Storefronts for agentic commerce

A year ago “AI on a storefront” meant a chat widget in the corner. Today it means something different: an agent, sometimes the user’s own, sometimes their wallet’s, sometimes a comparison shopper, landing on your page and making a buy decision without ever rendering it visually.

That changes what a storefront is for.

What an agent actually wants

Crawlers parse markup. Agents reason over structured facts. The lift from one to the other is smaller than people think, most of the work is already in your schema.org Product blob if you took the time to fill it in.

A useful checklist:

Field	Why it matters to an agent
name	Disambiguates from other SKUs
sku / gtin	Universal join key across marketplaces
price	Filterable
availability	Lets the agent skip out-of-stock variants
shippingDetails	Filterable on delivery-by date
hasMerchantReturnPolicy	Lowers perceived risk
aggregateRating	Tie-break vs comparable SKUs

If you only fix one thing, fix shippingDetails. Agents optimising for “arrives by Friday” need to know your cutoff window, your handling time, and the carrier transit estimate. Without it, you get filtered out before the price even gets compared.

Render once, expose twice

The pattern we use across Nomu storefronts:

import { Product, ProductSchema } from "@/lib/commerce";

export default async function ProductPage({ params }) {
  const product = await Product.get(params.slug);
  return (
    <>
      {/* The page for humans */}
      <ProductHero product={product} />
      <ProductDetails product={product} />

      {/* The same source of truth, exposed as schema.org JSON-LD */}
      <script
        type="application/ld+json"
        dangerouslySetInnerHTML={{ __html: JSON.stringify(ProductSchema(product)) }}
      />
    </>
  );
}

The trick is that ProductSchema(product) reads from the same object the UI is rendering from. If a field is missing from the JSON-LD, the field is missing in the data model, not because someone forgot to add it to two places.

What changes about copy

Humans skim. Agents read.

That means the SEO-stuffed product description (“Premium quality candle made with love by our team in Brooklyn from sustainably sourced beeswax”) loses twice: a human skips it, and an agent extracts very little from it.

A short factual description plus structured additionalProperty entries outperforms both:

const product = {
  // ...
  description: "Beeswax candle, 8oz, sandalwood + vetiver. 48-hour burn time.",
  additionalProperty: [
    { name: "burn-time-hours", value: 48 },
    { name: "wax-type", value: "beeswax" },
    { name: "scent-notes", value: ["sandalwood", "vetiver"] },
    { name: "container-material", value: "ceramic" },
  ],
};

The human still gets the headline. The agent gets a queryable record.

The take

Agents are not a new traffic source. They’re a new reader. You don’t need a separate storefront for them, you need the same storefront to be correct as a document and as a page. Build that once and you’re future-proofed for whatever comes next.

We’ll go deeper on the agentic protocol layer in the next post. If you want to be notified, the RSS feed is the canonical channel.