Agentic Commerce: What Happens When the AI Does the Buying

The most consequential commerce shift in progress is not a new marketplace — it is a new shopper. AI assistants increasingly research products, compare options and complete purchases on a human's behalf. Every merchant assumption about shelves, search and persuasion is calibrated to human eyes. Agents read differently.

What agents actually reward

An agent shopping for "a quiet dishwasher under €600 that fits a 45cm slot" does not scroll your category page or feel your brand video. It parses structured data: specifications, availability, price including delivery, returns terms, review distributions. Merchants win agent traffic on three things:

• Machine-readable truth. Complete structured attributes (schema.org product markup, clean feeds, accurate GTINs), honest availability, and delivery promises expressed as data. Vague listings that convert humans through photography are invisible to spec-driven queries.
• Verifiable trust. Ratings, warranty terms, seller performance — agents weight what they can verify and discount what they cannot. Fake-review-adjacent tactics stop working entirely when the buyer runs statistical checks by default.
• Frictionless checkout. Agent-mediated checkout flows — via emerging standards like agent-payments protocols and delegated-purchase APIs — favour merchants whose checkout is API-clean. Every dark pattern, forced account and surprise fee is a defection trigger for a buyer with zero switching cost and perfect memory.

The near-term merchant checklist

1. Audit your product data as if no human sees it. Missing dimensions, inconsistent units, attributes in image text only — all fatal. This is the same feed discipline marketplaces already pay for, now with higher stakes.
2. Expose live availability and delivery. An agent asked "arrives by Friday?" needs an answer in data. Estimated ranges scraped from a PDP footer do not survive.
3. Keep pricing coherent across surfaces. Agents compare everywhere simultaneously; channel price conflicts that humans never noticed become automatic lost sales — or parity-rule violations.
4. Watch your traffic mix. Agent-mediated sessions are already visible in some categories' logs — headless user strings, spec-heavy query patterns, zero dwell time before conversion. Measure them before deciding what they are worth to you.

The strategic question

Agents compress discovery, which compresses differentiation-by-advertising. If the buyer's agent optimises for verified quality per euro, brand equity migrates toward product truth — the reviews you earned, the spec sheet that does not lie, the delivery promise you keep. Retail media as currently constructed also gets renegotiated: agents do not see sponsored placements the way people do, and the platforms know it.

We are building agent-readiness into commerce clients' infrastructure now — structured data, availability APIs, checkout hygiene — because this shift arrives the way marketplace shifts always do: gradually, then suddenly, then everyone hires at once. Talk to us if you want the audit version of this article run against your own catalogue.