Two Different Angles On How AI Will Shop, And Whether Retail Is Ready

The Interline has spent several months now documenting the ways that AI is changing the web – and the implications that has for product discovery and personalised shopping – but this week we reached a new fork in the road, and  it’s time to ask just how much work  retail is going to be required (or willing) to put in in order to meet a very different model of machine-led shopping.

First up: yes, this is another week where the news is dominated by AI. Realistically, it’s a very difficult topic to ignore. This week, Nvidia hit a $4 trillion valuation (the highest ever for a publicly-traded company), driven by a gold rush of model-building, inference serving and GPU speculation. The hype around AI is clearly not going to subside any time soon – especially with provocateurs like Elon Musk dragging the steady march of new model capabilities into a street fight and, behind it, a culture war.

But beneath the ballooning valuations, new model upgrades, and deserved controversy, something more foundational is definitely unfolding. The buzzword most AI companies are banging the drum for right now is agentic AI, task driven models that can act, not just respond. 

Most people have a surface level understanding of the term, but if agentic AI makes good on its promises, it won’t just change what we use the internet for, it will change who, or rather what, is actually doing the browsing and the interacting with websites… or with whatever amalgam of frontend design and backend re-engineering is going to constitute a website in a world where consumers interact with chatbots rather than scrolling through catalogues.  

We’ve speculated before about the internet’s “new front door”, specifically when rumours first emerged that OpenAI was building an AI-native browser. This week that rumour hardened into something more tangible: OpenAI is reportedly preparing to release that browser soon.

Unlike new image or video models, it likely won’t arrive with the same visual fireworks, but it’s just as consequential all the same. Not least because it signals a rising realisation: that people seem to prefer engaging with chatbots than navigating websites, so one logical play is to design new applications where the chatbot comes along for the browsing and purchasing journey, with extra hooks into personal context that don’t currently exist inside of AI apps like ChatGPT and Claude, even if those apps include search capabilities.

And OpenAI is not the only company releasing an AI browser. On top of experiments like Dia, (a browser that lets you “chat with your tabs” by rolling them up and querying them) this week saw the invite-only release of Perplexity’s Comet browser, which does… something very similar? 

From the fashion and beauty retail perspective, these AI browsers are an interesting experiment, because they represent one of two potential directions that things could go in from here. By bringing the chatbot along for the ride, giving it greater access to users’ browsing lives, the argument in eCommerce will be that these browsers represent a slightly different slant on a trend we’re all already conversant with: humans using websites, with a little help from AI. The other direction, thought, is more foundationally different. And it raises the question of whether 

retailers should re-architect their infrastructure to serve AI-native journeys, or whether they’re comfortable continuing to populate their eCommerce storefronts the way they do today, and letting browsers-injected-with-AI deliver the next turn in consumer experiences.

One thing we can all agree on, right here and now, is that search is changing, although not all at once. Alphabet’s share price has dipped this year, despite owning juggernauts like YouTube and futuristic bets like Waymo, largely due to fears that generative AI could erode its biggest asset: Google Search.

There’s some truth to the anxiety. Chatbot usage is up nearly 81% year over year, and AI tools are replacing alternatives in everything from recipe lookups to holiday planning. But it’s not quite the collapse the stock market might suggest. Search engines still drive 34 times more visits than AI platforms, and Google’s dominance remains near total: 68% of internet activity still starts with search, and 90% of those queries go through Google.

If the internet is a city, Google isn’t just the map. It’s the roads, the traffic lights—the entire flow of movement for most users. Almost everything routes through it.

For now.

Beneath that stability, the landscape is evolving. Tools like OpenAI’s upcoming browser (when it arrives), Perplexity’s Comet, and most standard large language models are beginning to give AI a more active presence on the web, beyond just responding to search queries. They can surface answers, complete basic tasks, and, in limited cases, even carry out transactions. But they do so by interpreting content that was never structured with them in mind.

Rather than interacting with a web built for machines, these systems simulate human behaviour, clicking through interfaces, parsing visual layouts, extracting meaning from context. It’s a workaround, not a solution, and its inherent brittleness becomes more obvious the more it’s used: AI models mimicking human behaviour in a remote web browser is likely to always be a poor proxy for direct human interaction (albeit maybe a more convenient and hands-off one). And moving the browser window from a remote virtual machine to the user’s desktop or phone home screen might change part of the picture, but it’s unlikely to alter all of it.

That’s why these agents often hallucinate, miss key information, or bypass complexity altogether. They’re interacting with systems that simply weren’t designed for them, and in those scenarios the probabilistic tendencies of AI tend to rear their head.  And for fashion and beauty, this means that an AI-equipped browser isn’t automatically going to return truthful, accurate, or actionable information from a website.

Which, unsurprisingly, means that enterprising tech companies are looking to help support brands and retailers in building alternative infrastructure – of the kind The Interline analysed in the MCP portion of the recent “Agentic AI Outlook’ article.

One company working in that space is New Generation, a startup focused on AI-first commerce infrastructure. As co-founder and CEO Adam Behrens puts it:

““AI is fundamentally changing how shoppers discover and evaluate products, but most retail infrastructure hasn’t caught up. Brands have invested in personalised experiences on their own sites, but now they also need to meet shoppers where AI systems initiate the journey—whether that’s in a chatbot, voice interface, or embedded assistant.”

“The challenge is that there’s no consistent endpoint across these environments. It’s forcing brands to rethink not just how they market products, but how they structure their data, present inventory, and enable transactions in a format AI can actually use.”

So should brands take a wait-and-see approach, working off the assumption that “AI browser” as software category will be sufficient for keeping them relevant in any kind of agentic era of the web?  Or do they build structured, machine-readable infrastructure now?

Waiting is, obviously, the path of least resistance. It requires no new investments or architectural rethinks, but it also means surrendering visibility to systems that were not designed to interpret your website cleanly, and to platform holders who do not automatically have brands’ best interests at heart – especially as the real cost of running AI models comes home to roost, and advertising begins to worm its way into chatbot interfaces. As AI interfaces attempt to navigate existing retail sites through the kinds of AI browsers we seem to be getting, brands could then be left hoping that their current structure is legible enough to pass through that filter, and some might not be.

The alternative isn’t glamorous, and ultimately the jury is out on whether it will really be shown to be necessary.  It involves revisiting how product data is structured and exposed. It means making inventory, pricing and fulfilment logic accessible in a way that machines not humans can understand (using codified layers like MCP)and it means making a fairly substantial bet on the idea  that AI systems are the long term future of how consumers will engage with brands. 

If a new phase of retail discovery is indeed emerging, then that second route seems inevitable, and it feels very much as though AI browsers are a short-term dead-end, or, in more charitable terms, an experiment in user interfaces with a limited shelf life. So, just as the creators of enterprise systems are working to get their platforms ready for AI models to call them as tools, it feels likely that retailers will need to do the same. The question is how much time the AI browser stopgap might buy them…