The attention economy took thirty years to build. It is being dismantled by the one customer it was never designed to persuade — an AI agent that cannot be flattered, seduced, or worn down by repetition. The next platform war will not be fought over human eyeballs. It will be fought over which company the machines trust.
Every advertising dollar ever spent has rested on one quiet assumption: that a human being is on the other end of the decision, and that human beings can be influenced by repetition, emotion, social proof, and design. Google, Meta, and thirty years of digital marketing were built on refining that influence to the decimal point. That entire discipline is about to meet a customer for whom none of it works. When an AI agent buys your flight, negotiates your insurance renewal, or restocks your kitchen, it does not see the banner ad. It does not feel FOMO. It does not remember your jingle. It reads structured data, compares it against other structured data, and picks the best answer. Thirty years of persuasion infrastructure is about to be pointed at an audience that cannot be persuaded — only informed.
This is not a future scenario — it is a trend already visible in 2026. Letter 66 (The End of the Attention Economy) documented the early cracks: Generative Engine Optimisation replacing Search Engine Optimisation, AI agents that don't click ads, and a $258.6 billion advertising market being redistributed rather than destroyed. Letter 69 (Legible to Machines) went further, arguing that the winning brands of the next decade would be the ones structured as machine-readable, API-ready assets rather than emotionally compelling stories. Both letters were describing the same underlying shift from two different angles. This letter names what both were circling: the emerging battleground is not attention. It is trust.
An agent making a purchasing decision on a human's behalf needs exactly four things, in this order: verified price, verified availability, verified quality signal, and a track record it can check against other agents' experience. None of these are things a thirty-second video ad can supply. They are things a structured, queryable, continuously-updated data layer can supply — and right now, that layer barely exists. Product data is scattered across a thousand incompatible formats, review systems are gameable, and there is no equivalent of a credit bureau for "how trustworthy is this seller's data, according to other machines that have transacted with them before."
"Google won the last platform war by being the place humans went to ask a question. The next platform war will be won by whoever becomes the place AI agents go to check whether an answer is true."
Whoever builds the canonical trust layer — the infrastructure that AI shopping and service agents actually query before completing a transaction — inherits the gatekeeper position that Google Search and the Meta ad auction have held for two decades. This is not a small prize. It is the successor to the entire $258.6 billion digital advertising market, except the customer is a machine, the currency is verified structured data instead of attention share, and the moat is trust infrastructure instead of engagement algorithms.
The pieces are visibly assembling, even if no single company has yet won the position. Perplexity's Comet browser (Letter 115) is explicitly built as "the surface where an AI agent starts a task and finishes a purchase" — a direct statement of intent to sit at exactly this layer. Stripe and other payment infrastructure players are building agentic commerce protocols that let AI agents complete transactions autonomously, which requires exactly the kind of verified-data handshake this letter describes. OpenAI's Operator and similar agentic browsing tools face the identical problem from the demand side: an agent tasked with "find me the best flight" needs a trust layer to know which price is real. None of these companies have solved it yet. All of them are colliding with the same unsolved problem from different directions, which is usually what a real emerging market looks like before anyone has named it.
Three categories of company have a genuine shot, for different reasons. First, the agentic browser and assistant layer (Perplexity, OpenAI, and whoever else builds the interface agents actually operate through) — because they see the most transactions and can build reputation data as a byproduct of normal use. Second, payment and identity infrastructure (Stripe and similar players) — because verified transaction history is the single hardest trust signal to fake, and payment rails already sit at the choke point where money actually moves. Third, and least obvious: any company that already runs a large, structured, hard-to-game verification system for an unrelated reason — credit bureaus, KYC/AML infrastructure providers, even certain logistics and customs data providers — because trust infrastructure is expensive to build from scratch but cheap to extend from an adjacent, already-built system.
"The company that wins the trust layer will not look like an advertising company, because trust cannot be sold the way attention was sold. It will look like a credit bureau, a payments network, or a customs database that quietly became the place every AI agent checks before it spends your money."
Three real obstacles keep this from being a settled thesis rather than an early one. First, the chicken-and-egg problem: a trust layer is only valuable once enough agents query it and enough sellers feed it verified data, and nobody has cracked that cold-start problem yet at scale. Second, the open-standard risk: if agentic commerce protocols standardise on an open, non-proprietary trust layer — the way email standardised on open protocols rather than a single gatekeeper — the economic value accrues to everyone and no one, which is good for the world and bad for any single investment thesis built around capturing it. Third, regulatory risk: a trust layer that determines which sellers AI agents favour is a chokepoint regulators will eventually scrutinise the same way they now scrutinise search ranking and ad auction design, and that scrutiny could arrive well before the category has even matured commercially.
The attention economy is not dying because AI is replacing advertising with something flashier. It is dying because its next customer cannot be advertised to at all. Thirty years of persuasion infrastructure is being quietly replaced by a much smaller, much less glamorous requirement: verified truth, structured so a machine can check it. Nobody has built the definitive version of this yet. The company that does will not look like the advertising giants it replaces — it will look more like a credit bureau that machines trust with real money. That is a strange, unglamorous, and genuinely enormous business to be early to.
Founder, NextGen Economics · Bangalore, India · July 2026
Sources: Letter 66 (The End of the Attention Economy) · Letter 69 (Legible to Machines) · Letter 115 (Perplexity & the Airtel Question) · Letter 113 (Stargate) · Public disclosures on agentic commerce protocol development from Stripe and OpenAI.
Not investment advice. All investments carry risk including loss of capital.