Your eyewear catalog was built for optical software. AI agents are next

Optical frame data is unusually well standardized. A frame's style, color, size, measurements and status move cleanly from the maker into practice software. A shopper's AI agent will ask something that data was never designed to answer: is this one, in this color and size, at this store, today?

DIRECTIONA future direction. No present-tense capability claim.

Explore with AI

What optical already has

Frame data is organized the way a buyer thinks about it. The Frames Data API describes a hierarchy of manufacturer, brand, collection, style and configuration, where a configuration is one style in one size and one color. Each configuration has its own Frame Product Code, the maker's UPC, the eye size, bridge and temple, the lens width, height and diagonal, the color name and code, shape, material and a status of active or discontinued.

Practice-management and POS systems load that data. A barcode or a code fills in the frame when a store adds it to inventory, and the same systems track frames by location and move them between offices. Lab ordering platforms carry the finished job. Try-on tools show how a frame looks on a face.

Frames Data API Field GuideFrames Data (Jobson)
Optical solved frame identity years ago. That's an advantage, not a gap.

Why the variant matters

A shopper doesn't want a style. They want the tortoise one, in a 52, that fits. Every one of those words is a property of a single configuration: one color, one eye size, one bridge. An answer at the style level, “yes, we carry that frame,” sends them to a store that may only have it in black.

Measurements are stated, not estimated. An assistant reading a photo can't tell a 52 from a 54. It can repeat what the maker states, and it should say when a value isn't stated rather than guess.

The unit of an eyewear answer is the configuration, at a location.

Inventory is a separate truth

The catalog knows a frame exists and whether it's current. It doesn't know whether a given store has it on the board. That lives in the store's own system, location by location, and it changes every day.

The brand's store locator is a third source again. It says an account sells the brand, not that this store has this configuration today. Three owners, three systems, and nothing we found that joins them into one answer an outside agent is allowed to give.

What a frame is, whether you have it, and who sells it are three different questions.

Shopping agents are arriving

The general platforms are building the rails. Shopify and Google published the Universal Commerce Protocol for AI agents in January 2026. Google's agentic calling phones local stores to ask about stock on a shopper's behalf, and Google's local inventory program needs per-store availability data.

Each of these assumes the merchant can say what is available where. An independent optical store usually can't publish that today, and none of the optical software vendors we read offers an agent-facing way to ask.

The rails assume per-store answers that most optical stores can't give yet.

Qualifying a frame without guessing

Take a real request: a tortoise acetate frame, rectangular but not too boxy, around a 52 to 54 eye size with a wider bridge. A careful agent keeps the frames whose stated values fit, sets aside only the ones whose stated values contradict the request, and marks the rest as “confirm with the store.” A frame with no bridge measurement on record isn't a no. It's a question.

For a discontinued style, the only replacement is the one the maker names. A frame with the same shape and close measurements is worth suggesting, as one to try on, never as the same fit.

Stated values decide. Unknown stays a question.

Controlled actions

The first useful action is small: ask the store to confirm a specific frame, color and size for a shopper, and route it to a person at that location. Holding a frame, booking a fitting or quoting wholesale each need the business's own rules, an opt-in and a gate in front of them.

Some things never belong to an agent through Lehvel: prescriptions, measurements for a prescription, exam results, insurance benefits or anything about a patient. Those stay with the practice and its systems.

Start with a question a person answers. Add actions only behind the business's own rules.

Why the business layer matters more than any one AI

Shoppers will use ChatGPT, Claude, Gemini, Perplexity, and agents that don't exist yet. Protocols will change. What doesn't change is the business underneath: its locations, its brands, its frames and its stock, each with a source and a time.

That's the layer worth getting right. Built once, it can answer every assistant, and the business decides what each one may say and do.

Build the truth once; let every assistant read it under your rules.

The broader thesis

Eyewear is a clear case of a pattern we see in every vertical. Business Truth tells an agent who the company is. Product Truth tells it what the company sells. Operational Truth tells it what is actually available now. Controlled capabilities give authorized software a bounded way to do something with that truth.

Optical has an unusually strong start on the second. The opening is in the third and the fourth.

01 · Question
What does an AI agent need from an optical retailer, distributor or frame brand to answer a shopper, and where does the existing optical data stop?
02 · Why it matters
Before modeling eyewear we read how frame data actually moves: the frame catalog, practice-management and POS systems, lab ordering, try-on, brand locators, and the new agent-commerce protocols. The aim was to find what already works, so Lehvel reads it instead of rebuilding it, and the exact point where a shopper's agent runs out of answers.
03 · State
DIRECTION A future direction. No present-tense capability claim.
04 · What we built or tested
  • A tested eyewear model in Lehvel's repo: frames as variants of style, color and size; measurements in millimetres as the source states them; status and the maker's stated successor; availability per store answered only from a current reading. No retailer's or brand's record is loaded.
  • Not live: any agent-facing tool or inventory inquiry for an optical business.
05 · Evidence
U.S. optical industry (The Vision Council, Jan 2026)About $69.5 billion; more than 80% of frames and lenses are bought in physical locations.
Frame identity (Frames Data API Field Guide)A Frame Product Code for each style, size and color; status Active or Discontinued.
Catalog coverage (Frames Data, its own figure)Over 600 frame brands.
Store stock (Eyefinity help)Practices with several locations can transfer stock between offices.
Stock checks by phone (Google)“Let Google Call” phones local stores to check stock for shoppers.
Agent commerce (Shopify and Google)The Universal Commerce Protocol for AI agents launched January 11, 2026.
06 · What we learned
  • Frame identity is solved and owned. A configuration code, measurements, status and images exist for the frames in the catalog. Nobody should rebuild that.
  • Stock is a different truth with a different owner. It lives in the practice or POS system, by location, and changes every day.
  • The brand locator, the catalog and the store's stock are three systems with three owners. Nothing we found joins them into one answer an outside agent may use.
  • A frame that looks similar is not the same fit. An agent should offer it as one to try on, and use the maker's stated successor as the only replacement.
  • Unknown is not “no.” A store with no current reading should be “ask the store,” never “out of stock.”
  • The general agent-commerce rails assume a merchant can publish catalog and per-store stock. Most independent optical stores can't today. No optical software vendor we found offers an agent-facing commerce interface.
07 · Limitations
  • The Frames Data field guide we read is dated 2019; the live API listing matched its field names on October 7, 2026, but coverage may have changed.
  • Several items were reported rather than read at the primary source: brand B2B portals, the try-on library sizes, and how often a catalog-to-store gallery refreshes.
  • No search-volume data was available for how shoppers phrase these requests.
  • No optical business's record is loaded, so the model hasn't met a real catalog or POS export.
08 · What changes next
Build it with one independent optical retailer and one frame brand first: the retailer's stores and stock, the brand's styles and successors.
Negative results are published as results. A state on this page is a claim about what is true today, not about what is planned; a limitation is a limitation.
These tools belong to the people who made them. Lehvel reviewed and, where noted, tested them. Nothing here is a Lehvel product, and no performance figure attributed to a vendor has been reproduced by us.

Explore with AI