What is Agentic Commerce?

Why It Matters

A shopper who used to browse a category now asks an assistant, and the assistant returns a shortlist. That is a different contest from the one storefronts were built for: the product page, the photography, and the merchandising all still matter, but only to the products that made the shortlist, and the shortlist is assembled from structured data.

The practical consequence is that a retailer can be excluded without being out-competed. Nothing about the product changed; the feed simply did not answer the question the agent asked.

What Changed This Year

Three things landed at once. Google published an open standard for agent-mediated commerce covering discovery through post-purchase, alongside agent payments and its existing agent-to-agent and context protocols. OpenAI and Stripe’s protocol, originally the plumbing behind native checkout, shifted to feeding product catalogues into an assistant for recommendation. And Shopify opened its checkout to browser-based agents, letting them read a checkout, change it, and complete the order with the buyer’s authorisation.

Retailers are not unanimous about it. Some platforms and brands are actively blocking agents from purchasing on a customer’s behalf, which is a legitimate position and also a decision about who owns the customer relationship.

Where It Breaks

Readiness is the first break, and it is widespread: most retailers have no agent-facing manifest at all, which makes them invisible rather than merely unoptimised. Structured attributes, variants, availability, and policy pages are the entry ticket.

The second is policy legibility. An agent cannot respect a returns window it cannot read, so the terms that live in a PDF or a footer image are terms the machine will guess at.

The third is action authority. Once an agent can transact, it can also err: a wrong variant, a duplicate order, a refund issued twice. Whether the agent acts on your storefront or on someone else’s, the same question applies: what may it do on its own, and how is that undone?

How Flytebit Handles It

We treat the agent-facing surface as a first-class one: catalogue attributes, availability, and policy pages structured so a machine can read them, and the same product data readiness work that improves human conversion. Where an agent acts on your own store, its authority is written down and enforced as a bounded action, with a reversal window and a record. The industry application is on our E-commerce & Retail page, and the control design is our AI governance and risk work.