AI agents that buy for customers: what merchants must prepare

L'ADN reports that AI agents are moving beyond product recommendation: they now search, compare and buy on behalf of customers. The outlet frames this as a change in the rules of the commercial relationship, from small merchants to large platforms. The report does not specify which agents, which markets, what volumes or what timelines are involved, and it gives no measured impact. That absence is itself worth noting: on this topic, most available claims are still qualitative.

From recommendation to purchase

The distinction matters. A recommendation engine surfaces options and leaves the decision with the buyer. An agent that executes a purchase takes over the comparison step, the sequencing of criteria and, in some cases, the transaction itself. For a merchant, the surface that consumers once browsed is no longer the place where the choice is made. What remains visible to the customer is the outcome.

Well-established context: for years, digital acquisition has rested on winning attention at the moment of decision, through search rankings, ad placements or a product page. If the decision is delegated, the competition shifts toward being selectable and correctly interpreted by a machine, before any human sees the offer.

What a team can check this week

No one knows yet how fast this becomes a meaningful share of transactions. The report offers no data on adoption. But a few concrete checks are cheap and do not depend on that answer:

• Whether product data (attributes, stock, prices, variants) is complete and consistent across the site, feeds and marketplaces, since a comparable offer that is well structured travels further than a vague one.
• Whether prices and availability are exposed in a machine-readable way, and how often they change without a corresponding update.
• Whether the pages that carry the offer say who the merchant is, where it ships and what the return terms are, in plain terms an automated reader can extract without ambiguity.
• Whether the analytics setup can still tell you where a converted visit originated if the entry point is no longer a human click on a link.

Price clarity deserves particular attention. Monetisation models built on opaque conditions, forced bundles or gated information are fragile when the comparison is done systematically rather than occasionally.

The commercial relationship behind the transaction

Small merchants are not automatically the losers here. L'ADN's framing includes them alongside large platforms, and that is the useful reminder. An agent will not reward brand storytelling on its own; it will reward availability, specification and terms it can act on. That puts pressure on catalogues that were built for browsing rather than for selection.

What the report leaves open is the counterweight: the customers who still want to choose, and the merchants who turn human service into the reason to bypass an agent. Whether that becomes a defensible position is unknown. For now, the practical question for any team is simpler: if an automated reader came to your product pages today, would it leave with the right information?