The Edit is our weekly show, where Social Editor Grace Robinson quizzes editor-in-chief Ben Hanson on some of the most significant fashion and technology stories from the past seven days.
In this edition Grace and Ben dissect the state of live shopping through Whatnot’s half-billion-dollar raise, the market for MCP-access intelligence licences, what happens when AI chatbots mistake brand language for provable sustainability credentials, the unnerving echo of 2008 in data centre lending rules, and what Jo Malone is doing in Fortnite.
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 quickfire 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 25 minutes to give you our analysis on the stories we think really matter.
Ben Hanson: Hey, Grace. Good to see you again. It’s really hot in the UK and a whole bunch of other places at the moment. If you’re watching this on video and we collectively look a little bit frazzled, that’s probably the reason. I’m rocking my linen shirt. For anybody listening to the audio, it was the best combination of smart and relatively airy I could find. It’s about 31 degrees in this room, I think.
Live shopping consolidation: Whatnot’s $545m raise and the battle for the live commerce market
Grace Robinson: A big headline this week was that the live shopping platform Whatnot has raised an additional $545m in Series G funding, bringing their valuation to $20 billion. This is apparently the biggest raise in live commerce. So I’m interested to get your take on this news, and also if you think Whatnot is in its own lane in terms of live shopping since its mission is really to help small businesses grow, or whether it’s competing with the likes of TikTok Shop. As a reference point, the cosmetics brand P Louise set and then beat their own live shopping record on TikTok Shop this year, selling £2.7 million in 14 hours.
Ben Hanson: We’ve covered live shopping before — I think it was StockX’s expansion into it — and we’ve talked about the sheer scale of the live streaming market. This is interesting to me because it’s clearly a very big race, and clearly a company with a lot of support and belief behind it. They claim the live shopping market in the US, the UK and Europe is worth an estimated $20 billion, roughly in line with their valuation. They currently have about 60% market share of that.
It’s difficult for me to parse this. I’ve read through Whatnot’s own state-of-live-commerce report for 2026, and a bunch of the analysis, and it’s tricky to understand what the definition of live shopping is here. Is it its own vertical in the sense of dedicated apps that exist solely for live video selling, or does it encompass the entire social media landscape with all the platforms that have added transactional or referral functionality? TikTok Shop is the most prominent, but there’s a fair amount of commerce integrated into Instagram as well. You and I talked last week or the week before about how every social network is a selling and shopping destination.
I wouldn’t pretend to know how that’s going to play out. Unless I’m missing something, there’s no way that Whatnot owns 60% of the global potential market for people buying things through social commerce. For one thing, this excludes China, which is a major live shopping destination. It has to, I think, exclude TikTok and exclude Instagram. And if you stack this alongside some of the record-holding live shopping events — the cosmetics brand P Louise here in the UK set and then beat their own live shopping record a couple of times, shifting about £2.7 million of merchandise in a 12-to-14-hour window on TikTok Shop — there’s a huge amount of scale here.
The question for me is: if it’s dedicated applications, is that where individual sellers, micro-brands and smaller brands go? That seems to be Whatnot’s positioning. Or does it start to pull some of the larger brands who are setting records and shifting huge volumes through the big established social channels? I think the thing to watch here is not necessarily whether live shopping is big — it’s very clearly big — but who owns it. Is it likely to be owned by dedicated companies that just do shopping with a community aspect, or by social channels that are increasingly making shopping a big fixture of it?
For me, as someone who is not a social media user in my personal life, if people value selling with a community element, which I think is Whatnot’s edge, the cynic in me says eventually you bring more and more community onto the selling and the selling app becomes a social network in its own right, converging with TikTok and Instagram. But unambiguously, week over week, we continue to see a huge amount of money, interest and brand investment flowing into live and social commerce. I have an interview on the Thursday show coming up in the next couple of weeks about exactly this: the creator economy, its size and scope, and how to measure the value of investment in social commerce.
MCP access to intelligence: Edited and Adobe signal a new class of headless software licences
Grace Robinson: The next story is about a shift that could be happening with intelligence providers. We’ve seen this week that Edited has released Edited MCP — a server that users can connect to Claude Desktop or their own agent, which pulls from Edited resources and data without the user having to be on the platform. We’ve also seen something similar with the Adobe plugin for ChatGPT, which brings Adobe tools into ChatGPT Work and Codex. My question is: are we going to see more intelligence providers and technology companies do this? There seems to be a choice between selling people licences of your own software that contains your intelligence, or selling access to that intelligence wherever people want it. Should those be different business models or the same thing?
Ben Hanson: This is a really interesting one. MCP — Model Context Protocol — just to ground that for anybody not deep in the AI weeds: think of it functionally as a slightly enhanced version of APIs and tool calls. Edited is a retail intelligence platform, I think is the way they would put it. It’s not functionally equivalent to a forecasting service like WGSN; it’s something more tightly aligned to brands wanting to analyse the competitive landscape and make fairly quick decisions based on a huge number of variables.
The intelligence packaged into platforms like Edited is the value. If you are a subscriber user of Edited, you are there because you want data to back up the choices you are making about products, assortments, channels, prices, markdown strategies and so on. That doesn’t mean the Edited interface doesn’t matter 100%. What it does mean is that Edited and a bunch of other companies have started to reckon with the idea that the interface matters less than what it gives people access to.
As I understand it, Edited MCP allows people to pull all of that data into their AI agent harness of choice, provided they are existing Edited users or customers. There is no secondary tier between full platform users and MCP users. The same is true of the Adobe plugin for ChatGPT, but with a slight twist: you can do some basic functionality through the natural language interface without an Adobe account, but if you want access to generative tools — Firefly and so on — you need a pre-existing Adobe subscription.
What I very strongly believe we are going to see is a new class of licences for intelligence platforms like Edited, but also creative tools like Adobe and supply chain management. If there is value in the data a platform contains and the questions it can answer, and that value exists independently of the interface, we’ll see a new class of licences that are just MCP access. That feels inevitable. Would Adobe rather sell a hundred $33-a-month Creative Cloud subscriptions, or sell $10-a-month licences that give people access to core features through ChatGPT or their agent of choice? You would rather have the latter, just in pure monetary terms.
The thing is you don’t have to choose. I think we’ll see much more stratification: people who are desktop users and people who are MCP headless users of these products. Intelligence and creative tools are a good first step, but you can stretch the definition through to transparency platforms, traceability and lifecycle impact assessment, line planning, forecasting. There is so much you can put into an MCP that doesn’t require a full licence, and it significantly expands the audience for your platform in the process. We’re going to see a ton more of this: some cheap and easy attempts where people just make a few features available through ChatGPT, and then more serious attempts where companies realise they have a much bigger addressable market than they thought.
AI sustainability claims: When LLMs repeat brand language as gospel, and what watermarking means for greenwashing
Grace Robinson: Next I want to talk about an interesting piece from Business of Fashion this week. The piece reveals that AI chatbots could be making inaccurate sustainable fashion recommendations. As you know, LLMs are pulling from simply the words most frequently used to describe fashion brands rather than doing independent investigation into the truth of those words. Business of Fashion partnered with Quilt AI to analyse responses from ChatGPT, Claude and Gemini. When prompted to suggest clean, plastic-free and sustainable brands, they found that though the AI did suggest some well-known sustainable brands like Patagonia, it was still clear the results were simply based on language and therefore not totally accurate. Allbirds was among the results, which we know isn’t true.
This story was interesting enough, but it also came across our desk at the same time as an announcement from Anthropic: that all text generated with Claude in future will carry an invisible watermark, to comply with the EU AI Act. A lot of people have very strong opinions about this. What do you think? Is there a story here about what AI looks for in written content, and how companies should be thinking about how they create that content in the first place?
Ben Hanson: I’d always encourage people to go read our AI Report for a whole bunch of reasons. For this one, I’m going to pull from a piece I wrote that I think I titled ‘Turning Down the Temperature in the AI Discussion’. In there, I made two recommendations. First: everybody should, at some point, sit down and interact with a proper frontier model fully connected to all of their enterprise technology estate, a Claude Opus or whatever it is at the time of recording with connectors to everything you do, and you will quickly find that if LLMs are given the right context, the right information and the right grounding, they can give accurate and trustworthy answers.
The second thing people should do is run a really small local model with no internet access and ask it some questions about something they care about. That gives you the unique experience of being lied to with supreme confidence — receiving a reply that sounds believable but is based entirely on inference, not on any kind of source of truth. I suggest those two things because LLMs don’t know anything. There’s no investigation, no verification. To determine whether Patagonia is sustainable, you or I would do a reasonable audit of reports and word-of-mouth and a whole bunch of variables. An LLM does it based on the references to Patagonia that exist within its training data and whatever information it can gather at the point of inference.
Most LLM queries default to fairly quick web search because ChatGPT, Claude and so on are optimised for giving people quick answers. If I ask ‘recommend a sustainable menswear brand’ and the answer takes 15 minutes because it involves a big research fan-out, I’m just going to do something else. That use case isn’t there. So they pull from the most readily available and most heavily search-ranked content: a combination of reviews, recommendations, brand language, backlinks. The usual stuff that makes up the internet. None of it is independently verified or audited.
What this allows companies to do is make claims in writing — because the written word is the most important part here — either explicitly or accidentally through AI-generated copy, that is not backed up and doesn’t have to be backed up for LLMs to repeat it as authoritative. I saw a really interesting experiment on Substack where someone made a fake deodorant brand, put together a bunch of basically meaningless sustainability claims — magnesium-free, I forget the specifics — and wrote a Substack article, put a website up, and it took not a terribly long time for ChatGPT to start recommending that product as a sustainable deodorant. That’s the prime example: there is no independent verification.
To link this together with our previous story: if you were to pair ChatGPT or Claude with an MCP tool call to an index where people have actually done inspections, audits and certifications, all rolled up into an intelligence source, then great: you get more authoritative answers. But we live firmly in the era where you can just say things. You can claim to be something you are not, and LLMs will pick up on it and repeat it as gospel.
On the Claude watermarking piece: I know for a fact that a lot of companies do product detail page copy and consumer-facing written content through AI. It will be based on product attributes, technical specifications and design intent, but it doesn’t take much for an accidental certification or wording that would class as greenwashing to slip through. If LLMs pick up on it, before you know it you have a product in the market that doesn’t meet the claims you are making accidentally, and those claims get repeated verbatim ad infinitum by LLMs thereafter. The upshot: don’t trust LLMs to do this kind of research on your behalf, whether you’re an investor or a consumer.
Data centre lending risks: The SEC relaxes rules on asset-backed securities for AI infrastructure
Grace Robinson: I’ll keep my question brief on this one because I know this is an area you’ve been watching, and I know you’ve covered it in a recent long-form interview. I saw news this week that the US government has relaxed some of the financial rules that have helped de-risk investments since the 2008 financial crisis — specifically for data centres. What’s happening here?
Ben Hanson: I’ll keep the answer brief too, because this is a prime example of my cynical attitude manifesting in the real world. The specifics: data centre owners and builders can now issue what the SEC calls asset-backed securities, derivative products that people can invest in and purchase, pegged to an underlying asset class. The reason this is tied to the 2007-08 subprime mortgage crisis is that this was precisely the same mechanism — mortgage-backed securities. Everybody go watch The Big Short. The essential idea is that you can do a lot of speculation built on top of an underlying asset class if you believe that asset class has a future.
The 2008 financial crisis took a lot of people in the financial world by surprise because mortgages had always been considered a very stable asset class. There was a lot going on under the surface about rating different tranches of investment classes higher than they should have been. The underlying assumption here is that data centres represent that kind of stable asset class. The SEC is claiming that, and therefore data centre owners should be able to offer these derivative products on that basis.
I’m not sure any of that is true. Time will tell. Data centre usage is definitely on the up, but there is a lot of rebellion against the build-out of data centres, particularly in the US but also in the UK. This is an example of what people talk about when they talk about an AI bubble: a house of cards built on the assumption of future use, and the build-out of capacity to meet assumed and forecasted demand rather than guaranteed demand. If you then stack derivative financial products on top of that, you are making a whole series of layers of assumptions. It doesn’t take a great deal for that house of cards to collapse. The implications for AI companies and for the general economy are pretty profound. It’s something that CFOs across fashion should be watching.
Fortnite experiences: Jo Malone, Epic’s asset sales, and the strategy behind gaming activations for heritage brands
Grace Robinson: The final story is the news that Jo Malone has launched a fragrance with an in-game experience in Fortnite. The activation is to promote their new Sea Salt & Bergamot cologne, and it’s a two-week in-game experience that apparently gives consumers a more personal and immersive way to discover fragrance through exploration and storytelling. We’ve talked about gaming being the best way to reach younger generations in a world of social media bans, but in all honesty Jo Malone feels to me more like a mum brand than something that belongs in Fortnite. Do you think this kind of play to target younger consumers through gaming is viable for brands that already have a pretty clearly defined audience? And I know there was wider news from Fortnite’s parent company Epic this week, so tell me your thoughts on that too.
Ben Hanson: Your read on Jo Malone and my read are basically the same. It’s not a young person’s brand and hasn’t been for a while. I think it’s purposely cultivated that audience. Very clearly this is the result of some market investigation and reporting that says there is a target demo we can go after here, and the best way to reach it is to work with an agency. This is not an official Epic collaboration; it’s part of the Fortnite creator platform. They’ve partnered with somebody to build out this experience with the goal of reaching the kind of audience that you and I have talked about before: people who are not going to be on social media for much longer in the UK and elsewhere, when bans come into force for under-16s.
I don’t see any reason this play couldn’t work. The whole of fashion, fragrance and beauty is riddled with stories of brands that became a little fusty and did a successful job of appealing to younger demographics. It’s also filled with examples where that didn’t work. So I can’t make a cultural assessment on whether this will play out. What I find interesting is that Fortnite is what they turn to, because Fortnite has had a drop-off in engagement and players. And Epic has had a bit of a time of it recently: they’ve conducted a bunch of layoffs. They acquired ArtStation and Sketchfab about five years ago, and they have now offloaded those as of this week.
Those purchases were made, I think, on the assumption that real-time 3D was the future of the web. That was the bet. The original vision for something like Jo Malone in that world would have been: come to our own website or across other channels and interact with real-time 3D in some way. Now the bet is very clearly, from Epic’s side and from brands’ side, to focus back in on the core experience and the core IP where people actually are. I think we’ll just have to see if Fortnite engagement and numbers go back up the way Epic is predicting. In the near term, I think you’ll see more companies taking the agency-build route rather than direct partnerships with Epic and similar platform holders.
The last thing I would encourage people to do is listen to my interview from last year with Marcus Holmström of The Gang, a company that does these builds for brands in Roblox in particular but also Fortnite and other platforms, because it contains a lot of insight into the justification behind these kinds of choices. So that is a brand story, a technology platform shift and an acquisition story all in one — which is always an interesting way to bring us to a close.
