The Primary Consumer Is Now an Agent

On May 23, 2026, Axios published a headline that named what many had felt coming for years: the end of the internet’s golden age. The occasion was Google’s overhaul of the search bar — the largest change to search in twenty-five years. But the headline reached further than the news event. It named an era. The internet that felt empowering, where search tools surfaced the web and humans decided what to do with it, has given way to something fundamentally different.

What replaced it isn’t just a new technology. It’s a new audience.

For thirty years, the logic of digital marketing was built on a single assumption: a human is at the end of the pipe. You crafted the message, you sent it through the channel, a human received it, felt something, and decided. The entire apparatus — creative, copy, campaign strategy, channel mix — was engineered to move that human from awareness to action. The channels changed. The assumption didn’t.

That assumption is now wrong. And the research infrastructure of the marketing industry itself is saying so. And one of the world’s most recognized consumer brands has already restructured its entire enterprise around what comes next.

The evidence arrived all at once

Three moves, in the span of months, confirmed what was directionally true and made it structurally undeniable.

Shopify didn’t wait for the market to shift; they helped build the infrastructure for it. Together with Google they co-developed the Universal Commerce Protocol, an open standard for AI-driven transactions. They launched Agentic Storefronts, letting brands appear simultaneously inside ChatGPT, Gemini, and Copilot. And they made the intent explicit: every Shopify store is now agent-ready by default. Their CEO put the internal posture in writing — reflexive AI usage is now a baseline expectation, and before any team can request additional headcount they must first demonstrate why an agent cannot do the work. That is not a productivity policy. It is a declaration about where commerce is going and who they intend to serve when it arrives.

OpenAI moved on the transaction layer. ChatGPT crossed 900 million weekly active users in February 2026. They launched the Agentic Commerce Protocol — open infrastructure for AI-native commerce. And opened product discovery inside ChatGPT directly, integrating Shopify and Etsy catalogs without additional setup. The model is explicit: AI handles discovery, the merchant handles checkout, the human confirms what the agent already decided.

Google rebuilt search around AI Mode, conversational follow-ups, and autonomous agents that monitor the web on a user’s behalf. AI Overviews now reaches 2.5 billion monthly users; AI Mode crossed one billion in its first year, query volume doubling every quarter. The blue links that colored Google’s experience for decades have been demoted to a secondary offering. And at Google I/O in May 2026 they moved past discovery entirely: Universal Cart is a persistent, AI-powered shopping hub that follows users across Search, Gemini, YouTube, and Gmail — monitoring prices, flagging deals, routing to checkout through a protocol that authorizes agents to complete purchases autonomously. Google did not just reinvent how products are found. It is reinventing who, or what, completes the transaction.

Note what happened to the prediction that Google would be displaced by AI. It didn’t come true, because Google absorbed the shift. It had the rails, the data, the trust. The companies that built the early internet are the ones racing to disrupt it, to avoid being the disruptees of the AI age. The platform that invented search reinvented it as intelligence and emerged stronger.

Three companies. Three moves. One direction: the intelligence layer is now the primary mediator between consumers and the products they find.

The brand that read the room

Infrastructure moves are one signal. When a Fortune 50 consumer-goods company restructures its entire enterprise in response to the same shift, it becomes a different kind of proof. At Dreamforce 2025, PepsiCo declared itself on track to become the first global brand to operate as an agentic-AI-first enterprise — agentic in every part of the business by the end of 2026, connecting all operations, all processes, the way it strategizes and commercializes. Internal agents now oversee media buying, creative optimization, and demand forecasting.

PepsiCo didn’t change their product. They changed their operating assumption — the same assumption the marketing industry has run on for thirty years. Reinvention, not optimization. Not channel diversification. Reinvention.

A Fortune 50 brand did not change its product. It changed the assumption the whole industry runs on.

What the research already knows

Gartner, Forrester, and McKinsey reached the same conclusion through different lenses, and the convergence is the point. Gartner predicted traditional search volume would drop sharply as search marketing loses share to AI chatbots and agents — and, further out, its top strategic prediction for 2026 holds that 90% of B2B buying will be AI-agent-intermediated by 2028, routing more than $15 trillion through agent exchanges. (Read the term precisely: “intermediated” counts any deal where an agent shaped the research, not only ones an agent closed — a distinction worth holding onto. The direction is the point, not the decimal.) In a finding that inverts conventional intuition, they expect mass adoption of public LLMs as a search replacement to drive a doubling of PR and earned-media budgets, because the intelligence layer rewards credibility signals that cannot be purchased. Paid placements don’t transfer. Reputation does, which is the outermost layer of the agent surface: whether the intelligence layer can read what you actually are.

Forrester named the budget consequence directly: advertisers cutting display budgets sharply in 2026 as consumers leave the open web for AI-generated summaries and chat interfaces — a mechanism already visible in the majority of Google searches that now end without a click. McKinsey puts the scale of what replaces it in the hundreds of billions of dollars of marketing productivity driven by AI, most of it flowing through agentic systems. And HubSpot’s 2026 State of Marketing report found a clear majority of marketers calling this the industry’s biggest disruption in twenty years.

The research firms that CMOs pay to tell them what’s coming have already told them. The gap between what the research says and what most organizations are doing is the Knowledge Distance problem made visible in real time.

The channels you’re still using

The surface numbers on traditional channels look healthy. SMS open rates hit 90–98%; email reports average opens above 40%. These figures circulate in every benchmark deck, and they are largely misleading. Email opens are inflated by privacy features that pre-load tracking pixels whether or not a human ever reads the message; the real average click rate across industries sits near two percent, and the real SMS conversion — click plus completed purchase — under one.

This is not a story about channel decline. It is a story about where in the decision each channel operates. SMS and email reach humans after the consideration set has already been determined. The agent decided upstream; the message arrives downstream. A 98% open rate is real, and irrelevant to the outcome it purports to measure.

The discovery layer moved. The channels didn’t.

The last human surfaces

Social and YouTube are the exceptions, and it is worth understanding why. Social was never primarily an information channel. It is connection, identity, entertainment, cultural participation, human needs that agents don’t fulfill, and that behavior sustains human-native traffic longer than any information-seeking channel will. YouTube sits at the intersection of search and social in a way nothing else does, serving genuine information intent while keeping the creator dynamics that make social sticky. Agents are beginning to parse video transcripts and surface YouTube content directly; the channel is not immune to agent mediation, only more resistant to it. Both surfaces are on a convergence path. The timeline is uncertain. The direction is not.

The CMO competency floor shifted

The organizational consequence lands hardest at the top of the marketing function. The old floor was built on brand instinct, creative judgment, channel expertise, audience psychology — emotional intelligence at scale, the skills that got most sitting CMOs their roles. The new floor requires something else: data-architecture literacy, structured-content governance, an understanding of how AI systems evaluate and represent a brand, and the ability to build organizations that deliver machine-readable intelligence rather than emotionally resonant campaigns. The CMO who cannot reason about what an agent finds when it looks for their brand. And what to do about it — is working from an incomplete map. Most marketing organizations were built for the old floor, and that investment does not depreciate gradually. In an agent-mediated world, large parts of it become structurally insufficient faster than most leaders are prepared to acknowledge — the human-capacity side of the crossing, arriving at the top of the marketing function.

The opinion being formed

A vendor said something in a meeting that deserves to be named precisely: there will come a day when the models will have formed an opinion about you, and it won’t be fixable. Researchers now have a name for the mechanism — perception drift, made sticky by parameter lock-in. Andrej Karpathy framed the shift plainly: most content is still written for humans instead of models, he wrote, when “99.9% of attention is about to be LLM attention, not human attention.” And once a model has formed its opinion — positive, negative, or indifferent — it becomes exponentially harder to change later.

The mechanism works like this. Models train on vast portions of the public internet — news, reviews, documentation, social. A brand’s digital footprint becomes part of the model’s knowledge, and consistent presence in positive contexts across authoritative sources gets encoded into the weights. Absence, inconsistency, or negative signal gets encoded equally. When someone — or something — asks for a recommendation, the opinion is already formed before the question finishes.

The more urgent risk is not absence. It is misrepresentation. A model that describes your product wrong — stale pricing, outdated positioning, conflated with a competitor — is not making an error it will self-correct. It is forming an opinion that compounds, because every subsequent training cycle that ingests content referencing that output reinforces the distortion. Ignore it long enough and you face a perception problem no campaign budget reaches, because the budget cannot touch the layer where the opinion was formed.

The model forms its opinion before the question finishes. And once formed, no budget reaches the layer where it lives.

Marketing to an intelligence

The game did not evolve. It changed. For three decades marketing was the science of human behavior. You studied emotion, mapped psychology, optimized for the moment a human felt something and acted. Now ask a harder question: who are you actually marketing to?

human audienceagent audience

There are two audiences. The human audience responds to emotion, creative, brand feeling, social proof; it is reached through SMS, email, display, and social, and measured in opens, clicks, impressions. It is the audience the entire industry was built to serve. The agent audience responds to completeness, consistency, structure, and verifiable credibility; it is reached through content architecture, structured data, and API legibility, and measured in citation frequency, recommendation inclusion, and model sentiment. It is the audience most organizations are not yet building for.

Emotion works on humans. Structure, completeness, and verifiable credibility work on agents. Using emotional tools on an intelligent audience does not merely underperform. It is invisible. The agent doesn’t feel your campaign. It evaluates your data, and either includes you in the consideration set or doesn’t, before any human makes a conscious choice. Most organizations are running the wrong playbook on the wrong audience. The channels feel familiar, the metrics look like engagement, but the primary consumer of the intelligence layer that now mediates discovery is not reading your copy. It is parsing your architecture. Two audiences, two ways of being reached. And the boundary between what an agent proposes and what it is permitted to act on is the seam the rest of this work is built around.

This is Part I of The agent surface. It names the shift; Part II — Do You Know Who Is Using Your Product? — shows what breaks in the customer journey when the agent arrives. The whole surface, mapped in five layers: Own Your Surface.