Agentic Commerce
Last updated
Agentic commerce describes AI agents completing purchases on behalf of users: researching options, selecting a product, and processing payment, with the user authorising rather than manually executing each step. The commercial and technical infrastructure to support this is being built across the industry. The gap between that infrastructure and actual agent-completed transaction volume is currently significant. This article distinguishes what is confirmed, what is operational but early-stage, and what remains speculative.
What is the Universal Commerce Protocol?
The Universal Commerce Protocol (UCP) is an open standard for agentic commerce, developed by Google in collaboration with Shopify, Etsy, Wayfair, Target, and Walmart, and endorsed by more than 20 partners including Adyen, American Express, Best Buy, Flipkart, Macy’s, Mastercard, Stripe, The Home Depot, Visa, and Zalando.1 Announced at the National Retail Federation conference in January 2026, it defines a common language for commerce interactions between AI agents, merchants, and payment providers: product discovery, capability negotiation, checkout, and post-purchase handoff.21 A UCP-enabled site advertises itself at /.well-known/ucp, which is the quickest way to check whether your platform has implemented it.
UCP is protocol-agnostic at the implementation level: it supports REST, the Model Context Protocol (MCP), Agent Payments Protocol (AP2), and the Agent2Agent (A2A) protocol. Agents built on different platforms can interact with merchants through UCP without requiring bespoke integrations. AP2 is Google’s authorisation layer for agentic payments: it lets users set spending limits and product constraints, with tamper-proof digital mandates creating a verifiable record between user, merchant, and payment processor.
In March 2026 Google added three capabilities to the protocol. Cart lets agents “save or add multiple items to a shopping cart at once from a single store”. Catalog lets agents “retrieve select real-time product details from a retailer’s catalog where necessary, like variants, inventory and pricing”. Identity Linking lets shoppers “receive the same loyalty or member benefits they would on a retailer’s site”. Google also announced simplified UCP onboarding through Merchant Center.3 Cart matters most of the three, because single-item checkout was the specific complaint that sank the first generation of in-chat buying.
At Google I/O 2026, Google introduced Universal Cart as the consumer-facing implementation: an intelligent shopping cart that works across merchants and runs background operations once items are added: monitoring price drops, surfacing price history, flagging product incompatibilities across multiple retailers, and surfacing merchant offers and loyalty perks via Google Wallet.4 Checkout completes via Google Pay or transfers to the merchant’s site, with the brand remaining the merchant of record. Universal Cart is rolling out across Search and the Gemini app in the US over summer 2026, with YouTube and Gmail to follow; launch merchants include Nike, Sephora, Target, Ulta Beauty, Walmart and Wayfair, alongside Shopify merchants such as Fenty and Steve Madden. At Google Marketing Live the following day, Google set out the same cart running across Search including AI Mode and Gemini, with UCP also wired into Direct Offers, Demand Gen campaigns, and Shopping ads on YouTube.56 UCP checkout is expanding to Canada and Australia, with the UK to follow, and UCP is extending to additional verticals including hotel booking and local food delivery.
How does Stripe support agentic commerce?
Stripe has built a payment infrastructure layer for agentic commerce through three components:
The Agentic Commerce Protocol (ACP), co-developed with OpenAI and released as an open standard in September 2025, was the earliest and most consumer-visible of these standards.7 It is the protocol behind ChatGPT’s Instant Checkout, and it predates UCP by three and a half months.
The Agentic Commerce Suite allows merchants to upload a product catalogue and manage agent access from the Stripe Dashboard. The suite handles discovery, checkout, and payment for agent-initiated transactions. For merchants already using Stripe for payment processing, this layers onto existing infrastructure without requiring a new payment integration.8
The Machine Payments Protocol (MPP), co-authored by Stripe and Tempo, allows AI agents to complete transactions programmatically: microtransactions, recurring payments, card-based payments, and stablecoins via Shared Payment Tokens. It operates at the payment execution layer, distinct from MCP (which handles agent-to-tool communication more broadly), and Stripe’s announcement states support for both MPP and the x402 payment standard.9 ACP, UCP and x402 are designed to compose rather than compete, so a merchant is not choosing between them.
How does Perplexity’s Instant Buy work?
Perplexity has run in-chat checkout with PayPal since November 2025. Perplexity describes Instant Buy as letting “U.S.-based users search for and purchase products from a wide range of merchants directly on Perplexity with a seamless checkout workflow”, with “all orders fulfilled directly by the merchant and payment sent securely to the merchant”, so the retailer stays merchant of record.10 For merchants on neither Shopify nor Stripe, PayPal’s store sync and agent ready tools are the low-effort route into it.
What are Shopify’s agentic channels?
Shopify’s Agentic Storefronts connect Shopify merchant catalogues into ChatGPT, Google AI Mode and Gemini, Microsoft Copilot, and Meta.11 For Shopify merchants, this is a sales channel configuration in the Shopify admin, not a custom technical integration. Google AI Mode and Gemini is in early access and not yet available to every store.
What is the difference between agentic commerce and AI shopping recommendations?
The distinction matters for implementation priority.
AI shopping recommendations (covered in AI shopping agents and ecommerce visibility) describe AI systems surfacing product suggestions within conversational interfaces. The user still navigates to the merchant’s site and completes checkout manually. Optimising for this requires catalogue quality, structured data, and Merchant Center completeness: no new infrastructure.
Agentic commerce describes AI agents completing the entire purchase: selecting, adding to cart, and paying, with the user confirming or delegating the transaction. This requires the merchant’s systems to accept programmatic purchase requests from agents, not just browsing traffic. It is a fundamentally different technical relationship.
A merchant can be well-positioned for AI product discovery without any agentic commerce infrastructure. Conflating the two leads to either under-investment (dismissing AI shopping recommendations as “too speculative”) or over-investment (building agentic checkout infrastructure before demand exists).
What should merchants implement now?
The current cost-benefit ratio favours low-effort, infrastructure-adjacent steps:
Ensure Merchant Center data is accurate and complete. Required for Google AI Mode and Shopping Graph inclusion regardless of agentic developments.
Check your Agentic Storefronts channels if you are a Shopify merchant. Agentic storefronts is active by default for eligible stores, so the step is not enrolment: confirm which channels the store is participating in, accept the Agentic Storefronts supplemental terms, and note that Google AI Mode and Gemini remains in early access.11
Review Stripe’s Agentic Commerce Suite if you use Stripe. Product catalogue upload is a low-effort step toward agentic discovery infrastructure. Stripe’s announcement directs interested merchants to a waitlist rather than to general availability, so check the current status before planning around it.8
If you are on neither Shopify nor Stripe, look at Shopify’s Agentic plan. It syndicates a product catalogue to “ChatGPT, Google AI Mode and Gemini, Copilot, Perplexity, the Shop app” with no Shopify online store required, no monthly fee, and card rates from 2% + 25p.12 It is a catalogue service rather than a commerce migration.
Monitor AI-attributed orders in GA4. The ‘AI Assistant’ channel group and custom Perplexity segments establish a measurement baseline before the channel becomes more significant.
Can merchants block shopping agents?
The legal position shifted in August 2026. Amazon sued Perplexity in November 2025 over the shopping agent in its Comet browser, under the Computer Fraud and Abuse Act and California’s equivalent, and won a preliminary injunction in March 2026. On 4 August 2026 the Ninth Circuit reversed it, holding that the statutes regulate human action and that “however advanced the Assistant currently is, it is a tool, not a person”. The court’s reasoning was that “it is the user who accesses Amazon’s computers, with the help of the Assistant to carry out specific acts on Amazon.com”.1314
This is the first federal appeals ruling on whether an agent shopping on a user’s behalf breaches computer-access law, and the underlying suit continues. It cuts against blocking as a strategy, and merchants have taken opposite views: Amazon blocks agents, Walmart does not and says it does not plan to.15
What is still speculative?
Custom MCP endpoint development, bespoke agent APIs, and custom agentic checkout integrations require engineering investment that is not supported by published transaction volume data for most merchant categories.
The clearest evidence for that caution came from the company furthest ahead. OpenAI launched Instant Checkout in September 2025, then pulled back in March 2026, rerouting purchases toward third-party apps and the merchant’s own site rather than completing them in the chat.16 The reasons are more instructive than the reversal itself.
Merchant adoption was thin, though the exact figure is contested: the number reported at the time was about a dozen Shopify merchants live with the feature, while Shopify told Forrester it was closer to 30 and climbing.17 Against a merchant base in the millions, either figure is negligible. That thinness was not merchant reluctance, though. Shopify president Harley Finkelstein said the holdup was on the AI firms’ side.18
UCP adoption is at a similar stage. On Shopify’s Q2 2026 earnings call in August, Finkelstein put it at “dozens of retailers and platforms” since the protocol launched in January.19
The demand signal is more revealing. From November 2025 Walmart made around 200,000 products buyable inside ChatGPT. Daniel Danker, who oversees product and design at Walmart, told WIRED that conversion ran three times lower for that in-chat selection than for the items ChatGPT showed which required clicking out to Walmart’s site.15 Both sets were surfaced inside ChatGPT, so this measures two checkout routes against each other rather than ChatGPT against walmart.com as traffic sources. The two sets are not like for like either: Walmart deliberately withheld products where buying a single item in isolation is a poor experience, such as a television the buyer will want to bundle with cables.
Danker’s own diagnosis is narrower than a verdict on agentic checkout. The problem he names is basket fragmentation. Instant Checkout buys one item at a time, and shoppers “generally don’t want to split the checkout experience”, fearing five separate boxes when they wanted one delivery. That is a flaw in the first implementation rather than evidence that people will not buy through an assistant. The counterweight points the same way: Danker says ChatGPT brings Walmart roughly twice the rate of new customers as search engines do. Discovery through the assistant was working; the checkout inside it was not. Both figures are executive commentary rather than audited disclosure, and Danker did not present the underlying data.
That changes what the reversal means. Walmart’s replacement is not a retreat from AI shopping but a change of venue: its own chatbot, Sparky, runs inside ChatGPT and arrived in Google’s Gemini a month later, so the basket persists and checkout stays on Walmart’s rails. This is OpenAI’s pivot “to apps” in concrete form, and it is the shape the near-term opportunity takes. So the infrastructure is real, but “operational” needs qualifying: the specifications, payment rails and platform channels genuinely exist and work, while the consumer behaviour that would justify bespoke merchant investment does not yet.
That points to a specific order of work: maintain catalogue and data quality, which serves AI recommendations, Shopping and agentic discovery alike; take the low-effort channel steps; monitor the channel as volume data emerges; and make infrastructure investment when transaction data justifies it for your specific store. Merchants who built for the first wave built against a surface that was then withdrawn; the Agentic Commerce Protocol itself is intact, and OpenAI continues to work with Stripe on it for app-based transactions.16
Footnotes
-
Under the Hood: Universal Commerce Protocol — Google Developers Blog ↩ ↩2
-
Building the Universal Commerce Protocol — Shopify Engineering ↩
-
New Universal Commerce Protocol capabilities — Google Blog ↩
-
Google Shopping introduces Universal Cart, agentic shopping — Google Blog ↩
-
Google expands Universal Commerce Protocol — Search Engine Land ↩
-
How we’re helping retailers thrive with new Universal Commerce Protocol features and AI tools on Google — Google Blog ↩
-
Stripe and OpenAI launch Instant Checkout and the Agentic Commerce Protocol — Stripe ↩
-
Ninth Circuit: AI agent’s access to website is not a CFAA violation — Metropolitan News-Enterprise ↩
-
Perplexity has successfully overturned Amazon’s injunction on its AI shopping bot — Engadget ↩
-
Why Walmart and OpenAI are shaking up their agentic shopping deal — WIRED ↩ ↩2
-
OpenAI’s big ChatGPT Instant Checkout plan just changed — Search Engine Land ↩ ↩2
-
What it means that the leader in agentic commerce just pulled back — Forrester ↩
-
Why OpenAI’s checkout retreat spells trouble for its commerce strategy — Forbes ↩
-
Shopify (SHOP) Q2 2026 earnings call transcript — The Motley Fool ↩