Homesome Fieldnotes — September '26

Field Notes is back after skipping August (yeah, I know). The break was travel, and the industry did not slow down while I was gone. Since the last edition, two national marketplaces, one national grocer, and one of the largest technology companies in the world each shipped an AI that turns a conversation into a grocery basket. Three of the four do not own a single store.

That is what this edition is about, and I’m returning with a firmer view than I left with: grocery will be the first category consumers delegate to AI agents, and the retailers least prepared for that have the most to lose. Good to be back. Let's get into it.

🍓 Grocery Goes First
💬 Featured Insight: Who Owns the Customer? by Gary Hawkins

Every survey on AI shopping this year is hitting the same notes. People will let an agent buy on their behalf, but only for things that are cheap, routine, and low-regret. Product.ai's 2026 Trust in AI Commerce report found the most common spending ceiling for an unverified AI recommendation was under $25, chosen by 42% of shoppers. Gartner puts the share willing to let AI make a purchase decision alone at no more than 11%, even in household supplies. Croud's Consumer Index found 69% of Americans open to AI purchasing without per-transaction approval, with routine purchases drawing the most trust. J.P. Morgan expects agentic commerce to reach a quarter of US digital sales by 2030, with most of that volume in recurring, low-risk categories, groceries first among them.

Read those together and the conclusion is uncomfortable for our industry. The autonomy curve starts exactly where grocery lives: a $6 item, bought every week, where a wrong pick costs almost nothing and a right one saves twenty minutes. Apparel and electronics will wait years for the shopper's trust. Milk, eggs, and the Tuesday dinner plan will not. The category that has been slowest to adopt technology is the one consumers will delegate soonest, and the agents launching this month understand that better than most of the retailers they intend to shop from.

Here’s the part I find encouraging, because the regional grocer is not starting from behind. Intouch Insight's 2026 Grocery Value study found 79.2% of grocery shoppers use a digital tool before or during a trip. When those shoppers go looking for product information, the retailer's own app or website is the first source for 43.1% of them, ahead of search engines at 32.2% and the store shelf itself at 21.1%. The retailer's app already outranks Google and the aisle as the place the shopper begins.

That starting position is what every agent is competing for. A personal agent that wants to plan the household's week has to begin somewhere, and today it begins where the shopper already keeps the list, the loyalty account, and the weekly ad. The retailer already owns that starting point, but very few of them treat it like the advantage it is.

That position can also be lost quickly. Accenture's 2026 consumer research found 37% of behaviorally loyal shoppers, the ones who normally stick to one to three brands, would let an agent switch them for a better fit. That is a brand statistic, but nothing about it stops at the brand. If an agent will move a household off its peanut butter, it will move a household off its grocer when the other store's data answers the question faster. Numerator's share tracking shows Costco, Amazon, and Walmart each gaining grocery share over the past year while major supermarket companies posted declines. The agents did not cause that. They will accelerate it if the supermarket cannot be read.

The open protocols now arriving from Google, OpenAI, and Stripe describe the retailer's job plainly. Google's Universal Commerce Protocol lets an agent pull real-time catalog details, inventory, and pricing straight from the retailer, and its identity linking capability lets a logged-in shopper receive the same loyalty pricing and member benefits inside the agent that they would on the retailer's own site. Salesforce's analysis of UCP puts the implication bluntly: retailers have to stop building destinations and start operating as data-rich nodes on a network, because if inventory, price, or attributes are even slightly out of sync, the agent moves on.

That last part deserves a moment. In the old model an out-of-stock cost you one sale. In the agent model a stale inventory feed costs you the recommendation, and the shopper never learns you had the item. Retail Economics' Availability Effect report, audited across more than 100 UK grocery locations, found on-shelf availability at 89.7% and put £2.1 billion in annual sales at risk from stock gaps even in that disciplined market. A machine deciding between two stores will treat a gap in your data the way a shopper treats a gap on your shelf, except it will never come back to check.

Which brings me to the point I most want regional operators to hear. Your loyalty program, your personalized offers, your store-level pricing, and your item data are the best assets you own and the ones an agent most wants to read. A shopper asking an agent for the cheapest way to feed four this week is asking a loyalty question. If the answer to that question lives in a card program, a coupon vendor, a separate promotions engine, and a digital store that carries none of it, the agent will assemble a coherent answer from whichever competitor has one. The gap in spending between a regional banner and Walmart is not closable. The gap in readability is, and it is smaller than most executives think, because the data already exists. It is scattered, not missing.

The fastest response on offer to regional grocers right now is a white-label assistant from a national marketplace. Modern Retail and Grocery Dive both reported this month that the marketplaces are now selling exactly that: a conversational assistant that lives on the retailer's site and app, builds carts from a photo of a list, finds deals, and reorders, with regional banners already signing on because building one alone would take time and money they do not have.

I understand the appeal, and I want to be fair about it. The shopper sees your brand. The cart is your cart. The problem is quieter than a marketplace taking your customer. Every conversation that assistant has is a lesson about your household: what they will not eat, how far their budget stretches, which substitution they accepted. Schnucks, in its own assistant launch, said the thing that matters here out loud: the retailer maintains ownership of the customer relationship, the shopper data, and the brand experience. If your vendor agreement cannot say that sentence back to you, the assistant is teaching someone else's model how your customers shop, and the logo on the screen is the only part of it you own. Homesome has taken the position for years that a regional grocer's digital experience should run on its own loyalty, offer, and store-level data as first-party structure rather than a marketplace's catalog, and this is the month the reason for that stopped being a philosophy and became a competitive fact.

Treat agent readiness as a data project, not an AI project. Get item attributes, store-level inventory, and pricing accurate and current before anyone writes a prompt, because that same data is what a human shopper needs from your app today and what every agent will grade you on tomorrow.

Bring the loyalty program all the way into the digital experience and measure digital engagement against the store as a lift target. The program is the asset the agents want to read. Make it legible.

Before signing any assistant agreement, write down who owns the conversation data, who can train on it, and whether the learning stays with you when the contract ends. If the vendor cannot answer in one sentence, that is the answer.

And stop measuring yourself against Walmart's budget. Measure yourself against the question the household will ask its agent in eighteen months, and whether your store can be the one that answers it.

The customer relationship is moving to the agent that understands the household and acts on its behalf. Traditional retailers are in danger of supplying the products while someone else owns the intelligence that determines the basket, selects the seller, and learns from every decision.

Instacart released Clementine, an AI-powered grocery shopping assistant that converts a conversation, recipe, handwritten list, or simple request into a personalized cart. Meta released Muse, a personal agent that can work across applications, remember what matters to an individual, browse the internet, and make purchases with approval. At the same time, Schnuck Markets introduced its own AI assistant inside the Schnucks Rewards experience, retaining ownership of the customer relationship, shopper data, and brand experience.

Taken together, these developments make the emerging structure of agentic commerce visible. Three kinds of agents are competing for position: the marketplace agent that sees demand across retailers, the personal agent that understands the consumer across areas of life, and the retailer agent that connects household needs to a specific brand, assortment, loyalty program, and store network. The strategic question for retailers is no longer whether shoppers will use AI. It is which intelligence will represent the customer when purchase decisions are made, and what role the retailer will occupy in that decision system.

Read the full paper →