Issue 02 · July 2026 · The infrastructure layer beneath grocery’s AI moment
Last month we asked whether your AI initiative was confident in the model or confident in the data. This month, someone answered the question — by writing it into a standard.
In June, Target announced it had become the first mass retailer with shopping experiences live across three AI platforms simultaneously — Google Gemini, Microsoft Copilot, and OpenAI’s ChatGPT — reporting a 2,000% year-over-year surge in AI-driven traffic in Q1. That number is striking. What’s more important is what came alongside it.
Target didn’t just adopt the Universal Commerce Protocol. They helped write it. UCP is Google’s open standard for agentic commerce — the common language that lets AI agents discover, recommend, and transact across any retailer’s catalog. Target co-developed it alongside Google, Shopify, Wayfair, and Walmart. Which means their product data architecture isn’t just their problem anymore. It’s the reference implementation every other retailer will eventually be measured against.
“Target didn’t join the standard. They built it. Everyone else is now writing to their spec.”
This is how infrastructure standards get set in retail. Not by committee, not by regulation — by the first large operator to build something that works at scale and let everyone else integrate to it. The UPC was standardized the same way in 1974, when a single barcode was scanned at a Marsh Supermarket in Ohio and the industry quietly agreed to follow. Sunrise 2027 is the current version of that story, and we’ll come back to it.
The UCP’s requirement isn’t complicated: for an AI agent to discover, recommend, and transact on a product, the product data feeding it must be structured, complete, and accurate enough to satisfy the protocol’s disambiguation logic. Product attributes — ingredients, allergens, nutritional data, dietary flags, category taxonomy — are what the AI is reasoning over. Gaps in that data don’t generate errors. They generate confident wrong answers.
The 2,000% traffic surge isn’t a preview of the future. It’s confirmation that the future is already routing traffic through a product data layer most retailers haven’t reviewed since they built their current eCommerce stack. The next 18 months will sort grocery retailers into two groups: those whose AI investments compound because the data underneath is clean, structured, and complete — and those whose AI investments expose data debt they’ve been deferring for years. The retailers moving fastest on agentic commerce share one thing: they treated product data infrastructure as a strategic investment before the AI moment arrived, not after. The standard has been set. The question is whether your catalog is ready to meet it.
Agentic Commerce
Cooklist’s agentic AI shopping assistant launched across Wegmans and Kroger this month, serving 700+ stores and 10 million digital shoppers. Shoppers describe what they want in plain language; the assistant checks inventory and assembles a personalized, checkout-ready cart. That’s the part that made the press release. Here’s the part that didn’t: Cooklist is also generating reports from those shopper conversations and routing them directly to produce, meat, deli, and dry goods teams — connecting digital demand signals back to individual departments in real time. The agentic AI layer isn’t just a frontend experience. It’s a feedback loop. And that feedback loop is only as coherent as the product attribute data the AI is reasoning over to make its recommendations in the first place. The grocers getting the most out of this aren’t the ones with the most sophisticated AI — they’re the ones who went into it with a single, authoritative source of truth for their product catalog.
Grocery Dive: Meet the automated cart builder used by Kroger and Wegmans →
Infrastructure Signal
iHerb’s recent engineering job postings reveal more about their AI architecture than most companies would disclose in a press release. Two senior AI roles posted this month describe a conversational “Wellness Agent” powered by LLM-driven recommendations and — notably — RAG pipelines that retrieve directly from the product catalog in real time. This is a different architecture than most retailers are running. The AI isn’t querying a search index. It’s reasoning over raw product attribute data at inference time. That means product data quality isn’t a discovery-layer concern at iHerb — it’s the literal input to every recommendation the agent makes. If an attribute is missing or wrong, the retrieval is wrong, and the Wellness Agent confidently recommends the wrong thing. The gap between “we have a RAG system” and “our catalog is ready for a RAG system” is a gap most operators haven’t measured yet. The ones who close it first won’t advertise it. They’ll just have better AI outcomes.
Quiet Signal
The June 23 Delta brief marks nine straight weeks of named AI and loyalty signals at Loblaw — the longest sustained streak in the tracking database. This week’s addition: Specsavers Canada joining PC Optimum, extending the loyalty ecosystem beyond grocery into eyewear retail. PC Optimum’s personalization engine now has to handle dietary flags, allergen data, optical prescriptions, and lens specifications — simultaneously, across 18 million members. Loblaw isn’t building features. They’re building surface area, and every new surface is a new product data dependency. The question isn’t whether they’re moving fast. It’s whether the product data foundation is keeping up with the expansion. Organizations that have invested in standardized, centrally governed product data find that each new surface costs them less than the last. Organizations that haven’t find the opposite.
Worth Your Time
The retail barcode has not changed in any fundamental way since the first UPC was scanned at a Marsh Supermarket in Ohio on June 26, 1974. GS1 Sunrise 2027 is the formal name for the initiative that ends that era. By December 31, 2027, all compliant retail POS systems worldwide must be capable of reading 2D barcodes — QR codes and GS1 DataMatrix codes that carry not just a GTIN, but lot numbers, expiration dates, serial numbers, allergen data, and a Digital Link URI connecting the physical product to real-time information. Walmart, Target, and Kroger have all made public commitments to 2D readiness at checkout. Schnuck Markets discussed their Sunrise 2027 initiative publicly with GS1 US this month.
Here’s the part worth sitting with: December 31, 2027 is not a shutdown date for the 1D barcode. It’s the date by which 2D scanning capability must exist at every compliant POS. The practical pressure comes from what follows — retailers building 2D infrastructure will start expecting 2D-ready supplier data, and the window to get ahead of that expectation is closing. The companies treating this as an IT project will be late. The companies treating it as a product data readiness milestone will be positioned. The retailers best placed for this transition already have a single source of truth for product attributes — not because they anticipated the barcode standard specifically, but because that discipline pays dividends across every new surface that comes along. The UPC took 50 years to become the default. The 2D barcode won’t wait that long.