AI Commerce Status Report Q2 2026

From Promise to Division of Labour

This is the second instalment in a series begun in February 2026. It now extends the stock-take first published on 5 May 2026 with the decisive moves through the end of the quarter. It covers only those initiatives publicly announced or rolled out in production between 1 April and 28 June 2026 whose technical substance can actually be verified, and whose mechanics rest on genuine generative or agentic AI rather than a generously rebranded recommender system. Nothing from the Q1 Report is repeated for its own sake. Earlier themes reappear only where the second quarter has produced something materially new to say about them. An overview of all reports published so far is available on the Status Reports hub.

Executive Summary

The second half of the quarter has sharpened the picture considerably beyond what the May stock-take suggested. Two developments now define the period, and neither sits comfortably under the old vocabulary of polite consolidation.

First, the architecture of agentic commerce has visibly split in two. Discovery and transaction are coming apart, and they are increasingly operated by different parties. Walmart disclosed in March that its native conversion rate inside ChatGPT ran roughly three times below its own website, after which OpenAI quietly retired Instant Checkout in its first form. The June relaunch no longer runs on OpenAI's own payment stack but on Shopify's agentic infrastructure. Perplexity has chosen, with some deliberation, to remain a discovery layer with no transaction of its own. Shopify moved the Universal Commerce Protocol and the Catalog API into self-service with Spring 2026, Salesforce acquired the AI-native search engine Cimulate, and BigCommerce found itself functionally reduced to a feed provider for external discovery surfaces. Visa and Mastercard, on the same day, 10 June, took up the trust and identity layer for agentic transactions in what looked suspiciously like a coordinated manoeuvre. Europe's first production agentic payment was settled on 2 June in the Netherlands by Worldline, ING and Mastercard.

Second, the first documented cull has begun. Ssense dissolved its in-house photography and make-up teams in June and named AI explicitly as the substitute. OpenAI's first Instant Checkout failed visibly and had to be rebuilt on someone else's infrastructure. BigCommerce is defending an agency channel position that has slid, structurally, from full-stack provider towards data plumbing. commercetools posted a recovery quarter while quietly retiring the word composable and re-emerging under Autonomous Commerce.

The two developments are linked. Anyone who did not, in Q2, decide whether they are a platform, a product or a distribution layer will find that question answered for them in Q3, along the line that now divides discovery from transaction. The long-promised breakthrough of fully autonomous consumer agents has, again, declined to arrive. What has hardened instead is a division of labour in which brands must decide separately about visibility in the discovery layer and margin control in the transaction layer.

Methodological Note

The overview below is drawn from publicly documented announcements, press releases, peer-reviewed studies and reporting by established trade media. Only initiatives whose technical architecture stands up to inspection, and whose AI component goes meaningfully beyond rule-based automation or classical recommender systems, have been included. Where figures come exclusively from vendor marketing, this is flagged. Pure repositioning without new functional substance has been left out.

The Discovery and Transaction Split Becomes Architectural

The split between agentic discovery and transaction, still treated as a hypothesis in the Q1 report, has firmed up into an architectural fact over the course of the quarter.

On 24 March 2026 Walmart's EVP Daniel Danker disclosed that conversion inside ChatGPT's Instant Checkout ran at roughly a third of conversion on Walmart's own website.¹ A few days later OpenAI retired Instant Checkout in its original form and turned the shopping experience into a discovery-oriented format with no native purchase step.² By June, Instant Checkout had returned in a second incarnation, this time running not on OpenAI's stack but on Shopify's agentic infrastructure, with Spanx, Skims and Glossier as the first named merchants.³ OpenAI has, with no great ceremony, given up the role of merchant of record and become a distributor.

This is not an isolated move. Perplexity has continued to build out the Merchant API it opened in February. Operator briefings put it at roughly two billion dollars in annualised GMV, with conversion rates reported by merchants at two to three times that of Google Shopping clicks.⁴ No primary Perplexity announcement confirms those figures within the quarter, so the numbers are best read as directional rather than precise. The strategic posture is what matters. Perplexity has chosen, deliberately, to keep the transaction at the merchant and to restrict itself to discovery, on the explicit argument that the merchant should remain the owner of the customer relationship.

Shopify moved the Universal Commerce Protocol and the Catalog API into self-service with Spring 2026 on 17 June.⁵ Any developer can now register without approval, where the Q1 report still described a gated programme. Production MCP servers for Shopify and WooCommerce went live at the same time, in principle opening more than ten million storefronts to any MCP-compatible agent. Salesforce released Agentforce Commerce with three agents in general availability on 24 June, the core of which is the acquisition of the AI-native search engine Cimulate, now made available as Agentic Commerce Search on non-Salesforce sites as well.⁶

BigCommerce, in the same period, quietly changed shape. The Agentic Catalog layer opened in April via Feedonomics and the Q2 build-out of distribution to ChatGPT, Gemini, Microsoft Copilot and Perplexity describes a shift away from full-stack provider towards data and feed layer for external discovery surfaces.⁷ The shift is pragmatic, but it reduces value capture per transaction, because the transaction itself increasingly takes place on someone else's platform. Shopware, in parallel, used its Community Day on 10 and 11 June to unveil Nexus, Copilot, Payments and Experience Studio, four products that directly address the integration cost critique of composable architectures and that, with Shopware Payments, introduce a platform-owned monetisation of the transaction layer for the first time.³³

Visa and Mastercard placed themselves on opposite sides of the same announcement on 10 June, in what looked suspiciously like prepared choreography. Visa unveiled its Intelligent Commerce suite at the Payments Forum, comprising Agent Scoring for transaction risk, an Agentic Registry for verified authorised agents and a Large Transaction Model trained on the pattern signatures of agentic transactions.⁸ On the same day Mastercard launched Agent Pay for Machines with over thirty launch partners, including Stripe, Adyen, Cloudflare and Coinbase, as a formal product for fully autonomous machine-to-machine transactions.⁹ Eight days earlier Worldline, ING and Mastercard had settled the first production agentic payment in the SEPA zone in Amsterdam.¹⁰ Google had already donated the Agent Payments Protocol to the FIDO Alliance on 28 April, removing the authentication layer from platform control and placing it under neutral governance.³⁴

What this movement amounts to is not consolidation in the polite sense but an architectural separation. The discovery layer is settling into open, multi-protocol surfaces in which brands become visible through structured product data and answerability. The transaction layer is meanwhile settling into trust infrastructure with identity registries and agent-specific risk scoring, increasingly controlled by the card networks. The two layers are linked in practice but no longer operated by the same actor.

The First Cull Becomes Visible

Q2 is the first quarter in which the cull can be documented by name rather than predicted in the abstract. Four examples are sufficiently evidenced to be read as a pattern rather than as oddities.

OpenAI's first version of Instant Checkout, launched in September 2025 with a great deal of attention, was retired in March 2026 after the largest named merchant confirmed a meaningfully worse conversion.¹ The relaunch in June runs on somebody else's infrastructure. This is the first cleanly documented correction of a platform bet of the 2025 generation.

Ssense laid off its in-house photography and make-up teams on 22 June 2026 and named AI explicitly as the substitute in its public communications.¹¹ This is the first cleanly documented dissolution of an in-house production team in the premium segment with explicit reference to generative imagery. Zalando had earlier confirmed publicly that roughly ninety per cent of its content is now AI-generated, with seventy per cent higher output at flat cost and a campaign cycle compressed from eight weeks to a few days.¹² Zalando is the benchmark, Ssense is the first consequence. Debenhams justified a small number of UK redundancies in spring 2026 by explicit reference to AI image production, a single-digit move but unmistakeably in the same direction.¹³

commercetools reported in April that the first quarter of 2026 had crossed an annualised GMV of over one hundred billion dollars, with same-store growth of twenty-seven per cent and over seven billion dollars of new business in the quarter.¹⁴ The recovery follows the restructuring of February 2025 and a change of leadership. In the same breath commercetools subordinated the word composable as its organising principle and re-emerged under the heading Autonomous Commerce, with AgenticLift, the Smart Data Modeler and the AI Hub as its first products of the new generation. The recovery is real, the repositioning is honest. Together they mark the end of the composable marketing era.

Spryker closed a follow-on round in December 2025 in which only existing investors participated and the amount was not disclosed.¹⁵ Set against the rapid Q2 moves by Shopify and Salesforce, that is not necessarily a problem, but it is a signal that the B2B composable camp can no longer secure a meaningful valuation step-up.

None of these are insolvencies and none are mass layoffs. They do, however, show the same pattern across very different segments of the market. Anyone who failed in Q2 to decide whether they are a platform, a product or a distribution layer will begin in Q3 to drift into one of those roles, almost always in favour of the more dominant layer.

The K5 Conference, Read with Quiet Candour

The K5 Conference on 23 and 24 June in Berlin falls inside this quarter and offers a useful counterpoint. Three of its more prominent theses deserve careful weighing, because they understate the picture of the discovery and transaction split in several respects.

The thesis that LLM traffic at roughly 0.3% is disappointingly small and therefore manageable as a side topic describes the position from late 2025 into the first months of 2026 quite accurately. A peer-reviewed study in Marketing Science confirmed a ChatGPT share of under 0.2% of e-commerce traffic for the observation window August 2024 to July 2025 across 373 sites.¹⁶ Adobe Analytics, however, reported on 17 June 2026 that AI referral traffic grew 138% year-on-year in May 2026, and 1,324% since October 2024. In the same report Adobe documented an inversion of conversion quality. AI traffic now converts 54% better than non-AI traffic, where the same ratio had been 50% worse a year earlier.¹⁷ Conductor's enterprise benchmark across more than 13,000 domains finds an aggregate share of 1.08%.¹⁸ A realistic 2026 range for e-commerce sits between 0.5% and 1.5%, rising to around 3% in information-dense categories such as cosmetics and electronics. The K5 thesis measures the right quantity at the wrong time and ignores both the inversion of conversion quality and the documented dark traffic effect, in which AI-induced visits appear in standard analytics as direct hits.

The KPI thesis described as Habit Control, the proposition that brands should manage the frequency and routine of their use, is not, strategically speaking, a new mechanic. It describes retention and frequency under a fresher label. An industry that in the same period has lived through the separation of discovery from transaction does not gain agency from this choice of vocabulary, because the question is no longer how often a customer comes back, but in which layer they discover and in which layer they buy.

The thesis of production-led commerce against the classical demand-led logic, developed in Stefan Wenzel's K5 contribution, is operationally correct. Shein's five-day trend-to-production cycle, its in-house brands such as Sheglam, the Xcelerator programme and the logistics investments in Wroclaw and Zhaoqing all evidence the production-side depth. The thesis does, however, understate the speed at which the platforms are closing their one structural weakness. Shein's acquisition of Everlane in May brings established brand credibility into the portfolio.¹⁹ TikTok Shop added four new EU markets on 15 June and introduced Sell Across Europe as a single registration covering ten countries.²⁰ Temu received a DSA fine of two hundred million euros on 28 May, raising compliance costs significantly.²¹ The EU has introduced a three-euro flat duty on parcels up to one hundred and fifty euros from 1 July, narrowing the price gap.²² The production-led logic remains correct. The further claim that it is structurally superior to Western demand-led logic is, at the same time, being undermined by the platforms themselves, which are buying their weaknesses rather than enduring them.

Content and Cost in the Production Apparatus

The movement in image production has passed from experiment into operation during the second quarter. Zalando has built out its position as the European reference point, with forty-five thousand high-fidelity 3D twins in footwear in partnership with NVIDIA and ALLSIDES, and with the co-CEO's June statement that roughly ninety per cent of Concept Store content is now AI-generated.¹²²³ Otto has externalised its in-house MOVEX Virtual Content Creator, positioning itself alongside Claid, Pebblely and Booth.ai, with internally reported cost reductions of sixty to ninety per cent.²⁴ Those figures are vendor-reported and have not been independently audited.

Amazon began on 3 June to insert AI-generated contextual images directly into search result pages, without labelling those images as AI-generated.²⁵ Control over the visual presentation of the product is therefore moving onto the marketplace SERP and is partially leaving the merchant. Adobe took the Firefly AI Assistant into public beta in April and folded it into Photoshop and Premiere in June.²⁶ Google extended Product Studio with image-to-video via the Veo model and with outpainting, and moved Imagen 4 into general availability.²⁷

On the regulatory side EuroCommerce, whose members include Amazon, H&M, Inditex and Ikea, went public on 19 June with a formal request to exempt AI-generated advertising content from the labelling obligations under Article 50 of the EU AI Act, which take effect on 2 August 2026.²⁸ No decision had been issued by quarter end. In the United States, model Francheska Pujols's lawsuit against Rainbow Shops, filed on 22 May and refiled after a brief withdrawal on 15 June, has produced the first documented test of whether existing model contracts authorise AI-generated derivatives of the original shoot without further consent.²⁹

The reading is sober. The operational substitution of human image production has arrived in the premium segment, documented by Zalando and Ssense. The regulatory clarification, without which this substitution cannot scale in Europe without risk, is immediately ahead. 2 August is the next date that matters.

Chinese Platforms, Brand Rather Than Price

The convergence between Chinese and Western platforms sketched in the Q1 report has gained noticeable speed in the quarter. Three movements stand out.

TikTok Shop went live in Austria, Belgium, the Netherlands and Poland on 15 June, bringing ten European markets under a single merchant account.²⁰ Sell Across Europe materially lowers the barrier to cross-border EU selling and puts the programme in direct competition with Amazon's Pan-European FBA and Shopify Markets. Any Western e-commerce strategy in 2026 that ignores TikTok Shop is leaving on the table a channel whose growth curve has been triple-digit since August 2025.

Shein took direct management of its Dublin logistics facility on 8 June and launched Xcelerator in the Middle East on 16 June.³⁰³¹ The Everlane acquisition in mid-May is the strategic break in this sequence. Shein is, for the first time, systematically buying established brand credibility rather than building it algorithmically.¹⁹ In the same period Sheglam crossed four hundred million dollars of revenue, with a documented trend-to-SKU pipeline running through TikTok.³²

The EU three-euro flat duty from 1 July and the Temu DSA fine of two hundred million euros from 28 May mean that the regulatory asymmetry that flattered the cross-border model for years is beginning to close.²¹²² Platforms with local inventory absorb the duty more easily than pure cross-border flows. Which is precisely why Shein and Temu have spent recent quarters investing so consistently in EU logistics.

The reading for Western retailers is not that the gap is closing, but that the playing field has shifted. The price gap is being softened by regulation, the infrastructure gap is closing through investment, and the brand gap is closing through acquisitions. Anyone defending in 2026 on price or delivery speed alone is defending a line that has already moved.

Productive Assistant Agents Inside the Retail Surface

The wave of own-brand assistant agents inside the retail channels of established retailers, documented in the May stock-take, has held through the rest of the quarter. Macy's rolled out Ask Macy's on Gemini Enterprise in about four weeks and reported beta revenue per visit running roughly four and a half times higher among shoppers using the assistant than among those who did not.³⁵ Ulta Beauty and Cognizant released comparable implementations on the same day, all built on Gemini Enterprise for Customer Experience.³⁶³⁷ The three initiatives confirm the split. Discovery and value capture stay with the retailer, model infrastructure comes from the platform provider. What they do not answer is the question of which discovery layer beyond the retailer's own brand surface the retailer also needs to be findable in.

In B2B, SAP with the next generation of Ariba, Coupa with its May release and Sequentum together with Skyfire have provided the sober evidence that agentic procurement has reached productive implementations.³⁸³⁹⁴⁰ The conditions for authorised agentic action are easier to articulate in B2B than in the consumer market. The first scaled productive applications will therefore likely surface there, well before any consumer agent reaches operational meaning. Feedonomics has introduced Agentic Catalog Exports with Dell as the first named adopter, sharpening the conversation around agent-ready product data.⁴¹

What Can Be Drawn From the Quarter

Three consequences are sufficiently well evidenced to be stated as operating logic.

First, the question of agentic architecture is no longer a technical choice but a strategic one. Brands have to decide their position in the discovery layer and their position in the transaction layer separately. The discovery layer decides visibility and therefore demand, the transaction layer decides margin control and customer relationship. The two decisions are no longer covered by a single platform choice. Choosing Shopify does not answer in which discovery surfaces a brand turns up. Running an own headless system does not answer which trust infrastructure secures the brand's agentic transactions.

Second, the cull has begun. It runs quietly, as an architectural shift, rather than loudly. One actor becomes pure data plumbing, another gives up the role of merchant of record, a third dissolves its in-house production team. The pace is slow enough to be steered and fast enough not to be sat out. Q3 will, in all likelihood, produce the next examples, in the platform layer as much as in production.

Third, the operational lead in agentic commerce continues to move into B2B. Source-to-pay, procurement and reseller logistics offer the clearer authorisation conditions and are therefore the first scalable market for autonomous agentic transactions.

Implications for Retailers

The strategic situation for retailers has not only sharpened, it has shifted. Three areas of action stand out.

The first concerns visibility in the discovery layer. It is decided by agent-ready product data, structured answerability and presence on the relevant surfaces, from ChatGPT and Gemini through Perplexity to the Amazon SERP. Anyone who does not regularly check how they are described and with what priority they are returned loses visibility without noticing. A fuller methodology is set out in the piece on Decision Journey Analytics.

The second concerns the choice of transaction layer. Three options are visible. The first is the brand's own surface with an assisting agent on a hosted model platform, as Macy's, Ulta and Cognizant demonstrated in April. The second is the platform logic of Shopify or Salesforce, with native trust infrastructure across UCP, ACP and the FIDO authentication layer. The third is distribution through external discovery surfaces, as BigCommerce has done through Feedonomics. The three options are not mutually exclusive, but they require different organisational and contractual conditions.

The third concerns governance. Regulatory clarification is proceeding faster than most organisations clarify their internal responsibilities. EU AI Act Article 50 takes effect on 2 August, the UK FCA has named agentic payments a 2026 supervisory priority, and the Rainbow lawsuit in New York is defining the frame for AI-generated model imagery.²⁸⁴²²⁹ Anyone without a documented internal position in those areas is carrying a risk that will become visible in Q3. A fuller discussion of the organisational preconditions can be found in the piece on strategic consequences for commerce leaders.

The Q3 report should, with some luck, be in a position to indicate whether the cull moves into a more clearly stated second phase, whether the second version of Instant Checkout launched in June produces productive conversion data, and how the European Commission rules on the EuroCommerce request.

Sources

¹ Search Engine Land, "Walmart Says ChatGPT Checkout Converted Worse Than Its Own Site", 20 March 2026, searchengineland.com.

² CNBC, "OpenAI Revamps Shopping Experience in ChatGPT After Instant Checkout", 24 March 2026, cnbc.com.

³ Paz.ai, "OpenAI Instant Checkout Relaunch on Shopify", 15 June 2026, paz.ai.

⁴ Ecommerce Times, "Perplexity Commerce Hits 2B GMV Run Rate After Merchant API Push", 8 June 2026, ecommerce-times.com.

⁵ Shopify Newsroom, "Spring 26 Edition for Developers", 17 June 2026, shopify.com.

⁶ Salesforce, "Agentforce Commerce Announcement", 24 June 2026, salesforce.com.

⁷ BigCommerce Investor Relations, "Commerce Introduces New Product Innovations Across Storefronts", 30 April 2026, investors.bigcommerce.com.

⁸ Visa Investor Relations, "Visa Announces New AI, Stablecoin and Token Innovations", 10 June 2026, investor.visa.com.

⁹ Mastercard, "Mastercard Launches Agent Pay for Machines", 10 June 2026, mastercard.com.

¹⁰ GlobeNewswire, "Worldline and ING Complete a Live End-to-End European Agentic Payment in Production", 2 June 2026, globenewswire.com.

¹¹ The Logic, "Ssense Layoffs, Photography, Creative, AI", 22 June 2026, thelogic.co.

¹² RetailNews.ai, "Zalando Co-CEO David Schroeder on AI in Fashion Retail Europe", 15 June 2026, retailnews.ai.

¹³ Drapers Online, "Redundancies at Debenhams Group Amid Shift to AI Imagery", 15 April 2026, drapersonline.com.

¹⁴ commercetools, "commercetools Reports Strong Q1 2026", 22 April 2026, commercetools.com.

¹⁵ GlobeNewswire, "Spryker Secures Capital Boost to Accelerate Market Leadership in B2B Commerce", 11 December 2025, globenewswire.com.

¹⁶ Marketing Science, "ChatGPT Organic Traffic to E-Commerce Sites", peer-reviewed, May 2026, DOI 10.1287/mksc.2025.0489.

¹⁷ Digital Commerce 360, "AI Referral Traffic Hits New Peak in May, Adobe Analytics", 17 June 2026, digitalcommerce360.com.

¹⁸ Conductor, "2026 AEO and GEO Benchmarks", 15 April 2026, conductor.com.

¹⁹ CNN Business, "Shein Acquires Everlane", 18 May 2026, cnn.com.

²⁰ TikTok Newsroom, "TikTok Shop Expands Across Europe", 28 May 2026, newsroom.tiktok.com.

²¹ Politico EU, "European Commission Fines Temu 200 Million Euro Under DSA", 28 May 2026, politico.eu.

²² EU Council, "Customs Council Agrees to Levy Customs Duty on Small Parcels as of 1 July 2026", 12 December 2025, consilium.europa.eu.

²³ FTW PI, "Zalando Scales 3D Twins to 45,000 SKUs with NVIDIA Cosmos and ALLSIDES", 21 May 2026, ftw.pi.tv.

²⁴ Retail Optimiser, "Otto Group Offers AI Content Production Platform to Other Retailers", May 2026, retail-optimiser.de.

²⁵ TechCrunch, "Amazon Will Show AI Product Images When You Search", 3 June 2026, techcrunch.com.

²⁶ Adobe Blog, "Firefly AI Assistant Public Beta", 27 April 2026, blog.adobe.com.

²⁷ Google Merchant Center, "Product Studio Adds Image-to-Video via Veo, Imagen 4 GA", 27 May 2026, support.google.com.

²⁸ CyberNews, "EuroCommerce Lobbies for AI Act Article 50 Exemption", 19 June 2026, cybernews.com.

²⁹ PetaPixel, "Model Sues Fashion Brand After AI-Generated Pictures of Her", 1 June 2026, petapixel.com.

³⁰ RTE News, "Shein Opens E-Commerce Logistics Facility in Dublin", 8 June 2026, rte.ie.

³¹ PR Newswire, "Shein Xcelerator Debuts in the Middle East", 16 June 2026, prnewswire.co.uk.

³² Jing Daily, "Sheglam, 400 Million Dollar C-Beauty Brand That Barely Exists in China", 28 May 2026, jingdaily.com.

³³ PR Newswire, "Shopware Unveils Four Connected Commerce Infrastructure Products at Community Day 2026", 11 June 2026, prnewswire.com.

³⁴ BusinessWire, "FIDO Alliance to Develop Standards for Trusted AI Agent Interactions", 28 April 2026, businesswire.com.

³⁵ Google Cloud Press Corner, "A New Era for Online Shopping, How Macy's Built the Ask Macy's AI Agent in 4 Weeks With Gemini Enterprise for Customer Experience", 22 April 2026, googlecloudpresscorner.com.

³⁶ PR Newswire, "Ulta Beauty and Google Introduce Gemini-Enabled Shopping Experiences", 22 April 2026, prnewswire.com.

³⁷ PR Newswire, "Cognizant and Google Cloud Bring Agentic AI to Retail's Most Critical Customer Moments with Gemini Enterprise", 22 April 2026, prnewswire.com.

³⁸ SAP News Center, "Next-Gen SAP Ariba Is Here", 12 March 2026, news.sap.com.

³⁹ Coupa Newsroom, "Coupa's Newest Release Expands Agentic AI Collaboration and Orchestration Capabilities", coupa.com.

⁴⁰ Rebellion Research, "Sequentum Launches Enterprise AI Checkout Agent With Skyfire Integration", 4 May 2026, rebellionresearch.com.

⁴¹ GlobeNewswire, "Feedonomics Unlocks Agentic Discovery with Agentic Catalog Exports", 27 April 2026, globenewswire.com.

⁴² Payment Expert, "FCA 2026 Payments Regulatory Priorities Report", 25 March 2026, paymentexpert.com.