Two of our own stores check out for AI agents today — an assistant can read the whole shelf, open a real cart, and pay: on-chain in USDC, or by handing a normal Stripe checkout to a human. Live since August 2026. You can point an agent at them right now.
You can't sell agent-ready commerce to a client if you've never run it yourself. The protocols break in places no article mentions — and the only way to find those places is to own the store the machine is checking out of, hit every wall first, on our own money, before a client's store is ever on the line.
So we rebuilt two of our own stores for agents. Not a slide, not a demo on a staging URL. Real catalogs, real carts, real settlement, real things that get packed and shipped. We paid for the mistakes so a client doesn't have to.
That changes the shape of the conversation. When a client asks whether their store can be made shop-able by AI, the answer isn't a proposal full of conditionals. It's: we already did ours — here's what broke, and here's how we'd handle yours.
A sticker shop where an agent settles a real payment in USDC on Base with no account, no card form, and no human until someone physically packs the sticker. It reads the full catalog as clean markdown, opens a cart against the same system the website uses, and answers an HTTP 402 challenge to pay. Independently scored Level 4 by Cloudflare's agent-readiness scanner — not our badge, their crawler.
A full catalog store where an agent opens a real cart and gets the true total — items, shipping, and tax, all computed server-side, not guessed. Then it chooses a rail: pay the agent-native way over x402, or hand a normal Stripe checkout to a person for the final yes. Most real commerce wants that human confirmation, and this is what it looks like done right.
We started with stickers and print-on-demand on purpose. At $6.99 a unit, the things that usually make commerce scary — fraud, chargebacks, returns — barely bite. That's exactly where you want to be standing the first time an autonomous agent moves real money through your checkout.
The store advertises itself to agents through machine-readable declarations at its domain — a commerce descriptor, an agent card, and an OpenAPI schema — so an assistant knows the shop exists and how to talk to it before a human ever links it.
Ask for Accept: text/markdown and the catalog comes back as
clean, structured text generated from the real product data — not the
rendered HTML page stripped of everything the model needs.
The API answers with an HTTP 402 challenge. The agent signs a gasless USDC authorization and settles on Base through a non-custodial facilitator — or, when a human should hold the pen, it's handed a normal Stripe checkout instead.
A real order lands in the same pipeline as a human's. Someone packs a real thing and ships it. The agent was a customer, not a science project.
Designodin isn't a store, so agent-readiness here isn't a checkout rail — it's
content. This site serves a clean markdown version of itself to any agent that
asks (Accept: text/markdown), publishes an OpenAPI descriptor, and
lists our services in Stripe's Agentic Commerce catalog so an assistant can
discover and buy them directly.
What we didn't do is just as deliberate: no x402, no on-chain settlement, no commerce protocol theatre on an agency site where it would be cargo-culting. The stores needed those rails. The agency doesn't.
That judgment — the right standard for the right kind of property — is the actual deliverable. And the only reason we can make the call cleanly is that we've run both kinds and watched where each one earns its keep.
We're not pitching a category we read about on LinkedIn — we run it in production. Tell us about your store and you'll get a straight read: what it would take to make it shop-able by an agent, which standards actually fit it, and whether it's even worth doing yet.
