LY Traders · working draft — blog 35 · source: A16Z analysis on the agent economy (Olivia Moore, Nov 2026); Coframe AEO positioning
Traffic to bots.lytraders.com is still overwhelmingly human. That is about to stop being the default. A16Z's latest analysis puts numbers on a shift already underway: Shopify reported AI-driven traffic to its stores grew eightfold year over year, with orders from AI-powered searches up nearly thirteenfold. Vercel reported agent-triggered deployments up 28x in twelve months. Stripe reported agents account for nearly 40% of its documentation traffic and 70% of CLI requests. Agents have moved from answering questions to completing tasks — and the next task they complete on behalf of their operator may be evaluating a product, comparing terms, and starting a purchase.
For a business that sells software, this changes who the website is written for.
For twenty years the discipline was singular: know your visitor, write for your visitor, optimise for your visitor. That visitor is now a pair. A human still signs the invoice, but increasingly an agent does the evaluation — reading the specifications, comparing licence terms, checking compatibility, and shortlisting options before any person opens the page.
These two readers want opposite things. The human wants narrative, reassurance, and a reason to trust. The agent wants structured facts: price, terms, delivery model, technical requirements, refund grounds, support contact. A page optimised only for persuasion is hard for an agent to evaluate, and a page optimised only for data reads as a spec sheet to the person. The websites that win the next distribution wave serve both readers from one document — narrative layered over machine-checkable structure.
The emerging practice behind "generative engine optimisation" is less mystical than its name. An agent evaluating a product does what a careful human buyer does, only faster and more literally: it reads the product specifications, checks whether the claims are consistent across pages, looks for the legal terms that govern the sale, and verifies the facts it can verify. Inconsistency is costly in this regime — a price that differs between page and policy, a guarantee named one way in the FAQ and another in the terms, a product list missing an item. These are not just conversion killers for humans. They are evaluation failures for the machine reader, which resolves conflicts by trusting the more precise source or dropping the vendor entirely.
There is a quieter implication for how products get described. An agent parsing a product page rewards the same qualities a skeptical human does: specific requirements, stated limitations, terms that match reality. A page that says exactly what the software runs on, what the buyer must supply, and what happens if it does not perform is easier for both readers to evaluate than one written in adjectives. Hype optimises for the click. Structure optimises for the decision — and the decision is where the transaction happens.
The A16Z analysis is blunt about where the value sits: not in agent harnesses, but in the unglamorous layer underneath — identity, payments, fraud, and the interfaces through which agents transact. The same is true on the vendor side. The work that prepares a business for agent traffic is almost embarrassingly plain: structured product data in a schema agents can parse, legal pages whose terms match what the product page claims, a FAQ whose answers agree with the refund policy, delivery commitments stated in hours rather than adjectives.
None of this requires new technology. It requires the discipline of saying one thing, in one place, and keeping every copy of it true. Businesses that treat agents as first-class customers are, in practice, businesses whose facts are consistent enough to be trusted by a reader that checks.
The agent-economy wave is usually discussed as a Fortune-500 problem. It is not. A small vendor's site is, in one sense, better positioned than an enterprise's: fewer pages, fewer systems, fewer places for claims to drift apart. The whole catalogue can be made machine-readable in an afternoon. The terms can be verified end-to-end by one person in a morning.
That is the whole playbook for a business of this size: keep the structured facts exact, keep every page consistent with every policy, and let the narrative do its job for the human who arrives last. The agent decides the shortlist. The human signs. A site that serves both readers loses neither.
The next visitor does not blink at the design. It reads the terms.
Discipline is the mechanism.
Sources: A16Z, "The B2A Economy" analysis by Olivia Moore (November 2026) — Shopify, Stripe, Vercel, Human Security figures as reported therein; Coframe's AEO positioning for the machine-reader framing.