Can AI agents scale a physical-product business?

The research hypothesis

An AI agent should be able to expand a physical-product catalog from a bounded product contract, with people setting policy and handling exceptions instead of assembling every SKU.

Catalog growth is repeated coordination. Each product needs a subject, product data, copy, price, customer imagery, a manufacturable asset, and a fulfillment identity. Those parts have to agree when the catalog grows from ten products to hundreds or thousands.

World Patches became the test bed. A location brief enters a generator; code produces a supplier file, a store preview, and a catalog record. Fifty-nine products were enough to expose repeatability problems, manual work, and the point where a research pipeline could become a business.

The guide image

The first blocking technology problem was manufacturing geometry. “Leave room for bleed” is clear to a print designer and vague to an image model. The generator attaches the same 985 × 985 px PNG to every request. Three high-separation colours divide the file into a safe area, a bleed ring, and a sacrificial outer square.

The request sent to the model

Gemini receives the guide PNG and the product brief in one request. The brief supplies the location, landmark, visible labels, border colour, permitted palette, and explicit rules for each zone. Google documents image-plus-text input for Gemini image generation and provides a batch API for larger jobs. Batch capacity does not remove the need for acceptance tests.

contents = [zone_guide_png, product_prompt]

image_config = {
  "aspect_ratio": "1:1",
  "image_size": "1K"
}

candidate = generate(model, contents, image_config)
manufacturing = resize(candidate, 984, 984)
save_png(manufacturing, dpi=300)

Model output is treated as a candidate. Deterministic code resizes the first returned image to 984 × 984 px and writes 300 dpi metadata. A second pass masks the square image into a circular store preview and overlays the supplier border. The square manufacturing file keeps artwork outside the visible circle because those pixels are needed at the cut line.

Results from the first 59 products

We audited the repository on 10 July 2026. File presence and raster geometry are consistent across the catalog. Visual acceptance is still a human step.

Catalog records

59

Manufacturing files present

59 / 59

Store previews present

59 / 59

Manufacturing dimensions

59 / 59 at 984 × 984 px

Manufacturing density metadata

300 dpi

Automated visual acceptance

Not implemented

Audit scope: catalog JSON, generated files, raster dimensions, and density metadata in the World Patches repository. It does not include a new print run or an end-to-end paid order.

Failures observed during generation

The prompt became specific because loose versions failed in specific ways. Lofoten returned a tiny inner scene, buildings over the ring, a green ring, and incorrect inferred subtext. Faroe and Geiranger also needed individual reruns for boundary violations and scene clutter.

There is also a supplier-spec question to close. Prodigi’s patch page lists JPG or PDF at 300 dpi; its general Print API documentation accepts JPG, PNG, or PDF. The current pipeline submits PNG. A sandbox order and supplier confirmation should settle the product-specific contract before unattended generation.

What we would add before 10,000 assets

A batch endpoint makes generation faster. It does not make bad files cheaper to find. At catalog scale, acceptance must be expressed as tests with a saved result for every asset.

Physical samples remain part of the test plan. Pixel checks can prove that a file matches a template. They cannot prove legibility on fabric, colour under the supplier process, edge placement after cutting, or whether the rendered material matches what arrives.

From research pipeline to venture

The backend was built to test agent-run catalog expansion. Patches made the test bounded: one supplier SKU, a fixed print template, location-specific subject matter, and repeatable product fields. Producing a coherent 59-product catalog created enough working inventory to launch World Patches.

The storefront uses conventional Next.js, Stripe Checkout, and Prodigi order plumbing. Its value to the research is operational: every accepted asset must keep its product identity through catalog publication, payment, and the supplier request.

The fulfillment path also has an unresolved retry defect. A supplier-order failure is currently acknowledged to Stripe with HTTP 200, which marks the webhook delivery as successful. Stripe documents retries after non-2xx responses. Durable retry and reconciliation must move outside this request before the order path can be trusted.