When the buyer is an AI agent

For thirty years the job was to win a human's attention and a human's click. That assumption is quietly breaking. Increasingly the thing evaluating you, comparing you, and pressing "buy" is a piece of software acting on someone's behalf — and it shops nothing like a person.

"Agentic commerce" sounds like a futurist's word, so it's worth being concrete: the rails are already shipping and the behavior is already measurable. This piece sticks to what exists today — the protocols, the launch dates, the adoption numbers — and what they imply for whether a machine can buy from you at all.

This is already happening, by the numbers

The adoption curve stopped being hypothetical sometime in late 2025. Roughly 45% of consumers now use AI for some part of the buying journey, and 39% say they've bought a product based on an AI recommendation in the last six months.¹ The behavior also converts: Adobe found AI-referred visitors complete purchases at a ~38% higher rate than visitors from traditional search, and AI-driven retail traffic grew 805% year-over-year over Black Friday 2025.² Shopify reported orders from AI-powered search up 15× year-over-year through 2025.³

Discovery moves into the AI; checkout stays with you

The most important structural fact is where the agent does its work. After early experiments with checkout living entirely inside the assistant, the model that's settling in is discovery-first: the AI handles research, comparison, and intent, then hands the transaction to the merchant's own checkout.⁴ Shoppers, it turns out, prefer to pay where they already have an account and an order history. That split matters enormously — it means the discovery layer (where you can be included or excluded) and the checkout layer (which you still own) are now separate problems.

The protocol layer: ACP, UCP, MCP

Three standards now wire agents to merchants, each with a different owner and a different reach. They're real, shipping, and not yet interoperable — which is its own near-term tax on merchants.

Owner + role as of May 2026. The "hands off to?" column is the live question: today these are largely separate islands, so merchants supporting multiple protocols see ~40% more agentic traffic — a real, current cost of the fragmentation, not a forecast.⁵

Can an agent even buy you?

Strip away the protocols and there's a blunt precondition underneath: an agent can only transact with what it can parse, reach, and act on. Three gates, and missing any one makes you invisible to the agent regardless of how good your offer is.

The bigger wave is B2B, and it's quieter

Consumer checkout gets the headlines, but the larger shift is in procurement. Gartner projects that AI agents will intermediate $15 trillion in B2B purchases by 2028, and McKinsey estimates agentic models could redirect $3–5 trillion in global retail spend by 2030.⁶ In a B2B context the "agent" is a procurement assistant building a shortlist of vendors against a spec — which makes it a discovery filter with budget authority. That's the version most relevant to anyone selling considered, high-value services.

Where this is heading — flagged as observed direction, not prophecy: vendor-side agents will increasingly negotiate with buyer-side agents, and procurement agents will do more preference-shaping before a human sees a shortlist at all. We don't know the timeline. We do know the precondition is unchanged — an agent can only consider a vendor it can find and parse — which is exactly why the discovery layer is worth measuring now, before the stakes rise.

A perfect checkout is worthless to a buyer who never reaches it. In agentic commerce, the contest moves entirely to the discovery step.

What it means

Agentic commerce doesn't replace the visibility problem; it raises the stakes on it. The same questions decide whether an agent can act on you that decide whether an engine will cite you — can it find you, parse you into structured fact, and resolve you to a real, trustworthy entity. The new wrinkle is the protocol layer and machine-readable product/service data; the old wrinkle — being in the discovery set at all — is your Retrieval Surface by another name, now with a credit card attached.

← All insights

Guide · Published 2026-05-26 · [email protected]