The click was the unit of account for twenty-five years. It is being retired.
Over the past several quarters, a structural change has been recorded in the retail stack. Consumers are migrating away from point-of-sale interfaces — search bars, product pages, checkout funnels — and delegating the transaction itself to autonomous software agents. The terminology attached to this migration is "Agentic Commerce Optimization," or ACO, a phrase attributed to Scot Wingo and amplified in a quarterly document titled "End of Q3 Vibe Check" from a source called Retailgentic. I could not independently verify that report. I am recording the limitation now, at the top, because a ledger with an unverified entry is not a ledger. It is a rumor.
The direction of the migration is legible regardless. The signal lives in the architecture, not the narrative.
Retail has been organized around a single primitive since the late 1990s: human attention converted into a click, and a click converted into a purchase. Every intermediary in the stack — search advertising, affiliate networks, retail media networks, content publishing — monetizes a position along that funnel. Each charges rent for proximity to the human eyeball. The eyeball is now outsourcing the decision.
The essential claim of the ACO thesis is that commerce is shifting from discovery to execution. Under the old model, the customer discovered a product and then executed the purchase. Under the emerging model, an agent performs both — it discovers, compares, verifies, and buys on the user's behalf. The commercial consequence is not incremental. It relocates the point of monetization from the moment of attention to the moment of execution.
The article that surfaced this framing identifies four agent interfaces: answer engines, retailer-operated agents, personal agents, and merchant-owned site APIs. That is a taxonomy of interaction surfaces, not a technical breakthrough. It describes where an agent might connect. It says nothing about how.
I have spent enough of my career inside smart contracts and matching engines to distrust taxonomies that arrive without schemas. Based on my audit experience — four months reverse-engineering the EtherDelta order engine in early 2018, three weeks dissecting the Curve StableSwap invariant in 2020 — I have learned that the failure mode of any commerce system is never the interface. It is the settlement logic underneath. The interesting question is not which of the four surfaces wins. It is what protocol each surface requires, and who controls it.
Consider what "agent-ready commerce" actually demands at the machine level. First, structured product data — a schema the agent can parse without rendering a page. Second, real-time inventory and price endpoints, because an agent that buys against a stale cache will oversell a warehouse in seconds. Third, an identity and authorization layer, because an autonomous agent spending money requires a mandate, a limit, and a revocation path. Fourth, a payment rail that assigns liability when the agent is wrong.
None of these four requirements is a model problem. All four are standards problems. This distinction matters, because the current discourse treats agentic commerce as an artificial-intelligence race. It is not. It is a plumbing race dressed in model clothing.
The plumbing is where the difficulty concentrates. An agent must authenticate itself to a merchant. The merchant must trust the agent's mandate. The payment network must clear a transaction it did not witness a human authorize. Each link in that chain is a place where responsibility can be disclaimed. The framing documents glide past this with a phrase — "agent-merchant trust" — as if trust were a checkbox rather than the hardest unsolved problem in the stack.
There is a second layer almost no one is naming. Pages and APIs will soon serve two distinct audiences simultaneously: humans and agents. Call it the agent experience, or AX. The two audiences have opposed objectives. Human-facing interfaces are optimized for conversion, which frequently means engineered friction — the dark patterns, the pre-checked boxes, the urgency timers. Agent-facing interfaces require structured transparency, deterministic data, and honest availability. A merchant cannot optimize for both without contradiction. The resolution of that contradiction will determine which merchants are selectable by agents and which are quietly filtered out.
The standardization question is the one the optimistic framing refuses to confront. If every platform ships a private agent protocol — and the current incentive structure rewards exactly that — merchants inherit a multi-protocol adaptation burden. The cost of being agent-readable multiplies with each new default surface. The ACO services market that emerges from this is not a moat for merchants. It is a tax on them.
The economics compound the problem. The value pools flow toward whoever controls the agent's default entry point and the payment rail beneath it. The platforms that operate answer engines, the payment processors that clear the transactions, and the private protocol owners capture the execution margin. The merchant absorbs the retrofit cost — structured feeds, live inventory APIs, authorization infrastructure — with no guarantee of being selected. In a bull market, that cost is a rounding error. In the current bear market, it is a line item against a shrinking margin.
The bear market is the variable the ACO enthusiasm consistently underweights. Agent-readiness is capital expenditure. It requires engineering hours, API maintenance, and real-time infrastructure that costs money every time it is queried. The protocols that survive the downturn will be the ones whose execution layer produces measurable conversion, not the ones with the most elaborate pitch. Over the past several quarters I have watched infrastructure projects bleed engineering budgets to service demand that never arrived. The ledger does not lie, it only waits to be read. When the ACO revenue line is finally audited, most of it will be recorded on the service providers' books, not the merchants'.
There is a genuine paradigm shift buried in this, and the bulls are right about its shape. The move from discovery to execution is real, and it is structural. It compresses the funnel. It strips out the intermediaries whose only function was to stand between a buyer and a product and charge for the position. That is a legitimate and overdue correction to a bloated attention economy.
What the bulls get wrong is the beneficiary. They assume the merchant wins when the funnel shortens. They do not. The merchant loses the customer relationship to whatever agent owns the default surface. The intermediary does not disappear. It is replaced by a more powerful one — an agent that sees every price, every alternative, and every margin, and negotiates against the merchant in real time with perfect information. The buyer gains leverage. The platform gains control. The merchant gains a new dependency.
This is the same structural pattern I traced in the OpenSea floor-sale clusters in 2021: when information asymmetry collapses, the party closest to the settlement layer captures the surplus. The agent sits closer to settlement than any merchant ever will.
The forward question is not whether agents will shop. They will. The question is who writes the protocol they shop through, and whether the merchant retains ownership of the customer, the data, and the price. The ledger does not lie, it only waits to be read. The merchants who survive this transition will be the ones who read it before the agents finish writing it — and who insist on an open protocol rather than accepting the first private one offered to them.
The rest will discover, too late, that they optimized for a surface they did not control.