An agent submits a brief in plain language — over MCP, ACP, UCP, or AP2; Rendreal speaks all of them. Compass resolves the brief into a specification. Artifex perfects the artwork. Rendreal360 routes the order through real production. A physical parcel arrives. No human touches a configurator. No agent learns a Rendreal-specific spec.
Natural-language brief in. Compass resolves product, specifications, and quantity through dialogue, then submits the order with the artwork payload. Speaks MCP, ACP, UCP, and AP2 upstream — the agent calls the protocol it already knows.
The bottleneck for agent-callable physical commerce is not the agent. It is the artwork. Artifex perfects it — from intent and brand assets to the production-ready file, in whatever format the chosen partner needs. Catalog wrappers stop at the SKU. The product an agent actually needs usually doesn't have one yet.
Rendreal360 is the headless fulfillment platform underneath the protocol. AI can write code and design layouts. It cannot bind books, cut signage, or ship pallets. Rendreal360 is the physical backend — real partners, real machines, real parcels. Routes, manufactures, ships, and reports back via webhook.
We start where specs are tightly bounded, fulfillment is operational, and unit economics are known. The protocol generalizes outward from there into apparel, gifting, signage, and packaging.
Flyers, postcards, posters, business cards. Tight specs, fast turnaround, real volume. The cleanest loop to demonstrate the file-creation protocol live, with real production at the other end.
Multi‑page Artifex unlocks books, brochures, and longer‑form printed products. Mega Ingest brings Cimpress and additional supplier catalogs in — sellable through Compass with no UI for the long tail.
Apparel, gifting, signage, packaging. The same protocol — intent, specification, file, fulfillment — generalizes anywhere a product is configured and a partner is selected.
On the rails that don't yet exist for the things agents can't yet buy.
If you watch what AI agents buy today, you see something narrow. They book flights. They reorder coffee. They renew subscriptions. They cancel things. Whatever an agent buys tends to be a thing with a fixed shape — a SKU sitting in a database, a service with a known price, an item the human has bought before. Watching this, you might think agent commerce is small.
We think you are watching the wrong thing.
What you should watch instead is what they cannot buy.
I.Most of what people buy is not a SKU.
Most of what people buy has to be configured before it can be made. A pair of glasses for a particular face. A poster for a particular event. A logo on a particular shirt. A book printed for a particular occasion. A circuit board cut to a particular spec. A counter shaped to a particular kitchen. The interesting thing about this part of commerce is not that it is exotic. The interesting thing is that it is the majority — by volume, by economic surface, by the number of distinct businesses that depend on it.
It has not moved to agents. It has not moved because it cannot.
II.The gap is not that the agents are not smart enough. The agents are getting smarter at a rate that will continue to surprise people who haven't been paying attention. The gap is that there is nothing for the agent to call.
Buying something configurable requires a sequence of protocol-level operations that today exist nowhere as a callable surface. An intent has to be captured and resolved into a specification. A file has to be generated or accepted, and validated against the constraints of the thing being made. Pre-production has to happen — the steps between we know what to make and the machines that will make it can read the instructions. Fulfillment has to be routed to the right partner with the right capabilities in the right geography. Status has to come back through the same chain.
None of this is exposed to agents in any standardized form, anywhere in the world. Every company that makes physical things has solved each piece privately, behind a website built for humans.
A handful of catalog platforms have started to expose their static SKUs to agents — wrapping their existing storefronts behind a callable interface. That is a useful thing. It is not the same thing. A wrapped catalog can sell what already exists. The interesting half of commerce is the things that don't.
It is not a model problem. It is an infrastructure problem.
III.When rails appear in a category that didn't have any, they compound in ways that surprise the people who built applications on top of them. The reason is mechanical, not philosophical. Every integration between an agent and a fulfillment surface is a real engineering cost that nobody wants to pay twice. Agents adopt the protocol that connects them to the most capacity. Production partners adopt the protocol that connects them to the most agents. The protocol that arrives first does not need to be the most elegant. It needs to be early, fair to both sides, and well-designed enough to be worth not replacing.
The layer matters more than any single product that will eventually run on top of it. Products will come and go. Tastes will change. Categories will rise and fall. The layer, if it is built right, will be underneath all of them, doing what protocols do — getting more useful as more things connect.
IV.The agent ecosystem is forming in 2026 in ways that will not be reversible. Tool-using agents have crossed from research demo to deployed system. The protocol stack underneath them is being written this year. Companies that need an agent-callable surface for the things they make are looking for one. There is a window in which a thoughtful protocol layer can establish itself — roughly twenty-four months, by the most credible forecasts — before the category fills with worse implementations done in a hurry. The window is open. It will not stay open.
V.We are building this from the side of the world where things actually get made — not from the side where the abstractions are easy. The problem requires familiarity with how physical commerce works underneath the website: how specifications get negotiated, how files become finished goods, how fulfillment networks split the world between them. The protocol design choices that matter are the ones that come from having lived inside that world, not from having read about it.
VI.We are early. We are deliberate. We are confident about the shape of what we are building and conservative about how much of it we say in public.
Most of what we are doing is not on this site. Most of it will not be on this site for a while. We would rather be three months into building before we have to talk about it than three months into talking before we have anything built.
We're rolling out access to a small group of agent developers, channel partners, and pilot buyers. Tell us where you fit and we'll be in touch when we open the next slot.
We're a small, deliberate team building infrastructure for a category that doesn't fully exist yet.
We're hiring engineers who've shipped agent systems, payments rails, or platforms that move physical goods, and operators who know how things actually get made.