BriefingsWeekly Briefing15 min read

AI Commerce Weekly: Week 23, 2026

The UK's first agentic checkout, three consumer datasets that agree, resale's trust problem, and Kingfisher's line: technology can be bought, trust has to be earned.

Topics

agentic-commerce uk-retail consumer-behaviour payments ai-search events

9 June 2026

TL;DR

Week 23 ran the whole arc of agentic commerce and closed on one idea: when the models commoditise, trust is the moat. The UK got its first native agentic checkout, Hey Savi and PayPal with Debenhams Group as the launch retailer. Two more retailer-agent stacks productised with named customers: Adobe Brand Concierge behind Coach by Dick's, and Gopuff's Go on SpaceXAI. Three UK consumer datasets landed in the same week and broadly agreed AI now owns research while price still owns the sale. The plumbing showed itself, from Selltonomy's buyability audit to the Affirm and Stripe UK rail. Resale's provenance problem surfaced three times. And Kingfisher's Chief AI Officer said the quiet part out loud.

The UK gets its first agentic checkout

The headline UK event of the week was a checkout button. Hey Savi and PayPal launched what they call the UK's first agentic commerce experience with native in-app checkout, with Debenhams Group as the first UK retailer to bring end-to-end AI-powered shopping to British consumers. Hey Savi is a brand-agnostic AI fashion search platform for women: search by photo, screenshot or text, get results ranked by relevance across more than 10,000 brands, and crucially, ranked by relevance rather than by who paid for the slot. PayPal's Agentic Commerce Services supplies the payment and checkout layer. Debenhams Group here means Debenhams, Karen Millen, Boohoo and Pretty Little Thing.

Read it as the W23 escalation of a thread we have been following since W21, when PayPal's ACS pilot with Debenhams Group first surfaced. The new bit is the native checkout: the purchase completes inside the agentic surface rather than bouncing the shopper back to a retailer site to finish. That is the line everyone has been circling, because it is the line where the demand relationship genuinely changes hands. "UK's first" is the vendors' own framing and I would hedge it lightly, but the direction is not in doubt. The bit worth chewing on is the ranking promise. A relevance-ranked, non-sponsored result set is a real pitch to shoppers and a real problem for retail-media economics at the same time. If the agent won't sell placement, the thing you used to buy your way to the top of now has to be earned on merit, which is a theme that runs right through this week.

Two more retailer-agents go on sale

The platform map widened twice more. Dick's Sporting Goods is rolling out Coach by Dick's, an agentic assistant built on Adobe Brand Concierge, in its mobile app from this month, with Adobe Experience Platform and Adobe Journey Optimizer underneath it for data and orchestration. The same release carried a number worth keeping: Adobe says AI traffic to retail sites jumped 693% year-on-year over the 2025 holiday season. On the same Tuesday, Gopuff and SpaceXAI launched Go, a predictive shopper built into the Gopuff app, which prepares a suggested cart the moment you open the app, generates a TikTok-style shoppable feed from a custom image model, takes voice instructions through Grok, and delivers in as fast as 15 minutes from 400-plus micro-fulfilment centres.

For a UK Head of Tech the read is a vendor-map update, not a deployment moment. Adobe Brand Concierge now sits alongside the AWS Agentic Shopping Assistant (Kate Spade, last week) as a productised, buy-it-off-the-shelf retailer-agent with a named live customer. Stack that on top of Google's UCP, OpenAI's ChatGPT apps and Meta's in-app checkout and the platform-of-record slide is now six providers deep. The conspicuous gap is the same one as last week: none of these has a marquee UK-fashion reference customer attached. Gopuff Go is the more interesting object lesson. It is the clearest expression yet of the "cart before you ask" model, where the agent's job is to remove the search box entirely. It is US-only at launch, but Gopuff trades here, so UK availability is a watch item rather than a curiosity.

Three UK datasets, one shape

Three pieces of UK consumer research landed inside the week, and the useful thing is how closely they rhyme. Athos Commerce and Drapers' Connected Consumer 2026 found that 60% of UK consumers now use AI tools at least occasionally while shopping for fashion, with 39% comfortable buying fashion directly through an AI platform, and a generational gulf underneath that headline: 50% of Gen Z against 21% of Gen X. Only 14% say product details always match across social, marketplace and retailer channels. Akeneo's PX Pulse put the research-versus-purchase split in sharp relief: 43% have used AI to help shop or find deals ahead of Prime Day, yet 55% say final price is the single biggest influence on a purchase and just 3% name AI recommendations as the deciding factor.

The third one is the sharpest for a fashion audience. Marketing Signals' UK Fashion AI Visibility Index shows size and AI visibility have come uncoupled. Harrods tops the table of 59 retailers, driven almost entirely by editorial citations; Whistles comes second on just 217,000 monthly UK visits; Next finishes dead last despite 21.6 million monthly visits. The bottom ten retailers pull more than 54 million monthly visits between them and average a visibility score of 2.9. Put the three together and the curve is unambiguous: AI has moved firmly into research and discovery, it has not yet become the decision layer, and machine-readability now matters independently of how big your brand is. Sarah has made the point before about misreading an adoption curve, the Shopee Live miss in 2023, and the lesson holds here. "Used AI while shopping" is not "bought through AI", and the gap between the two is exactly where the next two years of work sits.

Can an agent actually check out?

If agents are going to do the buying, the unglamorous question is whether your storefront will actually let them. Selltonomy launched an "AI Buyability" platform that audits exactly that, checking a storefront across four layers, schema markup integrity, variant structures, price interpretation and checkout protocols, to see whether agents from ChatGPT, Gemini, Perplexity and Copilot can complete a purchase rather than just recommend a product. Their term for the gap is "Silent Failures": purchases that die before your analytics or abandoned-cart tooling ever see them. The launch leans on Shopify's Q1 2026 numbers, which are worth repeating with the right attribution: Shopify reports AI-referred shoppers converting around 50% higher than organic search, carrying 14% higher average order value, with AI-attributed orders up roughly thirteenfold year-on-year.

The payments rail moved in the same direction. Affirm and Stripe extended their North American partnership to the UK, so from July, British businesses on Stripe can add Affirm pay-over-time to checkout for the first time. The supporting numbers come from the US and Canada, where Stripe merchants enabling Affirm reported a 13.9% average revenue uplift on eligible sessions and a 21.3% conversion uplift on purchases over $250. The part that matters for this newsletter is the agentic arm: the two have committed to supporting Shared Payment Tokens, the mechanism that lets an agent transact pay-over-time without ever handling the card credentials. Buyability and agent-safe payment tokens are the same conversation from two ends. One asks whether the agent can read your catalogue; the other asks whether it can pay without breaking your security model. Both are the sort of thing that never makes a keynote and always makes the difference.

Resale, provenance, and a Sheffield mirror

Resale kept surfacing, and it kept surfacing as a trust problem. Optical-AI company Alitheon raised an $8m Series A1 led by Emerald Technology Ventures, with eBay Ventures participating. Its FeaturePrint technology reads an object's surface detail through a standard camera to create an unforgeable identity, "biometrics for things" under a Zero Trust framing where nothing is assumed authentic until verified. eBay Ventures writing the cheque tells you where the thesis sits: provenance in high-value recommerce. Then DRESSX teamed with eBay-owned Depop to put AI virtual try-on onto resale stock, starting with the DRESSX Mirror at Governors Ball in New York. The technically interesting detail is not the mirror, it is that the try-on renders over seller-shot, user-generated imagery rather than clean studio shots. That is the resale variant-explosion problem made visual, and if try-on works on the messy tail it works everywhere. Depop is UK-born, so the read-across to the British resale market is direct.

There was a physical-floor entry too, and it landed on home turf. 3DWD's Style Suite made its global debut at Meadowhall in Sheffield, an AI colour-analysis and styling tool you unlock from your phone, with results saved to a wallet-based profile that travels into the stores around the centre. I started out in print design and supermarket aisles, so a colour-analysis kiosk in a Northern shopping centre is the kind of thing I have a soft spot for. The serious point underneath it: provenance, fit and styling are all variants of the same question a shopper, or an agent acting for them, has to be able to trust the thing in front of them is real, will fit, and is what it says it is. Resale is where that question is hardest, which is exactly why it keeps leading this feed.

Event week: Olympia and Barcelona

The diary collided, as it tends to in June. London Tech Week opened at Olympia with an AI-dominated, sovereign-AI-centred agenda, Perplexity's Aravind Srinivas among the headline founders alongside enterprise deployment voices from Evri, Unilever, Heineken and Booking.com, with AWS and Microsoft as headline partners. The framing was sovereign AI, models and infrastructure not dependent on US or Chinese platforms, which is a worthy debate but a notch more abstract than a retail buyer needs. The W23 question to carry into next week is whether anything concretely commercial for UK retail came off the Olympia stages, or whether it stayed at the level of national strategy.

Meanwhile Shoptalk Europe opened in Barcelona with agentic commerce on the main stage, 4,500-plus professionals, roughly one in three of them C-suite. The session to mine is "Agentic Commerce: What's Here, Real, and Next?", pairing L'Oréal's Mark Elkins with Google's Dido Schmidt, and Vestiaire Collective's new CEO Bernard Osta setting out an AI-authentication play tied to the platform's profitability targets. Most UK teams had to make a delegate call between the two this week. The honest read is the same one I keep coming back to about the Northern Quarter office: the local convening matters when it is local, even when the international one looks louder on the day. Send one person to each, get transcripts both ways, and treat anything from Barcelona that isn't yet attached to a number as still-to-verify until W24.

Kingfisher's throughline: trust is the moat

The week closed on the clearest UK-leader articulation of the year. Kingfisher's Chief AI Officer Mohsen Ghasempour argued that AI is commoditising the very advantages it was meant to create. Writing after a Financial Times panel on loyalty alongside Bayes Business School and Unilever, the Kingfisher AI lead (the group behind B&Q and Screwfix here, Castorama and Brico Depot in France) made the case that personalisation, AI assistants and instant answers were genuine differentiators a few years ago and are now table stakes, which raises customer expectations to the level of the best experience they have had in any industry, not just yours. Loyalty, he argues, has shifted from being driven by memory to being driven by present-day expectation; convenience is becoming a commodity. The commercial backdrop is Kingfisher's multi-year Google Cloud partnership, upgrading to Vertex AI Search for Commerce to build agentic shopping beyond keyword search.

That line is the throughline for the whole week. Hey Savi promises a relevance-ranked result you can trust over a paid one. The Marketing Signals index measures who the machines find credible. Alitheon and Vestiaire are selling provenance you can verify. Selltonomy and Shared Payment Tokens are about whether an agent can transact without breaking faith with your security. Strip the announcements back and they are all asking the same question Ghasempour names: once everyone has the same models, what is left that is hard to copy? The answer keeps coming back the same, and it is not the model.

What to do this week

Treat native agentic checkout as live in the UK, not theoretical. Hey Savi + PayPal + Debenhams Group means the purchase now completes inside the agentic surface. Put it on the next board paper as a present-tense option to plan against, and decide deliberately whether you want to be in or out.
Audit your own buyability before a vendor does it for you. Run a manual check of whether ChatGPT, Gemini, Perplexity and Copilot can actually complete a purchase on your storefront, not just describe a product. Schema integrity, variant structures, price interpretation and checkout are the four layers to test.
Add the Affirm-on-Stripe UK rail and Shared Payment Tokens to the payments roadmap review before July. The agent-safe token path is the part that matters for agentic commerce; make sure your payments team has read it.
Benchmark your AI visibility, then act on the four levers. The Marketing Signals index shows traffic does not buy you machine credibility. Editorial citations, forum presence, unlinked mentions and schema-plus-a-strong-About-page are where the work is.
Pressure-test any AI business case that still rests on personalisation as a differentiator. Steal Ghasempour's framing: how much of that edge survives once competitors ship the same off-the-shelf models? Re-anchor the thesis on consistency and trust, the parts that do not commoditise.
Scope virtual try-on against your messiest imagery. The DRESSX x Depop work shows try-on escaping the studio PDP onto UGC stock. If you are scoping try-on, test it on your worst, highest-variant images now.
Mine Shoptalk Europe and London Tech Week for numbers, not vibes. Send one delegate to each, transcripts both ways, and carry only the claims that arrive with evidence attached into W24.

Sources