Machines’ case for narrower models to Beauty Pie’s move onto Shopify; the UK’s tighter rules on buy now, pay later and what AI-driven shopping means for overspending; StockX’s new mystery boxes; the rise of AI-native consultancies from OpenAI and Anthropic; and the deeper fit-and-development lessons behind Zara’s viral ‘deadly’ trousers — five stories about where technology genuinely helps, and where it quietly nudges us the wrong way.
This show is available in video format, embedded below, as well as audio, on Apple Podcasts and Spotify.
For deeper interviews with brand and technology leaders, our prestige interview show releases on Thursdays everywhere you find podcasts – and will be coming to YouTube soon.
Subscribers to our mailing list will receive notifications for both shows.
Grace Robinson: Welcome to The Edit from The Interline, the show where we run a 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. Together, we have less than twenty-five minutes to give you our analysis on the stories that we think really matter.
Ben Hanson: Hello, Grace. You’re joined by Ben from a very different location this time. Anyone who’s been watching the video versions will recognise the difference in background, and anyone listening to the audio versions might recognise a slight difference in sound. I’m away from home at the moment — I’m on a working holiday in Spain, and as it turns out, the Wi-Fi in the apartment I’m staying in is terrible. So I’m recording this from a café by a marina. You can’t see the other side of our screen, but there are some really nice boats and yachts over there. So, yeah, you’re getting a very different side of me this week. On that basis, Grace, what do we have news-wise?
General-purpose versus bespoke technology — reading the conflicting bets
Grace Robinson: Yes. There’s been a lot of news this week that highlights the conflicting bets companies are placing on general-purpose technology versus self-built or heavily customised technology — either built just for them or for their sector. On one side, you could say this is an AI story, because Thinking Machines, a company founded by former OpenAI employees, has released new research suggesting that narrower, cheaper AI systems trained on specific domains will start to outperform bigger frontier models.
On top of that, it’s also a general technology story, because this week we’ve seen the beauty giant Beauty Pie shed its own custom tech setup and move to Shopify. And we’ve also seen The Very Group — a UK-based company that sells fashion, furniture and the like — hand over its entire pricing strategy to agentic AI. But on the flip side, I also saw a story from Retail Dive saying that general-purpose chatbots like ChatGPT, Claude and Gemini are actually outperforming brands’ own custom bots, and we’ve seen IBM’s stock drop 25% in a single day.
So with all of these stories, I really want to get your take on what’s going on here. Is general-purpose tech the answer? Is custom tech the answer? Or, as I suspect, is it all a bit more nuanced?
Ben Hanson: It is all a bit more nuanced, so your suspicion is correct. IBM is arguably the name most synonymous with general-purpose technology. Outside of Microsoft, if you were to ask what the general software company is, it would be IBM — and you’d be very hard-pressed to define what they actually do. I think the slump in their stock is representative of the way the emphasis is shifting from companies spending on big enterprise software to the investments being made in the hardware infrastructure sitting behind all of the AI models.
The AI question is interesting on one level, but also kind of inevitable on another. For a long time — pre-ChatGPT, pre-generative AI — the division in AI was between what you’d class as narrow AI and general AI. Narrow being focused on very specific things: playing chess, playing Go, protein folding, that kind of stuff. And they made a lot of advances in that area. The focus then moved to ChatGPT, Claude, Gemini and the like — AI chatbots that can do anything. If you think about the parameter count, the size of a leading frontier model is huge; it’s in the trillion-parameter count and upwards. Parameters roughly correspond to capabilities and knowledge across different domains, so with Claude you can ask it anything from “what should I wear?” to “can you help me code a website from nothing, or build a game from scratch?”
What the Thinking Machines story is effectively saying is: that’s all well and good if you have one tool that’s a one-stop shop, able to do potentially everything — but it’s the perennial jack-of-all-trades, master-of-none situation. If you have one solution that does everything, it will eventually be outperformed by a narrower solution that’s trained on best practices, grounded in data and context, and given specific skills and reinforcement learning set to a specific task. So I think that’s the interesting part.
Now, the move away from custom tech to the likes of Shopify — which I nearly called Spotify there — is something you’ll see happening in a lot of cases. There’ll be companies that invested a lot in building out capabilities they saw as pioneering. They were new to them; they were things they couldn’t find in the open market, at least in the combination they needed. But the open market marches forward. So if you spend five to ten years building something, embedding it, training everybody to use it and integrating it with all your other systems, before you know it what’s available as a cheap subscription or usage-based solution is better than what you built for yourself. So I think we’ll see a lot more of that.
The general-purpose chatbot point is interesting too. What it’s effectively saying is that if you turn up to a retailer’s website through the e-commerce storefront and you want to ask questions about the product — “I’m looking for something that performs like this, that’s suitable in these scenarios” and so on — general-purpose tools like ChatGPT and Claude are better at that than the narrow, focused chatbots deployed on e-commerce storefronts. I find that fascinating, because presumably the narrow bots are better grounded and better anchored in the product mixes, the assortments and the data those brands and retailers have, whereas ChatGPT, Gemini and so on are not — they’re big and general-purpose, not embedded in the same way. I think that’s a reflection of pure model-capability overhang: frontier models are better at asking and answering questions than even the best-grounded examples of those bots.
I think the nuance is only going to get more complicated over time. If you’re a company that sees a lot of value in one-stop shopping — not one AI that does everything, but one technology stack that does everything — I think that’s valid, and especially valid as companies like Shopify really push the envelope on those things. If you do want a custom build, the options are still there. But I think those options look more like custom building with AI, or custom-training AI, than they look like fully building out bespoke solutions.
Buy now, pay later — regulation catches up with the debt vehicles
Grace Robinson: Now, in previous episodes of The Edit, you’ve mentioned the potential risks of different AI applications and AI voice-mode chatbots funnelling people towards purchases that include a buy now, pay later component. Interestingly, there’s been a lot of development on this specifically this week. The UK government has issued new regulation that will require more detailed checks from lending companies like Klarna, for example. We also saw a really interesting survey of 3,000 people from Park Christmas Savings — a company I’ll let you explain, Ben — but basically their survey shows that three quarters of the people who use buy now, pay later services spent more than they intended to. They also found that 35% of them later regretted using these services, and nearly a quarter are still paying off their borrowing more than a year later. So this is both a technology and a social story, and I wanted to get your take on what it means for fashion.
Ben Hanson: Yeah. I’ve long considered a lot of these buy now, pay later fintech companies as debt vehicles, just with a slightly different label on them, and the legislation in the UK is now catching up with that. I’m not the only person who thinks so. There are plenty of ways to spread spending, and what the UK legislation does is apply the same kind of scrutiny to the credit — to the people taking out the borrowing — as it would if you were applying for a credit card. So if I went into a bank and said I want a credit card with a £9,000 limit, they’d say, “Let me run some background checks and make sure this is affordable for you month on month, that it doesn’t exceed your means and isn’t going to put you in a debt trap.” This seems like a logical step, and I think it’s the correct step for regulators to be taking.
Park Christmas Savings is a very interesting company. They actually reached out to us with the results of this survey — I’d never really heard of them before, so I did some looking into it. It’s a way for consumers to spread the spending they do around the peak holiday season through a mixture of store cards, vouchers and a number of other things. They’re effectively another lender, or rather a “spending spreader” in their own right — they’re just not structured like a fintech company, and they’ve been running for a lot longer.
The statistics you mentioned are both interesting and depressing, because spending on things you can’t afford is an epidemic — particularly in the UK, but in a number of other places too. It comes about through a mix of cost-of-living increases and the fact that people simply can’t afford what they want any more. It also comes about because brands and retailers see this as a way of securing growth at a time when spending isn’t increasing — when the share of the wallet they’re competing for isn’t going up, and when they’re competing not just against entertainment and travel but against essentials like food.
The concern we’ve had, and that we’ve raised before on The Edit, is this: you can put a buy now, pay later checkout at the end of a traditional e-commerce journey. You can have it sit there alongside Apple Pay, Visa, Mastercard and so on, and it’s an option — people can use it as they see fit. The difference with AI voice mode in particular, and AI interactions in general, is that they are more persuasive, more captivating, more engaging and more likely to encourage people to make a purchase than a traditional e-commerce journey is.
If you then add the finding that three quarters of people spent more than they wanted to when buy now, pay later was presented to them, and that they’re still paying off their debt a year later, you see an uncomfortable pattern developing — where brands and retailers could see this not just as a way of securing growth because people spread their spending, but as a way of actually getting people to that point in the first place.
Now, somebody argued with me about this on LinkedIn — they said, well, how is this any different from traditional advertising? I see it as different because traditional advertising is primarily passive. There’s some personalisation to it, but looking at an image, reading a piece of text, or scrolling through a storefront or a product detail page feels different to me than having a conversation, in text or in audio, with an entity that is very good at keeping you engaged and very good at trying to get you to the next step of the journey. I might be proven wrong on this in the long run, but to me it feels like one unhealthy hook after another.
StockX’s mystery boxes — smart business or shopping as gambling?
Grace Robinson: Now, a few weeks ago we talked about StockX, and you mentioned that some of the new things they’re implementing — such as creating physical stores and building more peer-to-peer capabilities on their platform — were putting them on a completely different trajectory to the one they’ve traditionally been on. So I’m really interested to know what you make of the StockX news this week. Basically, they’re launching mystery boxes, which are effectively bundled-up verified inventory split into different tiers, with prices — and I quote — “based on current sales data from the platform, reflecting real-time demand and resale value.” Personally, I think the concept of mystery e-commerce is really interesting, but I wanted to get your take on it, and on what you think our listeners should make of it generally.
Ben Hanson: I’m going to start with a question back to you: would you buy something like this? Would you see value in it? The claim they’re making — and I’m sure it’s substantiated — is that the value of the contents of every mystery box is greater than or equal to the cost, so you’re not going to spend and end up with something cheap. From a pure monetary point of view, that seems logical. From a behavioural point of view, I’m interested to see how you feel about it.
Grace Robinson: I think I would do it if I already trusted the brand and had shopped with them before. And if the price was right — if I really felt like I was getting a lot more for my money — then overall, yeah, I think I’d be tempted by it.
Ben Hanson: Okay — I don’t think I would. And I wonder if that reflects an age divide. Coming out of the buy now, pay later conversation, I’m going to sound a bit cynical again, but it’s not great that everything’s becoming gambling. This isn’t gambling strictly, but everything is becoming a matter of playing the odds. You have to read this alongside the rise of Polymarket and Kalshi and so on — none of which are authorised to operate in the UK, as we’re all aware — because people do like playing the odds, at times where spending is concerned and at times where it isn’t.
The interesting part is that the mystery boxes are all made up of what StockX refers to as their verified inventory. When we talked previously, the different models they were trying were the live-shopping, creator-studio stuff, the permanent physical premises they were setting up and maintaining in New York City, and the fact that they were allowing trader-to-trader, peer-to-peer direct interactions. The reason I called that a different type of company last time is that selling peer-to-peer and existing as just the pipes in between is the eBay model — it’s the thing StockX has never been. StockX has always been a company that says: send in the inventory, we’ll verify it, we’ll grade it, and we’ll help to dynamically price it.
You can take the dynamic-pricing component of that, which is data-driven, and that’s great. But what I suspect they have is a lot of verified inventory. To me, this reads like a company that has excess stock sitting in warehouses that it needs to move — and again, that sounds like a cynical read. Behaviourally, though, I think this is on point. I think StockX are doing interesting things; I think they’re right to be going after live shopping, and from a pure business perspective it makes sense. It’s just not something I would invest in, because unless I’ve missed the point, you don’t know what brand you’re going to get, and you don’t know what product category you’re going to get. There’s a lot of uncertainty in it.
Now, the uncertainty is the fun, and the uncertainty is the game for some people. I’m not a gambler; I don’t like uncertainty, so I don’t think I’d use it from that point of view. It is a very interesting tech story, though, because the pricing is dynamically driven by data, and the inventory is presumably packaged, tiered and bundled according to the information they have about sale rates and everything else. Long story short: if you have a lot of products and a lot of visibility into what they’re worth at a minute-to-minute level, you could start to build some really interesting things on top of that.
So I’m watching StockX through two lenses — on one hand, I think it’s a damning testimony to the way people shop; on the other, I think it’s really smart business doing a lot of interesting stuff.
AI and the consultancy model — who still needs the service days?
Grace Robinson: The next story addresses the shift in how big companies are using AI — and I know this is something The Interline wrote about not that long ago, so I’m really interested to get your take on it. The idea used to be that advanced AI models were so capable that businesses could use them without much help. But now companies like OpenAI and Anthropic are spending heavily on advisory services to help their customers put AI into practice better.
Tying into this, the Business of Fashion ran some really interesting analysis this week arguing that AI itself could be replacing consultants. They concluded that typical consulting work — building strategic decks, gathering general research, and one-off projects — is all work that’s now being replaced by AI, and that companies would instead be better off paying consultants to help them optimise their product search for AI, help with AI and agentic shopping, and design new AI workflows within their business. I wanted to get your take on all of this — I know you have some strong opinions about the service-days model. So, what do you think of it?
Ben Hanson: So the strong opinions I have about the service-days model stem from the fact that I’m old-ish — I’ve been around long enough to remember a time when the majority of software projects came with software licences, which you’d expect to be the primary spend. If you were buying a big PLM or ERP, or some other big enterprise software, you’d say the software is where the bottom line sits — that’s what this is going to cost me. In reality, a lot of those projects were heavily indexed on management consultants: either your big third parties, so Deloitte, Kurt Salmon and so on, or consultants employed directly by the technology companies to come in and say, “If you can’t find the value, let us find it for you. If you can’t make this work, let us do it for you.” Those are what we classed as the service days — how many people do you need to bring in, and for how many days? In reality, it ends up being months of custom work.
The reason I have an issue with the service-days model is that it persisted for far too long. It’s a relic of the on-premise, custom-deployment era, but it carried over into the cloud-native, software-as-a-service era because it was the model a lot of companies were built around. In theory, AI is supposed to replace a lot of that. And the Business of Fashion analysis is very interesting, because if you have a big, generally capable model — like we talked about before — and you’ve ever sat down and thought, “Hang on, how do I connect to this? How do I get value from it?”, and somebody’s said, “Just ask it, get it to do it itself, get it to build the skills,” then you’ll have an idea of where that difference comes from.
Now, I don’t think that’s going to prove to be the case, because I feel like we’re right back with AI where we were with traditional enterprise software — people saying, “We can’t find the value. This isn’t translating into measurable, quantifiable metrics for us. So we either need to pull back on software spending, or — what a surprise — we need to get external consultants, or first-party OpenAI, or some other consultant, to come in and help us architect and find that value.”
The two strata the Business of Fashion puts this in are correct, from my point of view. What they’re referencing is that people need consultants to help them build workflows, build logic and build different journeys, because that’s where they’re going to find the value of AI. People do not need somebody else to come in externally and put together a presentation to explain what that value is. They don’t need somebody to come in and work directly on implementations in quite the same way.
It’s going to be interesting to see how companies like OpenAI’s Deployment Company, and Anthropic’s new company — which is called Ode — play out. Ode, again, is forward-deployed engineers and so on. I’m keen to see whether they actually end up being different types of consultants and advisers than what we’ve seen before, or whether, in a year or two’s time, they end up looking an awful lot like the generic, best-practice model of consulting we’ve had in the past.
Zara’s ‘deadly’ trousers — a fit and product-development failure
Grace Robinson: Our final story is a comedic one at first glance, but it does hook into a deeper and more serious problem around the fit and product development of mass-produced products from big brands. It’s a story that’s garnered a lot of attention online: a lot of women are taking to TikTok to share how a very specific pair of Zara trousers is causing them to trip up — scraping their knees, getting bruises, or injuring themselves generally. The trousers are long, slinky and wide-legged, and it’s funny, because I actually had a very similar pair of Zara trousers in the past and I remember tripping up while wearing them. I’ll assume you don’t own this specific pair of Zara trousers — but with your knowledge of product development and fit, I wanted to get your take: how did we land here? How does a big company like Zara get it so wrong when it comes to fit and product development?
Ben Hanson: So there are two bits to fit: there’s objective, scientific fit, and then the subjective side — wear-testing and everything else. You can have a product like this that technically fits. It fits your fit model; it fits your 3D block, if you’re working on that basis, in the sense that it cinches in the right places, it’s loose where it should be, it’s comfortable to wear, and it extends well across the size range and the grades, and so on. You can do all of that and still get fit subjectively wrong, because nobody is actually wearing the thing, taking it out and using it.
Companies like Zara aren’t going to do a lot of field testing. Field testing is traditionally reserved for outdoor wear, running and that sort of thing, where you really need athletes and people to go and test it and make sure it works. I’d say there should be more field testing for general apparel, on that basis, because it wouldn’t have taken more than somebody walking up and down a corridor to identify something like this as an issue. So the quick answer is: more field testing.
The slightly comedic answer I can give you as well is that I don’t own a pair of these trousers — however, going back to our age disparity, big jeans were huge when I was a teenager. JNCOs, and the like — big, baggy, wide-leg jeans. I remember everybody at my high school, when we were allowed out of uniform, had jeans where the entire back hem was frayed, because you just walked on them in your Vans or DC Shoes. And no doubt people tripped over plenty of them — the same people who probably got their hands caught in their wallet chains and everything else. Let’s see if we’re talking about wallet chains in the future, because I feel like things are coming full circle.