May 25, 2026

The production numbers arrived this week, and they came from the sales floor rather than the pitch deck. Lowe's said its Mylow assistant now handles more than a million customer inquiries a month, and shoppers who use it convert at three times the rate of those who do not, according to Constellation Research's tally of retail AI results. Walmart put a figure on Sparky back in early March, when Morgan Stanley pegged a 25% spend lift, and the number has since grown. CEO John Furner now says Sparky users place orders 35% larger than non-users, with weekly active users more than doubling last quarter and response quality up 40% this year. These are not projections. They are operator metrics from two of the largest retailers in the country, and they describe agentic shopping that already moves money at scale.

Breadth matters here as much as depth. Williams-Sonoma has scaled AI personalization across its entire brand portfolio and rolled agentic training tools out company-wide, with CFO Jeff Howie tying the effort to a "perfect order on time, damage-free every time" standard. Target has handed its customer experience overhaul to OpenAI. Urban Outfitters parent URBN is deploying both Gemini and Claude across its teams rather than betting on a single model. The same Constellation analysis frames the shift cleanly, noting that AI at these retailers has moved out of standalone innovation projects and into the operational layer, where it cuts friction and shows up in productivity numbers. The production gap this digest tracked through the spring has, for the category leaders, effectively closed.

All that demand is now the prize in a four-way fight. Google shipped its Universal Cart on May 19, a single cart that follows a shopper across Search, Gemini, YouTube, and Gmail and checks out through Google Pay. The company keeps positioning itself as the neutral layer. "We are not a retailer, we are not a marketplace, and that approach continues to guide us in this agentic era as well," VP Ashish Gupta told Digiday. The competition is not waiting on Google to define the terms. Amazon retired Rufus for Alexa for Shopping last week. TikTok Shop booked $4.9 billion in U.S. sales in the first quarter, nearly double a year earlier. Meta is testing its own AI shopping-research feature to rival Gemini and ChatGPT. Whichever surface captures the shopper's first query captures the transaction, and four of the largest platforms in technology now build around that fact.

The shoppers those platforms are fighting over are not sold. The same Digiday reporting cites survey data showing 54% of Americans find the idea of an AI agent reaching into their shopping history unappealing, and 73% say they feel uneasy about how AI might use their personal shopping data. That unease sits directly on top of the capability. An agent barred from purchase history cannot personalize, and personalization is the entire value proposition, so without it the agent is just a slower checkout button. The industry has spent a year solving for whether agents can transact. Whether shoppers want them to is the harder question, and the numbers this week left it wide open.

Trust has a legal dimension too, and one side of it moved this month. This digest flagged in April that Regulation E offers no clear answer on whether a shopper keeps dispute rights when an authorized agent botches an order, and that the card networks were laying rails faster than the rulebooks could define them. Visa has now closed part of that gap. Its Core Rules carry express provisions for agentic transactions, requiring identity verification under the Intelligent Commerce specification and use of a provisioned token. Mastercard's public Transaction Processing Rules still say nothing about agents, handling the subject through product programs like Agent Pay instead. The asymmetry is telling. A network that writes agent behavior into its binding rulebook is making a different, more exposed bet than one that keeps it in a product.

Step back and the week tells one story. The supply side of agentic commerce is no longer the open question. Walmart and Lowe's have the receipts, four platforms are racing to own the surface, and Visa is already writing agents into its rules. What stays unsettled is everything on the other side of the transaction. A majority of shoppers remain wary of letting an agent see what they buy, and the law still cannot say who absorbs the loss when an agent buys the wrong thing. Infrastructure outran both trust and the rulebook this spring. Merchants posting numbers like Lowe's and Walmart can treat that gap as the opportunity. Everyone else will find it is the thing that decides who actually gets to use what the engineers have already built.