A New Agentic Shopping Standard Reignites Arguments Around Big Tech’s Tightening Hold On Online Selling

Depending on where you sit on the continuum of AI pessimism-to-optimism, the idea that AI agents could become a retail channel shift of a similar magnitude to eCommerce will seem either silly (or at least crazily premature) or inevitable. But whichever of those vantage points you occupy, you won’t be surprised that this new frontier is going to have us relitigating a lot of very familiar arguments about alignment with, or independence from, big tech.

This week’s news is a case in point: by far the biggest headline announcement, where AI commerce is concerned, was the unveiling of the Universal Commerce Protocol, by Google and partners like Walmart, Shopify, Etsy, and Target, at the National Retail Federation “big show” at New York’s Javits Centre. 

The Interline wasn’t in attendance this year, but the developer-centric documentation provided by Google covers all the essentials. It emphasises the open source nature of the standard, presents the huge surface of potential deployments (from Google’s own “AI Mode” to embedded applications in retailer’s own storefronts and checkouts) as evidence that distribution and ubiquity make adoption pretty much assured, heads off criticism by pointing out that UCP integrates with MCP – more on that in a previous analysis –  and puts the consumer benefits front and centre, offering a reduced-friction, low-fragmentation journey “from product discovery to decision, so you can shop the brands you love, with peace of mind, ensuring you get the best value inclusive of your member benefits.”

All of which sounds a lot like splitting a very difficult difference between consumer advocacy – pressing AI agents into service as essentially negotiating on their user’s behalf – and a new kind of eCommerce hegemony, where yet another retail channel ends up being mediated by the world’s biggest technology companies.

That dual possibility feels likely because it is, after all, how every other new strand to the online shopping bow has wound up. Social discovery, social selling, search, livestreaming, personalisation, and plenty of other seemingly open-ended possibilities have, over time, wound up being controlled by an extremely small number of platform holders – most of whom are technology giants, not retailers. 

To be clear: retailers and direct to consumer brands have used all those channels to great effect, and there are armies of brands that exist solely because those channels exist, but none of them are retailer-owned-and-controlled. And the business of selling online is, as anyone who’s done it long enough will tell you, subject to the caprices of Google, Amazon, Meta, ByteDance et al. 

Where the platforms lead, the companies selling through them must follow.

More often than not, the whims of those technology giants wind up aligning with the interests of their customers and users – at least up to a certain threshold. TikTok Shop now dominates that platform partly because people already conditioned to watching influencer content suffused by brand deals were clearly happy to make the short hop to buying directly from those influencers (with close to 40% of Americans now being likely to make a purchase through TikTok). And it also thrives because brands see sales through that new channel as a green field opportunity, additive to existing routes to market. It just so happens that that channel is one that the platform holder is certainly happy to scrape its ongoing share from through referrals, transaction fees, and affiliate links.

And platforms like Google Shopping have grown because they provide retailers and brands with heightened visibility during high-intent searches, as well as putting appropriate products in front of shoppers – all while giving Google new revenue streams and new datapoints about its users to funnel into the ad-bidding market.

When new channels and platforms spring up, for a while at least, everyone wins.

Eventually, though, these channels and platforms have a tendency to morph into a shape that extracts more value, from everyone using and perpetuating those platforms, than they create, but with users and advertisers having limited recourse, and being too locked-in to move on. 

We only need to look at web search to remind ourselves that what began as a noble endeavour (uniting people with content) wound up becoming something of a raw deal for publishers, advertisers, online sellers, and end users – all of whom have had to alter their behaviours to service a closed-book model where the main beneficiary is the middle man, but there are no other middle men left to renegotiate a better arrangement with.

This is a pattern that, based on the platform-holder push that deepened this week, could easily be about to repeat itself with AI commerce. And while it would be easy to shrug and say “well, that’s how these things go,” AI commerce is not intended to be, in the way that social commerce is, a parallel or supplementary channel. It’s intended, provided the big tech model becomes the prevailing one, to take big bites out of the channels that already exist, and to reduce the agency of buyers and sellers in the long run.

And there are, realistically, limited alternatives. It’s clear that AI shopping is a thing that people want, and The Interline is not naive enough to think that either local, open-source AI, or cloud-hosted-but-runtime-encrypted AI models are going to become the dominant mode. It’s already clear that AI apps, and the models behind them, from OpenAI and Google will win this race, and that those apps (and those models, as they make their way into more operating systems) will be where a lot of shopping happens.

And this is precisely why both sellers and shoppers should treat the coming roll-out of UCP with healthy skepticism, because, as famed privacy advocate (and Signal creator) Moxie Marlinspike puts it:

“When you use an AI service, you’re handing over your thoughts in plaintext. The operator stores them, trains on them, and–inevitably–will monetize them. You get a response; they get everything.”

For retailers, the upside of AI-native commerce has been well-articulated: people are keen to shop with AI, and the amount of information that platforms like ChatGPT and Gemini (particularly when that latter model series is hooked into users’ emails and photos, as Google began trialling this week under the banner of ‘personal intelligence’) make them inarguably the deepest reservoir of triggers for personalisation that exist.

But the downsides are, The Interline believes, going under-discussed. The “everything” that retailers part with in this scenario refers to more than just catalogues and inventory. The more the channel mix tilts towards AI, the less control a brand or retailer has over how its brand and its products are positioned and understood, and there is a very real possibility of the entire search, discovery, engagement, and transaction funnel being monopolised by a single company. 

This is the same partially-poisoned bargain that drove so many brands towards Amazon, and then saw such notable ones pull back once they realised how much value was accruing to the marketplace owner, and how much of the relationship with the buyer was changing hands.

This sounds gloomy, but it’s based on prior form. Google is becoming the “first reference implementation” of UCP the same way that it was with the Chromium web rendering engine – an ostensibly open-source initiative that has inarguably benefitted from Google’s resources, but one that’s primarily expressed through a single product – Chrome – that has maintained a greater than 50% market share since 2017, and that currently owns more than 70% of the desktop web browser market.

In practice, that means that the web – and, for our purposes today, eCommerce storefronts – is designed to meet Google’s needs, because the majority of people using the internet are doing so through Google products and services. Getting this right confers advantage to the brand or retailer, sure, but every successful web project (i.e. a site or web application that renders correctly in Chrome) reinforces Google’s grip on the projects that come after it.

The same pattern is now being teed up to repeat itself with AI. Indications are that Google is already winning the LLM battle, so while tools for language models and agentic checkouts are, in theory, both available to anyone, the most likely reality is that they conform, over time, to where the biggest audience is. And when we consider that UCP offers unmatch personalisation potential, better discovery, and definitionally better search and natural language interaction, that audience is almost certainly going to build on Google’s side of the river.

And that’s before we even consider this week’s announcement that the next iteration of Apple’s Siri (which currently processes something in the order of 1.5 billion requests per day) will be powered by Google Gemini, presumably with easy connectors into UCP, and likely with the ability to “assist with more tasks, such as booking travel”.

In other words, the shopping agent in your iPhone is basically guaranteed to use UCP, because the model behind it belongs to Google. And it doesn’t take much thinking to connect that to the white-knuckle grip that Google maintains on its preferential search deal with Apple, whereby Google Search is the default in Safari. Distribution is everything, and the distribution channel for AI commerce is already looking extremely narrow.

Finally, this should all also give consumers pause. The “everything” they trade in their interactions with agentic AI models is, unless otherwise specific, everything that the graph memory layer behind their preferred AI application knows about them. And while Google denied that UCP is opening the door for a new age of personalised pricing – rebuffing a consumer watchdog who made that claim this week – the company’s own blog cites Direct Offers as a headline new capability that integrates into UCP:

“This new Google Ads pilot allows advertisers to present exclusive offers for shoppers who are ready to buy — like a special 20% off discount — directly in AI Mode.”

And while dynamic pricing sits on a fast-evolving legal frontier in the US, the EU, and The Interline’s home market of the UK, where current investigations by the Competitive Markets Authority cast a dim view on ‘hidden’ pricing dynamics, saying that “dynamic pricing can lead to poorer outcomes in certain circumstances, such as when consumers are confused or concerned because prices change rapidly and they are unsure why”.

With generative AI already being such a black box where the mapping of inputs to probabilistic outputs is concerned, it seems inevitable that this implementation of agentic shopping will lead to situations where offers, prices, and communications vary user-to-user in a way that resists external scrutiny.

The Interline wants to make it clear, before wrapping this analysis, that we are not against AI commerce in principle. There’s a huge amount of innovation taking place in this space, and readers need only listen to our Editor’s podcast discussion with the CTO of Daydream – along with an upcoming episode in a similar space – to hear a nuanced, pragmatic debate about where this is all headed. 

Where we do want to caution readers, though, is the willingness to see agentic shopping as a blank-slate opportunity, instead of recognising that it also represents a chance for damaging patterns from previous generations of online and social commerce to repeat.