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.

This week on The Edit, Grace Robinson and Ben Hanson discuss Zalando’s plans to send AI agents shopping at other retailers, Claude Fable 5.2’s leap in remote-work automation, Itsperfect’s AI tools and the spread of copilots in fashion software, ChatGPT’s new business profiles and agents, and Adobe’s experiments in generating new viewpoints and editing images through on-canvas controls.

Watch the video above or listen to the audio below. You can also find the show 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.

THE INTERLINE PODCAST · 28 MINZalando's AI Agent, Remote Labour Automation, Table Stakes AI, ChatGPT Profiles, And MAX Sneaks
Follow the show

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. And together, we have around twenty-five minutes to give you our analysis on the stories that we think really matter.

Ben Hanson: Hey, Grace. How are you doing? I am cold. If you’re watching this on video, you can tell by the, like, cable-knit jumper that I’ve just fished out of the wardrobe. The temperature has suddenly dropped.

Grace Robinson: Yep. It’s cold over here, but still sunny today, which is quite nice.

Ben Hanson: Yeah. I’ve got that going on as well. It’s at minus two or something. I am going to apologise up front. This is an AI-centric episode, but in a fairly diverse way, I think. A few different angles on AI here.

Certainly not all positive, certainly not all negative either. I think we’ll do a good job of splitting down the middle. But yeah, if you’re not in the mood for an AI conversation, this week is not the one for you, I’m afraid.

Grace Robinson: Yeah, that’s very true, but like you said, a lot of different angles. So we can get into the first story.

Who benefits when Zalando shops its competitors?

Grace Robinson: Firstly, we have to talk about Zalando because this week, the brand announced plans to launch its own AI-assisted checkout on the Zalando Extended platform in 2027.

So this service will allow customers to buy products from other retailers without actually leaving the Zalando app, and it will also have AI agents to help complete purchases as well. Now it’s currently in the pilot stage, but it seems like Zalando has kind of created this in order to remain competitive amidst the agentic AI landscape.

However, on the flip side to this, interestingly this week, eBay officially banned Meta’s Muse from browsing their platform. And this comes after eBay previously banned the use of tools like buy-for-me agents or LLM-driven bots to place orders without human review.

So it seems that if more and more brands ban agents, it kind of makes Zalando Extended already obsolete. So let me know what your take is on this and what your thoughts are.

Ben Hanson: Yeah. It’s a weird one. So this is unusual in that the personal agents that exist, so Meta Muse is the obvious one, ChatGPT Dots, Claude, Duet, I think it’s called, Instinct. A lot of these are not fully available in the UK at the moment, although I think ChatGPT Dots is about to drop here.

All of those are retailer-agnostic. Right? They’re created and provided by AI labs, and the purpose of them is to allow people to do a whole bunch of things that they don’t want to do. That seems to be the billing, you know, for a lot of personal agents. It’s a case of you don’t want to cancel your Netflix subscription. Let us do it for you. You don’t want to phone your utilities company for whatever reason. Let your agent do it for you. You don’t want to shop around for the best mobile phone, cell phone plan, or whatever. Let your agent do it for you. Don’t want to do returns. You don’t want to monitor for discounts and sales with your favourite brands. Let your agent do it for you.

The promise of all of that is that it’s an LLM supported by a little Linux computer that runs in the cloud, and it can go do those things for you, either through brute force, through its own browser, or in the case of Muse in particular, but I think ChatGPT as well through partnerships, through like retail partnerships. I know Muse launched with a few in the US. I think Walmart was in there. I think Dick’s Sporting Goods was in there. I think maybe, don’t quote me on this, but Sephora was in there.

But the purpose of those is to cross shop. Right? The purpose of those is to be able to do anything.

Then you have AI chatbots. I’m hesitant to call them agents, but you have AI chatbots that are embedded on retailers’ own storefronts. Amazon has been our go-to reference for this in the past with Rufus, but there’s a huge amount of those out there.

What we haven’t really had is a retailer-owned agent that doesn’t just shop the retailer’s own catalogue, but is envisioned to go out and complete purchases and browse catalogues across other retailers and then to transact afterwards.

Now I can only assume that this is not just a play to remain relevant in the kind of agentic AI era, because as we talked about last week, it doesn’t seem like there’s actually huge volume of that going on when you look at the case study we looked at for Uber bookings, for instance, through agents versus what McKinsey was saying was the full agentic distribution workload.

So it seems like the model is to make money out of this somehow, I assume, because agents are not free to run. That tiny Linux computer in the cloud, fine. It scales. It’s very elastic. It scales to zero when it’s not being used. Scales up when it is being used. They’re not running massive frontier models for these kinds of things. But if Zalando offers this service to all of its shoppers, it is not free to them to offer. It incurs a cost every time somebody uses it. So the only reason that a retailer would do this is because they think they can make money back on it afterwards.

If it was just Zalando’s own catalogue, then that would make sense. Right? Because they make the money through selling the products. They don’t make the money through selling other retailers’ products, unless there’s some real kind of backroom dealings and partnerships that we don’t know about and that aren’t disclosed as part of this release.

So the only rational read on this is that it’s taking the same model that Muse takes, which is a cut of every transaction. That seems to be the philosophy, or I can’t imagine this working any other way.

It’s a weird position for retailers to put themselves in. Like, to me, some of this agentic stuff in general is starting to feel a little bit desperate. When all of these bans are coming into place across eBay, Amazon, other storefronts, and things like that, when you have retailers who will either go, okay, I’m banning external agents, but I’m permitting my own, or they go, I’m not banning external agents because I think this is beneficial to me. It’s weird to go, I’m allowing my own, but I’m also going to send my own out to work on behalf of my customers to go shop from other retailers.

And Zalando doesn’t just sell its own labels, I believe. And Zalando sells clothes from other brands. Presumably, there’ll be a situation where something is available cheaper elsewhere. What does the agent do in that scenario? What is the incentive that’s in place?

Very, very weird story, and one that I’m kind of struggling to wrap my head around. You and I, we do still need to do a bit of investigation into agentic shopping as it becomes more available in the UK. I know we could VPN and try this out, but it’s no use for me buying from US retailers who aren’t going to ship here anyway. So I am committing to doing some more of our own research into this, see how it plays out.

But, yeah, when this story came across your desk and mine, my initial reaction was… and that’s still my current reaction. And I don’t know.

Grace Robinson: No. I agree. It seems to go against Zalando’s business model for the exact reason you said. If they find a cheaper alternative for a Nike shoe on a different retailer’s website, what happens then? So it’ll be interesting to see how this pans out.

Ben Hanson: Well, the bet is that agentic commerce becomes so popular that taking a cut of transactions is lucrative. That’s what all of this is predicated on. But payment processors, there’s a lot of people coming for their share of that, and I don’t know how much of that share is left for somebody like Zalando in that situation. But let’s see how it plays out.

Could general-purpose AI automate fashion work sooner than expected?

Grace Robinson: So next we have to talk about Anthropic’s new AI model, Claude Fable 5.2, and this has actually made headlines this week after achieving a significant leap in its ability to complete real-world professional tasks. So according to Scale AI’s Remote Labor Index, its score jumped from around 18% to 41% in just one model update.

Now the bigger question here seems to be that if AI is becoming twice as capable at completing work traditionally done by freelancers, then how close are we to AI replacing entire categories of remote labour? So talk to me about this story and what it means for fashion.

Ben Hanson: Yeah. And it is an interesting one because I’d never really come across the Remote Labor Index before. This came across our desk as a corollary to the release of Fable 5.2, which, as the numbering suggests, is not like a major milestone release in theory. You know, a point two is like a point upgrade.

It seems like it represents a fairly significant jump in the ability to just ask for finished work. Right? This seems to be where AI labs in particular and, you know, certainly in fashion, but in a bunch of other industries where people are headed is this idea that you just ask for finished work, and you leave it running overnight, and you come back and you get finished work.

Now the Remote Labor Index is made up of a bunch of different tasks. You might think from hearing it that it’s all admin, like putting data into different systems or reconciling spreadsheets or building pivot tables or answering emails or what have you. There’s a video editing component to it. There’s a 2D CAD component to it. There’s a graphic design component to it.

It’s a synthetic benchmark made up of all of the tasks that the average company gives to, like, maybe junior is the wrong word, but a non-senior, non-in-house creative assistant. Like, so it’s a synthesis of remote work in general.

Now we’ve always said, and I think I stick to the idea that for that kind of philosophy to take in fashion, for fashion brands to be able to go, we now have access to, call it, like a drop-in remote worker who can do creative design, pattern making, costing scenarios, assortment planning, channel allocation, whatever. Like, pick a job from fashion.

We’ve always said that that’s something that requires grounding and fashion-specific expertise and potentially fashion-specific, if not models, fashion-specific fine-tunes or deployments and best practices of models, the same way that you have Anthropic doing deployments for the legal sector, for the banking sector, or so on, where they’re taking general-purpose capabilities and making them hyper-relevant for a particular sector.

Stuff like this gives me pause from that point of view because if you assume that the same is possible again, a doubling to call it 80 points, 80%, I wasn’t clear whether it was percentage points or just points, then you very quickly arrive at a place where basically any non-senior work, you can just have done without having to hire anybody to do it, full time or part time or freelance.

And when you combine this with some of the advances in computer use and things that have happened, and you and I have talked before about AI models going and actually using fashion web applications because they have access to web browsers. That’s all you need to do it.

There’s something here. There’s something happening here that I think is starting to challenge my own long-held assumption that the only way fashion would start devolving work and giving it out to remote AI workers would be if they were incredibly specialised.

I still think that’s the case for product development. I still think that’s the case for pattern making. I think it’s the case for material development, like anything that is really specific.

But I think there’s a whole class of work that falls into that. Well, it’s a bit of admin. It’s a bit of creative. It’s a bit of graphic design. It’s a bit of this. It’s a bit of that. That stuff, that, like, general-purpose work might fall to automation way quicker than I thought it would.

This is one of those moments where I’m like, oh, I could be wrong about this. The way I’ve been framing this could be wrong. Let’s see.

Why AI copilots belong inside ERP and PLM

Grace Robinson: So as you mentioned at the start, the next story is also unsurprisingly about AI. So this week, fashion ERP provider, Itsperfect, announced that it’s rolling out two new AI features. One is Ask AI and the other is called AI Photo Studio.

Ask AI is essentially a personalised chatbot that can answer any quick questions someone would usually ask a colleague or have to go through a spreadsheet for. So for instance, this could just be a question like how many units did we sell last week?

The second is the AI Photo Studio, which is essentially an image generation tool that can help with everything from creating AI-generated product flat lays all the way to a fully generated campaign shoot.

Now what’s interesting about this story is that these tools are completely free to use, making AI even more accessible for fashion in a way. So talk to me about this story and what fashion should be thinking about it.

[Editor’s correction: Ask AI is included for Itsperfect users at no extra cost. AI Photo Studio is a paid module with a fixed monthly fee and a one-time implementation fee.]

Ben Hanson: Yeah. I’m going to do this in a way that I hope doesn’t offend the team at Itsperfect. I’ve never heard of it. It’s an ERP platform that I’ve never heard of.

Now I’m not the oracle on fashion and technology solutions, particularly at the minute. There are a lot out there. There’s a lot of new startups and things like that. There’s two things I think people should take away from this.

One is you certainly don’t need to wait for a big enterprise ERP or PLM provider to roll out AI capabilities in order for you to access them.

When we’ve talked before, you and I, or when we’ve said in the AI Report or in any of the one-to-one interviews and stuff that I do on Thursdays, when we’ve said AI has just diffused incredibly quickly, it’s available to everybody, the cost is marginal, the technology underneath it all is basically a commodity, this is exactly what we mean. Like, this is table stakes stuff that you could put in any product now.

If you have a product that contains a lot of information that people put in, that has complex workflows and variables and tables and financial plans and scenarios and this kind of stuff, and you want people to be able to ask natural language questions of it, super easy to do.

I don’t want to trivialise the work that goes into, like, business logic and data structure and things like that to get you to where you can have a structured database to query, but the act of putting in place a natural-sounding, natural-feeling language interface that allows you to just ask questions about your data is, like, it’s table stakes. It’s nothing.

And the same is true of image generation. But every image generation model out there is a full-on commodity. It’s available for pennies on the dollar. It’s available in the dedicated applications, so ChatGPT, Gemini. It’s available in all the multi-model aggregator workspaces, the interview we did with FLORA and so on. Nothing complicated about any of it.

Now, that sounds dismissive. I don’t mean it that way because what I’m actually driving at is that I’ve always said that PLM and ERP, places where there’s a ton of information, are the right places to put copilots. Forget image generation. Anybody can do image generation. They’re the right places to put copilots, the right places to be able to ask questions.

I don’t know about you. I don’t need a Copilot in my email. My email is not that complicated. I can search through it. I can write my own responses. It’s not that helpful. I don’t need a Copilot in Google Docs when I’m writing something, which I do by hand. I will use AI for research purposes and stuff, but I don’t need it there. It’s not beneficial to me there.

If my job was to sit in ERP or PLM and wrestle with data around the clock across multiple seasons and product hierarchies and different categories and different markets and different pricing structures and planned markdowns and all that kind of stuff, and all that information was in one place, I would love to have the ability to ask natural language questions of it.

I think that’s the frontier, not of models, but the frontier of actual use cases. And you also don’t need really expensive frontier models to do it. You can accomplish this with relatively affordable, open-weights models, which is what makes it easy to integrate these things into solutions as well.

So, again, no offence intended to Itsperfect, and hopefully none taken. I think it’s just very interesting that the technology has commoditised and diffused to the extent where any company can start to build these things into their solution, as long as they can justify them afterwards, as long as their customers enjoy using them, and the interface is intuitive and everything else.

That’s why this one jumped out to me. Not because it’s special, not because it’s unique in ERP or PLM solutions or whatever, but because it demonstrates just how much AI has spread out, how quickly it’s spread out, and how affordable and easy it is to embed into things.

Could ChatGPT become the next Google Shopping?

Grace Robinson: The next story is that OpenAI has announced its plans to launch a new feature that will allow businesses to create profiles on ChatGPT, update product information, share pricing and availability, and also see analytics into how often they’re being discovered by users.

Now it’s also introducing Business Agents, allowing customers to ask questions, compare products, and explore brands directly through AI.

Now we’ve spoken a lot in previous weeks about the importance of brands optimising for AI search. But then just last week, we also spoke about how perhaps less people are actually using chatbots as shopping agents than some of the media would like us all to believe. So I wanted to know what you make of this story. Is ChatGPT set to become the next major shopping and discovery platform, or is it not?

Ben Hanson: I don’t know is the answer. Because Google Shopping is one of the biggest shopping and discovery platforms in the world. When you consider the scale of its network, the sheer amount of information that it has on brands and products, it is apparently incredibly successful.

I’ve actively avoided using Google Shopping. It’s just never come up. I’ve always just chosen to go to retailer websites.

The point of that anecdote is to say that you can have something that is, by all objective metrics, really successful, and really does end up leading to sales and conversions, and you can still meet a whole bunch of everyday people in your life who go, what? I’ve never used that. I don’t use it for that. That’s not a thing.

I think ChatGPT has a chance of becoming that. I think it has a chance of becoming the equivalent to Google Shopping over time for two reasons. The first: of all of the AI apps, of all of the chatbot apps that you can install on your phone, ChatGPT is the one that is the most focused on being a consumer application. It’s the one, you know, yes, they have a big enterprise push with Codex and stuff, but ChatGPT on your phone is primarily conceived as a consumer product, designed to do consumer things, the same way that Google Search is. Google has a lot of enterprise products, but search is primarily like a consumer individual product. So I think ChatGPT has a good chance on that basis.

Then if we look at the conversation you and I had, I think it was last week about the amount of search volume that ChatGPT is doing compared to Google. And I think the claim was that it was around a fifth, so 20% of Google’s overall search volume.

Now I did look into that data a little bit more after that conversation, and it seems like a generous interpretation, because I think they frame it as any time you ask a query that does a bit of search or a fan-out or what have you and goes to the web, they count that as web search.

But put those two things together, ChatGPT is doing clearly meaningful search volume. People use it as a search engine. It is a consumer-facing application.

Of all of the AI companies, it has done the best job of getting brands on board. Like, this week and last, there was a big deal about the fact that Prada has a ChatGPT plugin.

I don’t know that your average, like, free ChatGPT user is necessarily a Prada customer. I feel like those are two very different price points to one another, but maybe. Like, it certainly could work from that regard.

And then when you add the other two elements to it, that’s where I think it gets more compelling for brands as well, because OpenAI abandoned their checkout approach.

So at one point last year, they were trying to do full end-to-end transactions with, surprise surprise, the idea that they would cream off a percentage on every transaction that went end-to-end through ChatGPT, i.e. the same model that Muse is going for, and the same model we assume that Zalando is going for.

So they’ve abandoned that side of things, but they have definitely not abandoned being the place that people discover, chat to, and interact with brands.

When you add the Business Agents, which is a weird concept, but again, like, I can currently go to a retailer or a brand store, and I can chat to their own, like, homegrown agent on the web, or I can have a conversation with ChatGPT that involves searching their product catalogue, or somewhere in this weird middle ground, I can have a conversation with a chatbot within ChatGPT that is maintained and shepherded by the brand. It’s a weird sort of setup, but, again, that could work.

And they also seem to be doing quite a lot in terms of giving brands visibility into what people are searching for, what people are discovering, like some measure of the answer engine optimisation, generative engine optimisation, AEO, GEO stuff that brands get from third parties at the moment rather than directly from ChatGPT.

So I don’t know. But OpenAI does a lot of stuff, and they don’t tend to stick with a lot of it for very long. But they seem to be making a very earnest play to become the next Google, and I think that will show up in shopping the same way that it will show up in search.

I don’t know. I’ve just yacked at you. Do you use Google Shopping, like, intentionally?

Grace Robinson: Surprisingly, no. I’ve used it in the past when it’s kind of been convenient, but it’s definitely not my go-to. No. I usually go to retailers’ websites.

Ben Hanson: Yeah. And that’s the case with everybody I’ve ever met, but then when you look at the statistics, you’re like, oh, this is actually one of the biggest discovery platforms in the entire world. Things can fly under the radar, I guess, is what I’m driving at.

When does an AI-edited image stop being a photograph?

Grace Robinson: And then the final story this week is that Adobe has unveiled a new wave of experimental AI tools at its annual MAX conference. Now they’re both exploring how AI can give creatives more control over image making and also video production.

So two particularly interesting examples are Project Sight Span which turns a single image into an interactive 3D environment allowing users to explore new angles and also create entirely new shots.

Then there’s also Project Hands On which lets users physically manipulate elements within an image using AI-generated controls rather than relying on text prompts.

So I wanted to get your take on this story as well. What do you think it means for fashion, and could these tools transform fashion campaigns and photoshoots even more?

Ben Hanson: Yeah. So I’ll be really quick about this one. Adobe are doing some interesting stuff that seems to be annoying the creative community just because of the sheer focus on AI, but they know where their bread is buttered, I think, and they know what they’re doing here.

The big announcement was that the kind of headline entries in Creative Cloud, so Photoshop, After Effects, Illustrator, I want to say Premiere, are all available now for third-party agents to drive over MCP.

So I know you use Premiere Pro. If you want to try saying, okay, Claude, ChatGPT, go edit this video for me. This is what I want to accomplish with it. You can now go do that without having to use computer use for it. You can go and use the cloud application.

It’s interesting. They know where it’s going. I’m still not 100% sure how this whole, like, post-interface thing is going to shake out because these are applications that people very much want to use hands on. Not Project Hands On, but they’re very much applications that people use as desktop apps, and they consider themselves to be expert users.

I don’t know how many people go from that to asking for stuff. Maybe that’s a behaviour that shows up over time.

I suspect it’s more new people who are not desktop-class users, but I think I’ve said before on this show or maybe on the main interview show, Adobe would love, I don’t have any insight here, I’m guessing, Adobe would love to sell thousands and thousands and thousands of cheaper licences versus one desktop-class licence. They would love to sell just core capabilities to people.

So that’s that. Then there’s the other two things you mentioned. So Sight Span and Project Hands On. What they create is so far away from photography now that I don’t really know what to call it.

Because you can take a picture with a camera, and you can use AI to clean it up, or you can generate a plausible-looking campaign photo or a really good product detail page photo or what have you with AI. And people generally seem more accepting of that over time. The results are typically getting a little bit better. They’re already really good.

If you then reshoot it from a different angle, which is what Sight Span is supposed to accomplish, there’s no magic there. You are just generating new angles and new pixels and new content.

If you do the same with Project Hands On. Right? So if you think about, okay, I’ve done a campaign, and in one of these real photos that I took or one of these generative photos that I did, this model’s coat isn’t quite hanging the way that I want it to. I’m just going to move it around, like, after the fact, and then I’m going to shoot it from a slightly different angle. And then I wish it had been windy, I wish that her hair was blowing. You just get so far away from ‘this is an image that represents a real thing’.

And I’ve had so many interviews now on the main show week over week with people saying AI has to deliver on the product promise.

Stuff like this is technically fascinating. Like, absolute wizards work at Adobe on this kind of stuff, on these sneaks and things in particular, and they do show up in products and stuff over time.

This is one where I’m like, fine, if this is what the world wants, if this is what everybody agrees, I’m not going to rebel against it. I don’t have really, really strong ethical or environmental objections, although I do have mild versions of those things.

It’s just like, what is this? At the end, once you’ve gone through that workflow, you’ve shot or generated something, and then you’ve reframed it from a different angle, and then you’ve moved bits of it around, none of those are edits in a way.

They’re all fresh generations. Like, it’s a ship of Theseus situation. How much of that is a photo anymore? Do we care? It’s a very, very weird space.

I don’t know if I’m being grumpy here. Again, technically fascinating, as Adobe’s sneaks at MAX always are, but very weird and slightly like eating its own tail example of synthetic content to me, anyway.

You May Also Like
Who Gets The Most Value Out Of Wardrobe Intelligence?
Read More: Who Gets The Most Value Out Of Wardrobe Intelligence?

Who Gets The Most Value Out Of Wardrobe Intelligence?

The average garment is under-utilised. Platforms that aim to get people wearing what they already own seem like sustainability-minded consumer plays, but the real value of the data they create runs much deeper for an industry fixated on newness.
If The Product Is Everything, Then Product Data Has To Be Everywhere
Read More: If The Product Is Everything, Then Product Data Has To Be Everywhere

If The Product Is Everything, Then Product Data Has To Be Everywhere

A product promise is made up of images, a written description, and a set of attributes. Those elements define eCommerce today, and they are fundamental to a vision for agentic commerce that depends on that data being available and being true.
A model in a tailored jacket with translucent pattern pieces beside a pixelated counterpart, set against a waveform.
Read More: PHP, Generative Tech Packs, And The State Of Clothing Repair

PHP, Generative Tech Packs, And The State Of Clothing Repair

This week on The Edit, Grace Robinson and Ben Hanson discuss the PHP security flaw and Google Cloud outage, The Fabricant’s generative tech packs and sewing-pattern tools, and the state of clothing repair and resale.
What Fashion Can Learn From Beauty’s AI Feedback Loop
Read More: What Fashion Can Learn From Beauty’s AI Feedback Loop

What Fashion Can Learn From Beauty’s AI Feedback Loop

Every fashion brand is pursuing personalisation and questioning how AI changes the way they engage with their consumer, but the most important case studies might be found in beauty, where regulated claims, shopper participation, and outcome measurement over time create a potent loop.