June 08, 2026

Britain Gets Its First Agentic Checkout, OpenAI Turns Codex Into a Store Builder, Mastercard Folds the Agent Into the Core

June 2-8, 2026. After the Google I/O blitz, a quieter week that still moved the map. The agent crossed the Channel into UK retail, OpenAI turned Codex into a business builder, and Mastercard put agentic and stablecoin payments under the same roof as the card.

Last week we covered Google shipping the Universal Cart, Amazon renting out its shopping agent, and Microsoft opening Copilot Checkout. That was the loud fortnight. This week was the quiet one that tells you where the loud announcements are headed. On June 2, PayPal and a London startup called Hey Savi launched what they bill as the UK’s first agentic commerce platform with checkout that completes inside the app, and they brought Debenhams Group in as the first retailer. The same day, OpenAI shipped a major Codex update with a Sites feature that, through a Wix integration announced June 3, lets a person go from an idea to a live store without leaving the agent. Also on June 2, Mastercard reorganized its leadership so that stablecoins and agentic payments now sit inside the consumer payments org rather than off to the side, and the follow-up coverage of Amazon’s AWS shopping-agent rental put a hard conversion number on it.

Here’s what happened.

PayPal and Hey Savi put the first agentic checkout in Britain

On June 2, PayPal and Hey Savi launched what they describe as the UK’s first agentic commerce platform with native in-app checkout. Hey Savi is a brand-agnostic fashion search app that uses computer vision and conversational AI: a shopper searches by photo, screenshot, or plain text across more than 10,000 brands, and the results are ranked by relevance rather than by who paid for placement. PayPal supplied the part that makes it a store rather than a search box. Its Agentic Commerce Services connect merchant storefronts to the app with real-time pricing and inventory, and PayPal’s payment layer lets the purchase complete inside Hey Savi with no redirect to an external checkout. Debenhams Group, which owns Debenhams, Karen Millen, Boohoo, and PrettyLittleThing, is the first retailer live on the platform. Hey Savi co-founder Sarah Daniel framed the pitch at merchants directly, calling it a step change where brands “of every size can be discovered by the highest-intent shoppers in fashion” on a platform that “democratises discovery ensuring relevance now directly drives revenue.”

Why it matters for merchants: This is the European angle becoming a product instead of a forecast. For months the story for UK and EU sellers has been “the agentic checkout is coming, but it is routing to the US first.” Hey Savi is a UK-built app, with a UK retail group, completing purchases on a payment rail that already holds the consumer relationship across Europe. The discoverability mechanic is the part to read closely. Hey Savi ranks by relevance, not by ad spend, and surfaces brands “of every size,” which means the determinant of whether a Debenhams Group label shows up next to an independent boutique is the quality of the product data, not the size of the media budget. That is the same shift we have flagged since the start: in agentic surfaces, structured, accurate product data is the shelf, and a small brand with a clean feed can out-rank a large one with a messy catalog. The concrete read for European fashion sellers: this is the first surface where you can test agentic discovery on home turf, and the entry ticket is a complete, relevance-ready catalog, not a partnership negotiation. (See our coverage of the 99% catalog-readiness gap for why most feeds are not ready, and our coverage of Europe’s regulatory lag for the timeline this jumps ahead of.)

OpenAI ships a major Codex update and turns it into a store builder

On June 2, OpenAI released a major Codex update aimed at work beyond software engineering. Codex now has more than 5 million weekly active users, more than a sixfold increase since the desktop app launched in February, and knowledge workers (not developers) are now about 20% of the base and growing three times faster than the developer cohort. OpenAI shipped six industry plugins (data analytics, creative production, sales, product design, equity investing, and investment banking), each bundling the integrations and context a role needs to work out of the box. The piece that matters for commerce is the new Sites feature: Codex can now output its work as a hosted, interactive website rather than as files on a developer’s machine, with launch partners including Wix, Figma, and Replit.

Wix made the commerce implication explicit a day later. Per its June 3 announcement, the integration connects OpenAI’s agentic platform to Wix Headless, so a Codex user can pair a front end built in the agent with Wix’s backend business stack (Wix Stores, Bookings, Events, CMS, and Payments) and go “from a business idea to a live, revenue-ready operation without leaving Codex.” This is the productization of the move we flagged last month, when OpenAI told staff it was merging ChatGPT, Codex, and the developer API into a single agentic platform under Greg Brockman.

Why it matters for merchants: The thing to notice is the collapse of the distance between “I have a product idea” and “I have a store taking payments.” For an existing merchant, the Codex-plus-Wix path is not the headline. The headline is the new competitor it creates: the marginal cost of standing up a fully functional storefront, with payments wired in, just dropped to a conversation with an agent. The defensible asset stops being “I have a website” and becomes “I have a catalog, a brand, and customer relationships the agent cannot generate for free.” For merchants already on a platform, the read is that the platforms are racing to be the backend the agent reaches for (Wix here, the same way Shopify positioned its Storefront MCP), so the question is whether your stack is one the agents can build against. The wider signal is that OpenAI is executing the unified-platform plan on schedule, which means the shopping conversation and the store-building conversation are heading for the same surface. (We covered the Brockman merger that set this up in May.)

Mastercard folds stablecoins and agentic payments into the core

On June 2, Mastercard announced a leadership reorganization that reads less like a personnel memo and more like a statement about where it thinks payments are going. Ling Hai, who ran Asia Pacific, Europe, the Middle East, and Africa, becomes Chief Financial Officer effective August 3. Outgoing CFO Sachin Mehra moves into a newly created Chief Business Officer role, Linda Kirkpatrick becomes Chief Services Officer, and Craig Vosburg becomes Vice Chair. The detail that matters for this newsletter is the product org: Chief Product Officer Jorn Lambert continues to lead consumer payments, now explicitly bringing together stablecoins, agentic capabilities, and the core card business under one organization. Per Payment Week’s reporting, the company tied the move to “artificial intelligence opening paths to agentic commerce” and to stablecoin interest amplified by US legislation.

Why it matters for merchants: A year ago, agentic payments and stablecoins were the experimental wing, the stuff a card network runs in a lab while the real money flows over the card rails. Putting them inside the same org as core consumer payments is Mastercard saying the experimental wing is now part of the main building. For a merchant, the practical consequence is consolidation of the rails you will eventually accept. Mastercard Agent Pay (the credential-and-authorization layer that lets an agent transact on a shopper’s behalf) sits alongside UCP, Google’s AP2, and the Stripe and Visa machine-payments work we have tracked, and the network is signaling it will offer card, agentic, and stablecoin settlement as one product family rather than three. The action item is not urgent this week, but the direction is worth pricing in: the payment options an agent can present at checkout are converging on a smaller set of standardized rails, and the merchant who builds to the protocol layer (UCP or ACP) rather than to a specific provider inherits the agentic and stablecoin options as they turn on, instead of integrating each one. (See our UCP vs ACP breakdown for the protocol map these rails plug into.)

Update: A week later, Mastercard shipped the protocol that uses this consolidated org. See our coverage of Agent Pay for Machines and Visa’s move into ChatGPT.

Amazon’s shopping-agent rental gets its conversion number

The wider reporting this week filled in the detail on the AWS Agentic Shopping Assistant, the service we covered last week when Amazon put its shopping-agent engine on AWS for retailers to rent. The new detail is the performance claim and the plumbing. The assistant runs on Amazon’s Bedrock AgentCore infrastructure, draws on data from Alexa for Shopping queries that ended in a sale, and Amazon says conversational shopping sessions convert at 3.5 times the rate of traditional keyword product searches. Tapestry’s Kate Spade AI Gift Concierge, built on the service, was tested for roughly two and a half months before going public. Retailers can stand up their own branded agent in weeks.

Why it matters for merchants: Nothing in the build-versus-buy calculus changed from last week, but the 3.5x conversion figure is the number to file. It is Amazon’s own framing of why a conversational agent on your storefront is worth the strategic dependency, and it sits above the roughly 2x conversion lift OpenAI reported for ChatGPT shopping ads in our May coverage. The figure cuts both ways: it is the upside of putting an agent on your site, and it is also the reason the agentic surfaces you do not control are pulling conversion toward themselves. The decision we flagged last week stands: a storefront agent is now a weeks-long project, and the real question is whether you want it running on Amazon’s stack, with the catalog and conversion read that hands Amazon. (See last week’s full breakdown of Amazon’s three-posture agent strategy.)

Where the AI commerce stack stands this week

Layer	Update this week
European agentic checkout	PayPal and Hey Savi launch UK’s first in-app agentic checkout, Debenhams Group live
Agent-as-store-builder	OpenAI ships Codex Sites, Wix tie lets you build a revenue-ready store in the agent
Payment rails	Mastercard folds stablecoins and agentic payments into the core consumer org
Agent-as-a-product	Amazon’s AWS Agentic Shopping Assistant posts a 3.5x conversion lift vs keyword search

What merchants should do this week

1. If you sell fashion in the UK or EU, treat Hey Savi as a live test of agentic discovery on home turf. This is the first European surface where the checkout completes in-app and the ranking is by relevance, not ad spend. Audit your product data the way you would for any agent: complete attributes, accurate pricing and inventory, real answers to the questions a shopper asks. A clean catalog is the whole entry ticket, and it works for a small brand as well as a large one.

2. Re-price your moat now that standing up a store is a conversation with an agent. Codex plus Wix means a competitor can go from idea to a payments-ready storefront without writing code. The website is no longer the asset. The catalog, the brand, and the customer relationship are. Spend this quarter on the things an agent cannot generate for free, not on the things it now can.

3. Build to the protocol layer, not to a single payment provider. Mastercard is consolidating card, agentic, and stablecoin settlement into one product family, and the other networks are doing the same. The merchant who connects to UCP or ACP inherits these rails as they switch on. The merchant who integrates one provider at a time pays for each one and lags on all of them.

4. Use Amazon’s 3.5x figure to make the storefront-agent decision deliberately. A branded on-site agent is now a weeks-long build, and the conversion lift is real. So is the dependency. Decide whether you want that agent on Amazon’s stack, on a more neutral one, or on your own platform’s version before you start the clock.