The Empty Thrones

AI agents aren't disrupting one part of e-commerce — they're emptying every throne in the chain at once

Estimated reading time: ~9 min (≈13 min in audio)

On a Tuesday afternoon, a shopper says out loud, “find me a waterproof running shoe under $200,” and never opens a browser. No Google search. No banner ad. No product page. No checkout form. An agent hears the sentence, searches across catalogs, weighs the options, picks one, and pays. The entire funnel that took twenty-five years and trillions of dollars to build collapses inside a single spoken request.

The conventional reading of this moment is that AI is coming for payments — that stablecoins and instant checkout are about to rewire how money moves at the register. That reading is real, and it is the least interesting part of the story.

Here is the deeper claim: the agent is not disrupting one layer of e-commerce. It is collapsing the entire funnel into a single request, and each stage it swallows — discovery, advertising, payment, fulfillment, delivery — leaves a throne empty for a new occupant. Call it the Empty Thrones. The reigning kings of online retail are not being killed outright; their subjects simply stopped showing up to court and sent a proxy instead. The thesis is falsifiable, and deliberately not on a clock: if the incumbent owners of each funnel stage keep control of their layer as agent-routed buying scales, and no new entrant captures a single throne, the framework is wrong. The early money is already moving the other way.

The scale of what is being reorganized is not speculative. ARK Invest projects that AI agents could facilitate roughly $8 trillion in online consumer spending by 2030, with the intermediation revenue around that flow growing from about $20 billion today to nearly $900 billion — a 105% compound annual growth rate (ARK Big Ideas 2026, via Yahoo Finance). Juniper Research independently estimates agentic commerce will generate $1.5 trillion globally by 2030 (Juniper Research). That is close to a trillion dollars of intermediation margin, emerging almost from zero, with no settled owner. The useful way to think about it is through Ben Thompson’s Aggregation Theory: value migrates to whoever controls demand (Stratechery). The agent is the new aggregator of demand. Every empty throne is a layer where control of demand is changing hands.

Throne one: discovery, and the next Google

The first throne to empty is discovery. For two decades, the act of finding a product began with a Google query, and Google monetized that moment through search ads. When the agent does the searching, the human never types the query — and the blue link, along with the ad above it, has no one to show.

The throne does not stay empty. It passes to whoever controls the search the agent performs — and, one layer above, to whoever owns the agent the consumer actually talks to. That distribution layer is already in motion: ChatGPT’s share of measurable AI referrals fell from about 89% in mid-2025 to roughly 63% by early 2026, as Claude, Gemini, and Perplexity absorbed the displaced traffic (Higoodie, 2026 AI Search Traffic Report). The single front door is fragmenting into four, and a brand optimized for one of them now reaches a third less of the landscape than it did a year ago. Beneath that contest for the agent itself sits the contest for the agent’s search. This is why Exa Labs matters out of proportion to its size: the company builds a search engine designed for AI agents rather than humans, and in May 2026 it raised a $250 million Series C at a $2.2 billion valuation in a round led by Andreessen Horowitz (Exa) — roughly tripling the $700 million it carried eight months earlier (Bloomberg, via Built In). The market is repricing the discovery layer in real time. The discipline once called SEO becomes GEO — optimizing to be the answer an agent returns, not the link a human clicks. The first throne already has a visible claimant.

Throne two: advertising, where the margin actually lives

The second throne is advertising, and it is the most valuable. If the human no longer sees the ad, the entire apparatus of display and search marketing loses its audience. Yet the spending does not disappear with the eyeballs; it migrates to a new surface. The agent itself becomes the thing that must be influenced, and shaping which product an agent recommends is the form advertising now takes.

ARK’s own decomposition puts the weight here rather than at the register: in its analysis the large majority of agent intermediation revenue comes from agent-directed advertising rather than transaction take rates, and it expects AI-native ads to take share from traditional search advertising with roughly a two-year lag before that share monetizes (ARK Invest). a16z frames the same shift as the end of advertising as we have known it, replaced by permissionless protocols where agents pay for what they need directly (a16z Crypto). And the throne is no longer purely theoretical: OpenAI has opened a self-serve ads manager inside ChatGPT, and more than a thousand brands are already running campaigns through Criteo’s ChatGPT integration, with early AI-referred conversion rates reported at roughly twice those of traditional search (MarketingProfs). The throne is enormous, the most valuable on the board, and the first claimants are only now stepping toward it.