July 13, 2026

The Proof Arrives: Walmart’s Own Agent Drives the Sales, AI Shoppers Outconvert Every Channel, Visa Turns On Live Agent Checkout Across Europe

Week of July 7-13, 2026. The biggest agentic-commerce news of the period was not a launch, it was a set of receipts. Walmart put concrete numbers behind Sparky, the shopping agent it runs itself: agent-attributed sales up 150 percent, units up fourfold quarter over quarter, baskets 35 percent larger, and roughly half its app users already engaged. The Prime Day post-mortem settled into view and flipped the trust story, with AI-referred shoppers now converting far better than every other channel. And the payment layer moved from pilot to production: Stripe and Cross River shipped card infrastructure built for agents, and Visa turned on live agentic checkout across Europe with real banks and real stores.

Last week we covered Salesforce putting a transactional agent on the merchant’s own storefront, the payments industry rallying behind one stablecoin, and Europe’s transparency clock starting August 2. That was the infrastructure hardening. This week the evidence caught up to it. A retailer running its own agent showed what the agent actually did to sales. The biggest shopping event of the summer produced data that reverses the “shoppers do not trust agents” narrative we have been tracking all year. And the money rails that spent the spring being announced started clearing real purchases, including on this side of the Atlantic.

Here’s what happened.

Walmart’s own agent stops being a story and becomes a number

Walmart put hard figures behind Sparky, the AI shopping agent it built and operates itself. In its most recent quarter, the value of goods sold attributed to Sparky rose 150 percent, and units sold through the agent more than quadrupled from the prior quarter, driven largely by replenishment of consumables. Roughly half of all Walmart app users have now interacted with Sparky, and CEO John Furner has said app users who engage the agent spend about 35 percent more than those who do not, with Sparky-attributed baskets running about a third larger. The agent has moved well past a search box: it is now live in stores, handles meal planning, automates household restocking, and speaks Spanish. Walmart has also pushed Sparky beyond its own properties, embedding it inside ChatGPT and Gemini so the agent travels to where shoppers already are.

Why it matters for merchants: This is the first time a mainstream retailer has put a clean number on what a shopping agent does to the top line, and the number is large. Last week we argued that Salesforce shipping a transactional agent onto the merchant’s own storefront was the move merchants had been waiting for. Walmart’s Sparky results are the proof that the own-agent bet pays. A 150 percent jump in agent-attributed sales and baskets a third larger are not pilot-scale curiosities, they are the kind of figures that move a quarter. And notice what drove the unit growth: replenishment, the low-consideration, buy-it-again category that agents win first, exactly the wedge we flagged when Amazon put voice checkout into pizza and tickets. The lesson for a merchant on any platform is not “go build a Sparky,” because most cannot. It is that the agent channel is now demonstrably a sales channel, not a demo, which means the catalog and inventory hygiene that lets an agent recommend and reorder your product is a revenue question, not a hygiene one. Walmart’s own numbers put a dollar figure on being the product the agent picks. (See our earlier coverage of the first agentic-commerce numbers going public.)

Update: Walmart extended the disclosure in August, telling its Q2 call that Sparky users spend 40 percent more per order with the user base up 70 percent year over year, alongside comparable figures from Lowe’s, Target and Albertsons. Worth reading with the selection-effect caveat we set out there: these are user versus non-user comparisons, not controlled tests.

Prime Day flips the trust story: AI shoppers now outconvert everyone

The single most important data point of the period came out of the Prime Day post-mortem. Amazon’s event ran June 23 to 26 and drove a record 26.4 billion dollars in US online spend, up 9.3 percent, but the number that matters for every merchant is about who was doing the buying. According to Adobe Analytics, traffic to retail sites from AI sources jumped nearly 90 percent year over year during the event, and those AI-referred shoppers converted roughly 40 percent better than visitors from every other channel, including paid search, email, and social. That is a reversal. A year earlier, during the 2025 event, AI traffic converted 23 percent worse than non-AI sources. AI-referred shoppers also spent about 50 percent more time on site, viewed 20 percent more pages, and added to cart at a 33 percent higher rate than traditional visitors.

Why it matters for merchants: For a year the counterargument to agentic commerce has been trust. Shoppers would research with an agent but not buy through one, the data showed people trusted AI to compare prices far more than to place an order, and that gap was the reason to treat all of this as a future problem. This Prime Day broke that argument. When AI-referred shoppers go from converting 23 percent worse to roughly 40 percent better in twelve months, the trust gap is not closing gradually, it has crossed over. And the direction of the engagement numbers tells you why: a shopper arriving from an AI conversation has already been qualified, has already compared, and lands on the page closer to a decision than someone who clicked an ad. That is the shopper every merchant wants, and they are now the highest-converting traffic on the site during the biggest event of the year. The practical read is blunt. If AI-referred traffic is your best-converting channel and it is growing 90 percent a year, then being discoverable to agents is no longer a defensive hedge against a future shift, it is where your best customers are coming from right now. The merchants who win this traffic are the ones whose product data answers the agent’s question before the shopper ever sees the page. (See our earlier coverage of the trust gap that this data just closed.)

The payment rails stop being demos: Stripe ships cards built for agents

On the infrastructure side, the theme was production, not announcement. On July 1, Cross River and Stripe expanded their issuing partnership to deliver bank-grade card infrastructure built specifically for AI agents. The mechanism is single-use virtual cards scoped to one transaction: an agent buying on a shopper’s behalf presents a one-time card number, never the shopper’s real credentials. For consumers, Stripe’s Link agent wallet mints those single-use cards automatically. For developers building their own agent-driven checkout, Stripe exposes programmatic card issuance through its API, all running on Cross River’s banking core so agent-initiated transactions meet card-network rules, AML, and KYC requirements. The partnership extends a relationship the two companies started in 2019 on push-to-card payments, now pointed at the always-on, high-frequency flows a delegated agent generates.

Why it matters for merchants: Two weeks ago we covered the payments industry rallying 140-plus firms behind one stablecoin to settle what agents buy. This is the same story one layer down and one step more concrete: not a future settlement standard, but a live card that an agent can present at your checkout today, scoped so a single compromised transaction cannot leak the shopper’s real card. That matters to a merchant for a practical reason. The thing that would have made you nervous about an agent transacting on your site, that you are accepting a payment from software rather than a person, is exactly what single-use, agent-scoped cards are designed to de-risk. The fraud and chargeback questions that kept agent checkout theoretical get easier when the credential is one-time and the issuer sits on a compliant banking core. You do not have to integrate anything new this week. You do have to notice that the payment plumbing under agent checkout is now shipping as product, which removes one more reason an agent purchase would fail to complete on your store.

The European angle: Visa turns on live agent checkout across Europe, no sandbox

For European merchants, the most concrete development of the period was a switch being flipped. On July 2, at the Visa Payments Forum in Paris, Visa announced that AI agents are now executing live purchases with real, independent merchants across Europe on behalf of cardholders, moving past the controlled test storefronts of earlier demonstrations. More than 30 issuing banks have completed live, agent-executed transactions, including Klarna, Barclays, HSBC UK, NatWest, ING, Revolut, BBVA, CaixaBank, Commerzbank, Lloyds Banking Group, Nordea, and Nexi Group. The first live merchants span sectors and countries: lastminute.com in travel, Frasers in retail, plus Cleverbridge and BrickDepot. The transactions run on Visa’s Trusted Agent Protocol and its Agent Directory, which let a merchant securely recognize a verified agent, with Visa Payment Passkeys authenticating the purchase.

Why it matters for merchants: We have argued since April that the European window is real and that the US-first, Europe-fast-follow pattern keeps compressing. This is the pattern arriving at its most important milestone: a live, cardholder-authorized agent purchase at a real European store, cleared by a real European bank, is no longer a lab result. When Barclays, NatWest, Klarna, and BBVA are among the banks that have processed one and lastminute.com is among the merchants that have accepted one, the agent channel in Europe stops being something a merchant watches and becomes something a merchant is either reachable through or invisible to. And the piece that makes it work for merchants is the Trusted Agent Protocol and the Agent Directory: they answer the question a European merchant would rightly ask, which is how do I tell a legitimate delegated agent from a bot, and how do I accept its payment without owning the fraud. Visa is answering that at the network level, for the banks European merchants already use. The catalog work we keep flagging is now due on this side of the Atlantic for the same reason it is due on the other: the checkout is live, so the only variable left is whether your product is the one the agent finds and buys. (See our earlier coverage of Visa’s Trusted Agent Protocol as it moved toward Europe.)

What moved this period

Development	What happened	Why a merchant cares
Walmart Sparky results	Agent-attributed sales up 150%, baskets ~35% larger	Proof a retailer’s own agent moves real revenue
Prime Day AI conversion	AI traffic up ~90% YoY, converted ~40% better than others	The trust gap flipped: AI shoppers convert best
Stripe + Cross River agent cards	Single-use virtual cards scoped per agent transaction	De-risks accepting a payment from software, not a person
Visa live agent checkout in Europe	30+ banks and real merchants clear live agent purchases	The European agent channel is live on banks merchants use

What merchants should do this period

1. Treat AI-referred traffic as your best channel, because the data now says it is. Prime Day showed AI-referred shoppers converting roughly 40 percent better than every other source, a full reversal from a year ago. Look at your own analytics for traffic arriving from ChatGPT, Gemini, Perplexity, and Google AI Mode, and check how it converts against paid search and email. If the pattern holds for you, being discoverable to agents is not a hedge, it is where your highest-intent customers already come from.

2. Read Walmart’s Sparky numbers as a target, not a curiosity. A 150 percent lift in agent-attributed sales and baskets a third larger is what happens when the agent can recommend and reorder your products accurately. Most merchants cannot build a Sparky, but every merchant can make sure their catalog is complete and structured enough that whatever agent a shopper uses can pick their product with confidence, especially in replenishable categories where agents buy first.

3. Expect agent-scoped payment as a checkout feature, and stop treating agent payments as a risk you have to solve alone. Stripe and Cross River shipping single-use cards for agents means the fraud and credential questions are being handled at the payment layer. Ask your payment provider what their agent-payment and single-use-card support looks like, so that when an agent presents a one-time card at your checkout, it completes rather than declines.

4. European merchants: the checkout is live in your market, so the catalog work is due now. Visa turned on live agent purchases at real European stores, cleared by Barclays, NatWest, Klarna, ING, BBVA, and 25-plus other banks. The infrastructure question is answered for you. The only variable left is whether your product data is good enough for an agent to find, describe, and buy it. Treat the European window as open, because this period it started clearing real transactions. (See our coverage of how far most catalogs still are from ready.)