Urgency and scarcity tactics on a product page — countdown timers, limited-stock warnings — can increase conversions by up to 332%. Sit with that for a second. The product didn't change. The price didn't change. A clock appeared, and a brain flinched. A third of the transactional power, in that context, comes from making a human feel like they're about to lose something.

Now remove the human.

Gartner's widely cited forecast puts 90% of B2B buying as agent-intermediated by 2028. Separately, McKinsey estimates agents could mediate $3 to $5 trillion in consumer commerce by 2030. Whether those numbers land precisely matters less than the structural question underneath: how much of the commercial web is built on the assumption that the entity at the other end can be frightened, flattered, or rushed?

More than we tend to acknowledge. A 2022 European Commission study found dark patterns on 97% of popular websites and apps. U.S. digital advertising hit $258.6 billion in 2024. Every layer of that spending is engineered for a specific audience: distractible, loss-averse, status-conscious, emotionally available. The commercial web runs on influence. Information is the substrate; persuasion is the product being sold.

We're starting to see what happens when agents actually encounter this architecture. The ACES study out of Columbia and Yale deployed AI shopping agents in a controlled e-commerce environment and found something genuinely strange: agents actively penalized sponsored product tags. Treated sponsorship as a negative signal. Meanwhile they rewarded platform endorsements like "Best Seller" and collapsed demand onto a narrow set of products. The $53.7 billion U.S. retail media market is built on sponsored placement. A buyer indifferent to your pitch is a familiar problem. A buyer that treats the pitch as disqualifying is something else entirely. Persuasion, inverted.

There's a precedent for this kind of structural break. When algorithmic trading replaced human traders on exchange floors, the persuasion layer of financial markets evaporated. The relationship brokers who read body language and exploited panic, the entire infrastructure of human-to-human influence that had organized price discovery for centuries, became vestigial. Spreads compressed. Commissions collapsed. The information layer survived. The influence layer didn't. Lab experiments found that even the remaining human traders behaved differently once they knew algorithms were in the market. The presence of non-persuadable actors changed the whole system, including the parts that were still human.

The commercial web appears to be facing something similar. You can already see the competitive axis starting to rotate: from emotional differentiation toward structured data, transparent pricing, machine-readable product attributes. Kearney describes the emerging imperative as becoming "agent-preferred," which in practice means becoming parseable. Legible to a process that has no patience and no ego.

Most of the conversation so far has focused on where persuasion budgets migrate. The deeper thing worth watching is what the migration exposes. Thirty years of commercial web development produced an architecture so thoroughly organized around human cognitive vulnerability that when you subtract the vulnerability, what remains is a persuasion network that has to figure out, possibly for the first time, what it looks like as something else. In financial markets, the answer turned out to be: smaller margins, faster execution, and competition on operational quality. The commercial web's version of that compression is still taking shape. But the direction suggests a web that has to become what it always claimed to be, and discovering how much less there is to charge for when it does.