For two decades, buying online meant a human doing the work: searching, opening tabs, comparing, adding to cart. Agentic commerce is the shift where an AI agent does that work on the shopper's behalf. Discovering options, comparing them against the person's stated needs, and increasingly completing the purchase itself.
Ask an assistant "find me a fragrance-free moisturizer for eczema under $30 and order it," and the agent fans the request into searches, assembles a shortlist, weighs the criteria, and, where the plumbing exists. Checks out. The person reviews a recommendation, not a search results page.
The real change: from browsing to machine selection
The consequential part isn't the convenience. It's who's doing the selecting.
When a human browses, every brand on the shelf gets a glance. When an agent selects, it builds a small consideration set first, and if you're not in it, you're eliminated before a human ever looks. Discovery moves upstream of your website and upstream of the human eye. The "shelf" becomes the agent's shortlist, and the agent assembles that shortlist from what it can find, parse, and trust about your products.
In agentic commerce, the buyer you have to win over first is a machine. If it can't find or understand your product, the human never gets the choice.
Why brands should care now
Three reasons it's not a "later" problem:
- The interface is already moving into chat. Shopping surfaces inside AI assistants, product cards in answers, and emerging checkout standards mean the path from "question" to "purchase" is collapsing into a single conversation.
- Machine selection rewards different things than a pretty PDP. An agent doesn't see your hero image or your brand film. It reads structured product data. Attributes, specs, availability, price, and the sources that describe you. Gaps that a human shopper forgives, an agent simply can't resolve.
- Readiness is uneven, so it's an edge. Most catalogs aren't structured for machine selection yet. The brands that get ready early get chosen disproportionately while the field catches up.
Two layers: visible, then buyable
It helps to separate two things:
- Visibility: does the agent find and recommend you at all? This is the AI-search/GEO layer: be retrievable, well-structured, and trusted enough to make the shortlist.
- Transactability: once recommended, can the agent actually buy from you? This is the newer layer that checkout protocols are being built for.
You need both, in order. There's no point being purchasable if you're never surfaced, and no point being surfaced if the agent can't transact. The next articles in this series cover the protocol making transactability real, and how to get your catalog ready for both. It's also the frontier of what Buffy Intel watches: not just how AI describes your brand, but whether your products are positioned to be chosen.