Variants Are Where Agentic Shopping Breaks

    Size, color, count and flavor are where agent purchases go wrong. Here is why variants are hard for AI shopping, and how JENNi is building toward a fix.

    October 23, 20264 min read

    Size, color, count and flavor are where agent purchases go wrong. An agent that buys a costume without asking the size isn't helping anyone.

    Product variants in AI shopping are the quiet failure. The agent finds the right product, the price looks fine, the checkout works, and the shopper still gets the wrong thing. Nothing crashed. The question just never got asked.

    We want to be straight about where JENNi stands. We're building toward letting a shopper pick size and options at checkout. It isn't live today. This post is about the problem and the direction.

    Why product variants in AI shopping are hard

    A variant looks simple: a shirt comes in small, medium and large. Under the hood it's three layers.

    • A parent product. The shirt as an idea.
    • Child options. Small, medium, large. Blue, black. Pack of 3, pack of 6.
    • Stock that differs by option. Medium might be on the shelf while small is gone.

    Different stores model these layers differently. Some list a shirt once, with a size menu. Some list each size as its own product. Some fold the size into the title and nothing else.

    That last pattern matters for agents. Sized goods are often listed one size per product, so today a good agent puts the size in its search query. If you want a medium, say medium.

    What a shopper expects

    Think about asking a friend to pick something up for you. If you say "grab me that jacket," a good friend says "what size?" before they leave. They don't guess and hope.

    Shoppers expect the same from an agent. The right behavior is:

    1. Notice that the product has options that change what you receive.
    2. Ask about them before buying.
    3. Confirm the choice in the total the shopper approves.

    An agent that skips step two is worse than no agent. A wrong purchase costs the shopper time, a return, and trust in the whole idea.

    The failure is easy to miss in a demo. The demo product is a roll of paper towels, which has no size. Real shopping is full of products that do.

    Shelf-level truth

    Asking the right question is half of it. The other half is asking one that can be answered.

    If the shopper wants a medium, the useful fact isn't "this jacket exists." It's "a medium is in stock at a store that can serve this address." Stock differs by option, and it differs by store.

    That's a shelf fact, which is what the SKU Graph exists to hold. Each store offer carries price, stock, source and when we last saw it. For variants, the same discipline has to apply per option. An offer for the jacket in general isn't an offer for the jacket in medium.

    It also means the question should be shaped by what's actually available. Offering a size that nobody near the shopper has is just a more polite way to fail.

    Where JENNi is today

    Here's what works now. Search shows in-stock items only. Agents can call search_catalog with a query and a 5-digit US ZIP, then get_buying_options for a given product_id to see availability, price, delivery fee, delivery promise and total before tax. Nothing is charged until the shopper approves the final total.

    Here's what we're building toward: variant-aware search and checkout, where size, color, count and flavor are first-class choices. The agent would see the options, ask the shopper which one they want, and have the stock behind each option checked before it asks.

    We aren't going to say it exists before it does. When it's live, we'll say so plainly.

    What agent builders can do now

    Until then, a few habits help.

    • Put the size in the query. "Women's medium rain jacket" will find better results than "rain jacket."
    • Check the product facts. Call get_product before you buy and read what it says.
    • Confirm with the shopper. Show the total from get_buying_options and have the shopper approve it before checkout.
    • Treat ambiguity as a reason to ask. If a product could mean several things, ask.

    That last one is the principle we're building around. An agent should earn a purchase by asking the right question first.

    Follow along

    If you build agents and want to see how products come back today, connect to https://shop.jennipro.com/mcp and try a search with a size in the query. If you shop, you can start at shop.jennipro.com, where you pick a product and every offer reads "Delivered by JENNi."

    Ready to connect?

    See how JENNi works for your business

    Whether you're a retailer with inventory, a brand, or building a channel — we'll show you exactly how the network fits.

    See the API