AI agents like Jowzi and Kroger's Gemini-based assistant are starting to plan meals, build carts and compare prices across retailers on a shopper's behalf, with Supermarket News reporting 26% of active gen-AI users say AI already pushed them toward a cheaper item and 21% saw their basket grow. For grocers and D2C sellers this means catalog data, promotions and fulfillment workflows now need to be machine-readable, not just human-readable, or the agent will simply route the sale elsewhere.
The grocery basket is no longer being built by a human standing in an aisle or scrolling an app. It is increasingly being assembled by software - an AI agent that plans the week's meals, checks what's already in the fridge, compares prices across two or three stores, and clicks 'buy' before the shopper has even opened the app. That is the shift Supermarket News flagged in its 4 September 2026 piece by Kelly Askew, and it changes something fundamental for every grocer and D2C seller: your customer is no longer only a person, it's also a process, and that process reads data differently than people do.
What exactly is an AI grocery agent?
An AI grocery agent is not a chatbot that answers questions. It's software that acts on the shopper's behalf across a full task, not just a single query. According to Supermarket News, these agents can plan meals, track household inventory, recommend quantities and automatically replenish essentials - moving grocery e-commerce from search-based browsing to what the article calls 'delegated shopping.' Instead of a shopper typing 'atta 5kg' into a search bar, the agent already knows the household goes through a 5kg bag every 18 days, checks which nearby store has it in stock and at what price, and adds it to a cart without being asked.
The key operational detail is that the agent is doing this with real data on budget, preferences and consumption patterns, which Supermarket News says lets shoppers decide with 'less guesswork and more precision.' For the grocer on the other end, that precision cuts both ways - it can win you a loyal, high-frequency customer, or it can quietly exclude you from consideration entirely if your data isn't good enough for the agent to trust.
How much are AI agents already changing what's in the cart?
This isn't theoretical. Supermarket News reports that 26% of active generative AI users say AI has already prompted them to buy a less expensive grocery item, and 21% say it has increased their overall basket size. Both effects can happen in the same household - the agent trades down on a commodity item like cooking oil while adding on impulse-adjacent items like snacks for a recipe it just planned.
Cooklist, whose AI shopping assistant already powers online stores for several major grocers in the US, reports double-digit basket size growth among shoppers who use it, according to the National Law Review. Cooklist is now rolling out to 10 additional grocery banners covering roughly seven million more shoppers. The exact percentage lift isn't public, but the direction is consistent with what Supermarket News is describing: agents don't just automate the basket, they actively grow it by surfacing complements a rushed human shopper would have skipped.
Who is actually building and running these agents right now?
Three examples make this concrete. Jow, operating as Jowzi in France and the US, runs an agent that IGD Retail Analysis describes as handling the entire journey end-to-end - meal planning, product search and checkout - across coverage of roughly 10,000 supermarkets, live with retailers including Carrefour, Intermarché and Chronodrive. IGD's most striking data point: around 80% of the products in a Jowzi cart come directly from the agent's own recommendations, not from the shopper manually searching. That's a basket where the human barely touches the keyboard.
Kroger has taken a similar route in the US, launching an AI-driven shopping agent built on Google's Gemini Enterprise for Customer Experience. Supermarket News reports the assistant can complete complex tasks from a single instruction - exploring meal ideas, building carts for large occasions, reordering past purchases and comparing product details, using Kroger's own proprietary data on assortment, pricing and availability.
Customers will be able to ask the integrated assistant to "create a shopping list based on their immediate needs, their budget and their family's unique preferences." - Yael Cosset, EVP & Chief Digital Officer, Kroger
Why is 'one-stop shopping' breaking apart?
For decades, grocery retail was built around the idea that a shopper picks one store and buys everything there for convenience. Supermarket News points out that AI agents undercut this logic because they compare price and availability in real time and route each item to whichever retailer best meets the shopper's need - not whichever store is nearest or most familiar. One basket might now become three or four smaller orders spread across different sellers, each chosen because the agent found a better price or faster delivery slot for that specific item.
Indian readers will recognise the early version of this already happening manually - a household ordering vegetables from one quick-commerce app, pantry staples from a subscription service, and specialty items from a D2C brand's own site. The difference now is that an agent does this comparison automatically, at scale, for every item, every time. A grocer that used to win the whole basket through convenience now has to win each item on data - price accuracy, real stock, and relevance - because the agent isn't loyal to your store, it's loyal to the outcome it was told to optimise for.
What does this mean for catalog data?
This is the part most grocers and D2C sellers underestimate. An AI agent doesn't browse your homepage banners or read your Instagram captions - it reads structured data: SKU-level pricing, live stock counts, unit sizes, substitution options, nutritional attributes, and delivery windows. If that data is stale, incomplete, or locked inside a PDF price list that gets updated manually once a week, the agent either skips your store or makes a recommendation based on wrong information, which damages trust the moment the customer's order arrives different from what was promised.
If your product feed isn't machine-readable and near real-time, an AI shopping agent won't argue with you about it - it will simply route the sale to a competitor whose data it can trust. There's no complaint, no lost review, just a sale that never happened.
What happens to promotions and fulfillment workflows?
Traditional promotions were built for human eyes - endcap displays, coupon codes, seasonal banners. An agent doesn't see any of that unless the discount logic is expressed as data it can act on: a machine-readable rule that says this SKU is 15% off until this date, or that a bulk discount applies above a certain quantity. Promotions increasingly need to be built as structured rules feeding into a catalog, not creative assets sitting on a webpage.
Fulfillment gets more complex too. Agent-built baskets mean more frequent, smaller, split orders across sellers rather than one large weekly shop, real-time inventory checks before every add-to-cart, and a higher chance the agent substitutes an item when stock is thin - which then needs to be communicated back to the customer without confusion. Grocers whose fulfillment stack still assumes 'one order, one store, once a week' will feel this friction first, in cancelled substitutions and mismatched delivery expectations.
How should a grocer or D2C brand prepare for this, and what would ODIV actually build?
This is exactly the kind of shift ODIV's ai-workflow-automation service exists for. The practical work isn't philosophical, it's plumbing: getting your catalog, pricing, stock and promotion logic out of manual spreadsheets and static pages and into structured, near real-time feeds that an AI agent - whether it's a shopper's own assistant or a marketplace's ranking system - can actually read and trust. That includes building the automation that syncs inventory across your website, your quick-commerce listings and your own ordering system, so an agent never sees three different stock numbers for the same product. It also means encoding your promotion and substitution rules as data the system applies automatically, instead of rules living only in a manager's head.
ODIV's engineers build this using modern AI coding tools like Lovable and Claude Code alongside conventional backend engineering, which is what lets a working, properly integrated system get built in a fraction of the time and cost of a traditional custom development project. The AI tools get you to a working build fast; the engineers make sure it's secure, connects correctly to your inventory and POS systems, and keeps working after launch. Grocers already using WhatsApp for order updates and repeat-order reminders can layer that on through ODIV Engage as one piece of the customer-facing side, but the real work here - and where ODIV's value sits - is fixing the data and automation underneath so agents, and the customers behind them, keep choosing you. If your catalog, promotions or fulfillment workflow needs this kind of overhaul, start a chat with ODIV on WhatsApp and we'll walk through what a first build would actually look like.
Frequently asked
A regular app waits for the shopper to search and click. An AI grocery agent acts on the shopper's behalf across a whole task - planning meals, tracking what's running low at home, comparing prices across stores and building or reordering the cart automatically, often with minimal manual input from the shopper.
Both effects are already showing up. Supermarket News reports 26% of active gen-AI users say AI pushed them toward a cheaper item (which can shrink spend per item), while 21% say it increased their overall basket size, and Cooklist reports double-digit basket growth for grocers using its AI assistant, largely by surfacing complementary items a rushed shopper would skip.
The main lever is catalog data quality: accurate, near real-time pricing, stock levels, unit details and promotion rules structured in a way software can read and trust. Sellers with stale or incomplete data risk being skipped entirely, since agents route to whichever retailer's data they can rely on, not necessarily the nearest or most familiar store.