AI Commerce Weekly: Week 24, 2026
The card networks built the agentic rails, Salesforce shipped the platform answer, and UK shoppers said the quiet part: they still don't trust it.
TL;DR
Week 24 was the week the agentic-commerce stack got built out end to end, then ran straight into its ceiling. The card networks laid the rails: Visa wired OpenAI into its network so agents can pay inside ChatGPT, then shipped a merchant-side toolkit, while Mastercard opened the machine-to-machine end. Salesforce's Summer '26 release went generally available, the platform incumbent's answer to a year of point-solution agents. NielsenIQ pushed the same argument up to the product record itself. Tuesday delivered the bookends: Fynd brought AI design-to-production to Britain, and The Interline marshalled UK data showing shoppers deeply uneasy about personalisation. The plumbing is real now. Trust is the binding constraint on all of it.
Salesforce ships the platform answer
The week opened with a release date finally resolving. Salesforce's Summer '26 release became generally available on 15 June, headlined by Storefront Next and Agentforce Multi-Agent Orchestration. Storefront Next is pitched as an AI-first, enterprise-grade storefront a merchant can run without a standing army of developers. Multi-Agent Orchestration is the more telling piece: it lets specialised agents collaborate on one end-to-end workflow while the customer sees a single point of contact, with intent, context and history persisting across channels and across agents. A billing agent hands to a service agent hands to a retention agent, and nobody has to re-explain themselves.
For a UK Head of Tech this is a vendor-map update with teeth. All year we have watched point-solution retailer-agents productise one at a time, the AWS Agentic Shopping Assistant, Adobe Brand Concierge and the rest. This is the platform incumbent answering, and it reopens the build-vs-buy-vs-platform question for anyone already standing on the Salesforce stack. The honest caveat is the same one I keep writing: a GA date is not a UK reference customer. Until someone British puts their name to Storefront Next in production, treat this as a planning input, not a proof point. The thing to put on the roadmap review is whether your agent strategy now belongs to your platform vendor by default, and whether that is a decision you made or one that was made for you.
Visa and OpenAI put a wallet in the chatbot
The headline of the week came out of San Francisco. Visa announced a strategic collaboration with OpenAI that embeds its payment network into OpenAI experiences, so an AI agent can browse, compare and complete a real-money purchase on your behalf inside ChatGPT. Visa supplies the rails underneath: tokenised credentials bound to a specific agent and use case, real-time authorisation, agent identification and continuous fraud monitoring. These are, in Visa's own words, the same capabilities it runs across "more than 300 billion transactions annually", now extended into agentic environments through its Visa Intelligent Commerce portfolio. The buyer stays nominally in command through guardrails they set: spending caps, merchant-category restrictions and approval thresholds.
The scope reaches past shopping. The same primitives are pointed at business workflows and at OpenAI's Codex coding agent buying APIs or compute within limits, which is a quietly significant tell about where this is heading. Visa's Jack Forestell framed it large: "AI will transform commerce more profoundly than the internet or mobile technology ever did." Worth keeping next to that ambition is a sobering piece of recent history: OpenAI's earlier native Instant Checkout in ChatGPT struggled to gain traction and was scaled back. So read this as a rails-and-standards play whose real-world pull is still unproven, rather than a finished consumer product. The infrastructure being laid is real. Whether shoppers walk onto it is the open question the rest of this week keeps answering, and not flatteringly.
The merchant-side rails: scores, directories, machines paying machines
The consumer-payments story got the headline, but the supplier-side announcements matter more to a retailer's technology function. At the same Payments Forum, Visa unveiled Agent Score, an Agentic Directory and a Large Transaction Model. Agent Score, built with a startup called New Generation, lets a merchant evaluate whether an AI agent can actually navigate, understand and complete tasks on their site, which is a machine-readability audit for the agent era in all but name. The Agentic Directory is a registry of agents and merchants Visa has verified as legitimate, attacking the two-sided trust problem from both ends. The Large Transaction Model improves fraud detection while cutting false declines. Visa also reported a roughly $7bn annualised stablecoin settlement run-rate as of March, with Forestell's neat division of labour: "AI is transforming the front end of commerce. Stablecoins are reshaping the back end."
Mastercard opened the other end entirely. Agent Pay for Machines handles transactions between systems rather than between a person and a merchant: agents buying services from one another continuously, including microtransactions of fractions of a cent, settled at machine speed. More than 30 launch partners are named, several of them part of the UK and European payments stack, Adyen, Checkout.com, Stripe, Cloudflare and Getnet by Santander among them. Chief Product Officer Jorn Lambert called it the start of a "superbloom of AI business models... very high volumes, very small values, very fast", while being candid he expects little near-term revenue. For a Head of Technology this is the consequential half of the week's payments news, because the agent-buys-its-own-compute pattern lands in your engineering budget and your procurement-governance backlog, not in the shopping basket. Agent Score is the one I would action first: it tells you whether your storefront survives an agent's inspection, which is a question you can answer this quarter.
The adoption numbers, and the UK trust gap underneath
The demand side produced three datasets, and the gap between them is the whole story. On the optimistic end, a report carried by FashionNetwork found one in three e-commerce brands now use AI agents to drive shopping, with fashion the leading adopter: 57% exploring use cases, 33% actively preparing to deploy, and 87% expecting AI-powered search to drive sales growth over the next year. Take the precise $291,626 average-investment figure as a vendor number until the underlying report turns up, but the direction is not in dispute. AI has crossed from experiment into budgeted line item.
Then the British consumer answered back. Checkout.com's Agentic Commerce 2026 report maps a UK trust gap sitting under the hype: just 23% of UK consumers expect at least a tenth of their purchases to be AI-driven within a year against 33% globally, 41% trust no organisation at all to run an AI shopping agent against 27% globally, and 37% say they will never delegate purchases to AI. Even the willing want a short leash, comfortable letting an agent spend only around £156 before checking back. Only 3% of UK transactions involve AI agents today. Layer on PSE Consulting's survey of 4,250 consumers, reported by InternetRetailing, and the picture sharpens: 43% would take a free assistant whose advice is shaped by advertising over the 27% who would pay for an impartial one, but 48% of Brits say advertising would erode their trust in AI recommendations, against 40% globally. Sarah has written before about misreading an adoption curve, and the lesson holds here. "Brands are deploying agents" and "shoppers will use them" are two different sentences, and this week the British consumer underlined the gap in red.
Make your product legible, or go invisible
If agents are going to decide what gets surfaced and bought, the unglamorous prerequisite is whether they can read your product data at all. NielsenIQ expanded NIQ Product Intelligence with a GDSN capability to deliver a single authoritative product record that travels across the physical supply chain, the digital shelf and agentic-commerce ecosystems. GDSN is the GS1-governed standard retailers and suppliers already use to exchange validated master data, so this is less a new idea than an existing plumbing standard pointed at a new consumer: the machine. NIQ's pitch is blunt. Products that lack structured, enriched, machine-readable data risk becoming invisible to recommendation engines, AI assistants and autonomous purchasing agents.
The asset NIQ is leaning on is scale: relationships with more than 8,900 retailers across 90 countries, a catalogue of 246 million-plus unique items and over 10 billion maintained product attributes. Read it alongside Visa's Agent Score and the shape of the week becomes clear. The same problem, is your product legible to a machine that is deciding what to buy, is being attacked from two ends at once, the checkout end and the master-data end. For a UK technology leader this is the least glamorous and most actionable item of the week. You do not need shopper appetite to fix attribute hygiene, and unlike most of this week's announcements, it pays off whether or not agentic checkout ever takes off, because clean machine-readable product data improves every discovery surface you already have.
Fynd brings the supply side to Britain
Almost everything else this fortnight has been about the front of house. Fynd launched Fynd Create in the UK, and it points the other way. Backed by Reliance Retail Ventures, the platform pulls trend intelligence, design, sourcing, cataloguing and logistics onto a single AI-powered stack, reading social trends, runway influences and competitor activity to compress the concept-to-production cycle. Founder Sreeraman Mohan Girija makes the speed-to-market case plainly: "trends emerge and evolve in a matter of days, yet many supply chains remain constrained by long planning cycles and fragmented processes." It folds in Fynd Snap, a generative engine that produces photorealistic on-model imagery from flatlays or 3D renders, the kind of catalogue work that traditionally means booking a photoshoot. Having started my own career in print design and supermarket aisles, I have a particular soft spot for tooling that attacks the bit between "we should do a drop" and "the garment exists".
Treat the claimed "up to 60%" design-productivity gain as a vendor figure until someone independent stands behind it. The interesting thing is the direction of travel, not the number. Fynd is aiming at the slow, fragmented, spreadsheet-bound front half of the fashion value chain, where the productivity case for AI is arguably easier to make and harder to argue with than it is on the storefront, because the baseline is genuinely manual. There is no named UK customer yet, which is the obvious thing to watch, and the integration question is the one to ask before the demo dazzles the merchandising team: a unified design-to-production platform is a lot of your value chain sitting on one vendor's stack. Decide deliberately what stays in your own systems.
The personalisation people don't trust
The week closed on the sobering counterpoint, and it is the more important read for anyone signing off a personalisation roadmap. The Interline argues fashion wants to personalise the experience and shoppers do not trust the data exchange that requires, citing new Chartered Institute of Marketing research drawn from more than 2,000 UK shoppers: 78% are uncomfortable with AI having access to their personal data, and 73% want AI-driven advertising more tightly regulated. The piece is honest about the contradiction, since the same people are merrily switching on chatbot memory and feeding intimate signals to the apps now brokering intent hand-offs to brands. But the regulatory teeth are visible. The Interline points to South Korea's Coupang, fined more than $400m this week per Reuters over a breach and a separate illegal data-grab covering some 11 million customers.
The framework The Interline leaves behind is genuinely useful for a technology leader. There are three modes of personalisation and two are mostly shut: experience personalisation is a data-governance minefield, product personalisation is out of reach for traditional models, which leaves on-demand manufacturing, and only the giants can do it. The proof point is Amazon, which this week let Alexa for Shopping generate artwork, print it on a blank tee and Prime it to your door. Put this beside Fynd and the steer for a fashion technology team is almost comically clear. The safe, defensible AI investment right now lives in the parts of the business the customer never sees: design tooling, attribute hygiene, sourcing intelligence, generated catalogue imagery. The customer-facing personalisation engine is where the regulatory and reputational exposure sits, and the data this week says shoppers are watching. If you want to behave like a technology company, you need the data governance a technology company is held to, and most fashion brands do not have it yet.
Sources
Salesforce ships the platform answer
Visa and OpenAI put a wallet in the chatbot
The merchant-side rails: scores, directories, machines paying machines
The adoption numbers, and the UK trust gap underneath
- One in three e-commerce brands now use AI agents to drive shopping, fashion leads, FashionNetwork UK, 10 June 2026
- Consumer demand for AI shopping is forming fast but cautious Brits are still not ready to give up control (Checkout.com Agentic Commerce 2026), Retail Times, 9 June 2026
- Trust gap slows UK take-up of AI shopping agents (PSE Consulting survey, 4,250 consumers), InternetRetailing, 11 June 2026
Make your product legible, or go invisible
Fynd brings the supply side to Britain
The personalisation people don't trust
Stay Connected
Follow LLCommerce on LinkedIn
Get the latest AI commerce insights, analysis, and industry news delivered to your feed.
Large Language CommerceAuthors
