In the final part of this three-part series, Nathalie Lethbridge, Founder, and Jess Carlin, Partner at advisory firm Atonik chart the rise of the ‘two-tier internet’ and its impact on the future of commerce. In this concluding piece, they move beyond diagnosis to consequence, examining what happens when discovery and transaction split across human and agent-driven layers.
When you look at what platforms announced at CES and the National Retail Federation’s Big Show in January 2026, there’s an overwhelming sense of inevitability. The message from every corner of the industry is consistent: agents are coming, and the smart move is to prepare.
But prepare for what, exactly? The two-tier internet doesn’t ask companies to choose between browsing and agents. It asks them to operate coherently across both simultaneously, competing on radically different rules depending on which layer matters for any given customer, product, or moment.
That’s much harder than it sounds.
If you read Parts 1 and 2 of this series, you already understand the framework: the human web is visual, exploratory, narrative-driven. The agent web is conversational, transactional, outcome-driven…
The 2025 holiday season demonstrated that conversational commerce is more than just theoretical now.. AI-driven traffic to US retail sites jumped 805 percent year-over-year during Black Friday. More than half of American shoppers expected AI to influence their holiday purchases. These were mainstream consumers under real purchase pressure, delegating discovery and decision-making to agents at scale.
What’s revealing is why. Holiday shopping compresses everything: timelines, budgets, patience. Decision fatigue peaks. Under that pressure, consumers have a clear indication that they preferred curation to retail overwhelm. That’s a behavioural choice not purely algorithm engagement.
And there’s a new level of urgency. On 16th Jan, OpenAI announced that ChatGPT will show sponsored product placements at the bottom of answers for free and Go tier users.
Eighteen months ago, letting discovery and checkout move upstream of your own website would have been unthinkable. Now it’s a core part of the infrastructure, and it’s generating real transactional value.
This creates a new form of discovery moat. It’s not SEO dominance. It’s not paid media supremacy. It’s being visible enough in agent-mediated recommendations that you’re consistently included in shortlists for relevant intent.. that you are relevant in that conversation.
Structured Data as the New Creative Medium
Here’s the new reality for brands: structured data has become your creative medium.
People call this Answer Engine Optimisation (AEO) but the label matters less than the reality behind it. How you describe your product to machines now shapes how humans encounter, evaluate, and ultimately buy it.
For the past twenty years, creativity lived in copy and design. Structured data attributes, metadata, product feeds were the boring infrastructure. Now, this has been turned on its head.
When an agent tries to understand what you sell and whether it’s worth recommending, it doesn’t read your campaign, it reads your data. Data isn’t supporting creativity anymore. It is the creative.
This creates a new kind of problem for brands. Product information has to work in two directions at once: persuasive enough for a human on your site, precise enough for an agent. It’s not traditional copywriting, and it’s not classic data management. It sits in between the two, and most organisations don’t yet have muscle there.
But the underlying rule applies to everyone. In agent-mediated commerce, data quality is visibility. If your information is incomplete, inaccurate, or inconsistent, you don’t get a second chance. You don’t get outranked, you simply get excluded and no campaign fixes that. You simply disappear.
Why Small Brands Might Win (But Probably Won’t, Unless They’re Strategic)
There’s an appealing idea circulating right now that the two-tier internet levels the playing field for small brands. A ceramics maker on Etsy no longer needs a Google Shopping budget, they just need a product that matches intent.
That’s true — but only under very specific conditions.
Agents don’t rank on brand awareness. They rank on relevance. When someone asks for “a handmade ceramic mug with a matte glaze from a UK maker under £30,” the agent evaluates data quality, accuracy, and fit. A small brand with clean, structured product data and genuine product-market fit can surface alongside, or even ahead! – of much larger players. That’s a real break from traditional search, where money and optimisation expertise often mattered more than merit.
But relevance alone isn’t enough. Agents can’t recommend what they can’t see.
Agents learn from merchant data, reviews, marketplace signals, and transaction history. A brand with no reviews, no presence across trusted platforms, and no historical signal doesn’t just rank poorly, it often doesn’t register at all. Trust signals still matter, and for now, they skew toward established sellers. New brands don’t get a free pass simply because their product is a good match.
Distribution matters too. Agents can only surface products from systems they’re allowed to access. Early OpenAI integrations were announced with Shopify and Etsy in the US, while in India, pilots tied into platforms like BigBasket and local payment rails. Since then, partnerships have expanded across retailers, marketplaces, and payments globally. If your product data isn’t connected to those ecosystems, intent matching is irrelevant, you don’t exist in agent-mediated discovery.
There is, however, a time-bound advantage most brands are missing.
In the early stages of agent-led commerce, agents are drawing from limited merchant datasets. Brands that structure their product information early tend to surface more often simply because their data is usable. That window won’t stay open. As agent-mediated commerce normalises, everyone catches up. Competition reverts to fundamentals: product quality, reviews, price, reliability, and brand strength. So take note brands: the time to act on this opportunity is NOW!
For small brands that move now, the two-tier internet really can be more equitable than traditional search. For those that wait, it doesn’t level the field, it just introduces a new one.
What Gets Lost in Agent-Mediated Commerce
One consequence of agent-mediated discovery is the loss of serendipity. Agents are precise by design: they surface what matches stated intent, not what a shopper didn’t know to ask for. That efficiency raises conversion but narrows exploration.
For brands, this changes the economics of discovery. Incremental purchases, adjacent category exposure, and early-stage browsing matter less inside agent flows. Visibility now depends on being explicitly relevant to declared intent, not on being stumbled upon.
Platform Power: The Real Battlefield
Beneath the noise about new checkout features, smarter recommendations and ads, the real contest is over control of the agent layer itself. Whoever controls the agent controls discovery, ranking, transaction flow, and the economics attached to them.
In Part 1 of this series, we mapped the layout of the land: open, interoperable rails championed by OpenAI and increasingly by Google, versus Amazon’s closed, walled garden approach (although their own agents are allowed to roam the internet freely!).
Agents are already mediating purchase decisions, which means brands and retailers no longer have the luxury of waiting to see which model wins. Because opting out isn’t an option. Agents will access your products whether you cooperate or not.
Agents will access your products whether you cooperate or not. You can choose to participate directly, through open, agentic commerce integrations that expose clean product data on defined terms, or you can be represented indirectly, via marketplaces, aggregators, or scraped listings, with far less control over accuracy, pricing context, and positioning. Either way, the agent layer becomes a gatekeeper you don’t fully control.
For brands and retailers, this creates a key question: where do you build? Do you participate in open-rail ecosystems like OpenAI’s, knowing they might not win? Do you focus on closed-rail platforms like Amazon, knowing you surrender more control but get more visibility? Do you hedge your bets and participate in both? There’s no safe answer yet because the infrastructure itself is still being contested.
The Convergence: Conversation, Video, and Proof
And then to add another layer of complexity, there’s video.
So far, agentic commerce has been largely text-based, but that’s just the first phase. The next iteration collapses conversation, video, and transaction into a single interface.
The underlying infrastructure is already emerging. The Agentic Commerce Protocol is format-agnostic, meaning it can support video as easily as images. Agents are becoming multimodal, capable of watching video, extracting information, and evaluating claims. At the same time, creators are actively searching for monetisation models that connect expertise to real economic outcomes, not just attention.
In that environment, video stops being primarily about storytelling or brand awareness, and it becomes proof. A short clip isn’t just there to entertain, it’s there to demonstrate that a product does what it claims.
For creators, this is potentially transformative. An expert in running mechanics doesn’t just build an audience anymore; they produce evidence. A video explaining overpronation or gait correction can directly influence what an agent recommends when someone asks for the “best running shoe for X.” That expertise becomes monetisable at the exact moment of decision, either through agent platforms or direct merchant relationships. The creator economy and commerce finally converge in a meaningful way.
For brands, the implication is becoming clear. Narrative-heavy video still matters on the human web. But inside agent-mediated commerce, what matters is demonstrable truth: how it works, why it works, and whether that claim can be verified. Video shifts from persuasion to validation.
And for agents, video introduces an entirely new ranking problem. Which clips count as credible? How is creator expertise weighed? Do agents surface multiple perspectives or optimise for a single “best” answer? These decisions will shape what shoppers see, and who gets paid.
In a two-tier internet, video doesn’t disappear. It changes jobs. On the human web, it builds desire. In the agent layer, it becomes evidence.
Preparing: What Each Player Actually Needs to Do
Everyone is adapting to the same two-tier internet, but in January, the stakes became immediate. On January 16, OpenAI announced that ChatGPT will begin showing sponsored product placements at the bottom of answers for free and Go-tier users in the US, clearly labelled and contextually tied to the conversation.
That single move turns agentic commerce from a structural shift into an economic one. Discovery inside conversational interfaces is no longer just about relevance, it’s now a monetised surface. And that creates pressure that lands very differently depending on where you sit in the ecosystem.
For brands, the implications are blunt. You’re now competing on two surfaces inside the same interface: the organic recommendation an agent makes based on relevance, and the sponsored placement that appears alongside it. Both depend on the same underlying truth: whether an agent can confidently understand, trust, and recommend what you sell. Storytelling still matters on the human web. Inside the agent layer, structure, proof, and credibility decide whether you’re even visible.
For retailers, the challenge is existential. As discovery and checkout move upstream into agent platforms, retail sites risk being repositioned from destinations to execution layers: fulfilment engines rather than discovery engines. Clean product data and reliable operations become help, but they don’t guarantee control. When discovery itself becomes a monetised surface owned by someone else, the question shifts from optimisation to dependency: whose rails are you feeding, and on what terms?
Media buyers face a completely different situation. Traditional performance channels still work for the human layer of the internet, but conversational ads don’t behave like search or social. You can’t bid your way into an agent’s answer (at least, not yet!) You can only appear adjacent to it. Success becomes less about reach and more about context, being present at the exact moment an agent is resolving intent. That demands new metrics, new expectations, and a much tighter coupling between data, relevance, and spend.
For publishers and creators, the opportunity is real, but narrower than it looks. As ChatGPT becomes a monetised discovery surface, credible content that functions as proof grows more valuable than content designed purely for attention. Expertise that helps agents evaluate products becomes a critical component of the infrastructure for commerce and transactional capability itself. Audience-building still happens on the human web. Agent-mediated discovery rewards authority, not volume.
What ties all of this together is the same truth: the two-tier internet doesn’t create one solution that works for everyone. What helps a brand gain visibility may deepen a retailer’s dependency. What enables a creator to monetise may weaken a publisher’s editorial model. These incentives don’t align, and they won’t resolve themselves.
The mistake most organisations marke is treating agentic commerce as a channel to optimise. It isn’t. It’s a structural shift that forces hard choices about where you participate, how much control you give up, and which parts of the value chain you’re willing to let someone else own.
That’s the work ahead…and it’s why this moment is less about tactics, and more about strategy.
The Question That Actually Matters
Agents reduce friction. They’re faster, more convenient, and increasingly good at collapsing complex decisions into a single answer. For time-pressed consumers, that’s real value, and it isn’t going away.
But agents are also black boxes. Consumers don’t see what was considered, rejected, or deprioritised, and brands lose control over how they’re represented, competing on data quality and proof rather than narrative or media spend. Platforms, meanwhile, gain extraordinary leverage. Whoever controls the agent controls discovery, ranking, economics, and ultimately, the flow of commerce itself.
That’s why the real question isn’t how to optimise for the two-tier internet as it exists today. It’s what kind of system we allow it to become. Closed rails recreate the worst dynamics of the current web, just faster and more opaque. Open standards offer a chance at competition, but only if brands, retailers, and platforms actively participate in shaping them.
You don’t get to sit this out. Agents will access your products whether you cooperate or not. The only choice is whether you participate deliberately, with clean data and defined terms or get represented indirectly, with far less control over accuracy, pricing, and positioning.
Read more…
Part 1 – The Two-Tier Internet: When AI Agents Became the New Gatekeepers of Commerce
Part 2 – The Two-Tier Internet: How the Holiday Season Showed Shopping Is Changing
