Research, guides and ecosystem updates
Original research on UCP adoption, the State of Agentic Commerce report, guides for each store platform, and the latest spec changes, written for developers and platform teams building for agentic commerce.
Protocol Design and the Validation Layer: Two Halves of Getting UCP Right
Tobi Lütke and Ilya Grigorik's UCP conversation is really about the craft of protocol design — negotiate don't blueprint, a boring core, discovery without a central registry, fail open. Those are exactly the principles that create the need for an independent validation layer. Here they are, paired with what we see across 6,391 verified stores and 1.5k+ agent sessions.
· 6 min read
Agentic Commerce's Messy Middle: The Payments and Trust Problem Between Browse and Buy
Discovery and checkout are the easy parts of agentic commerce. The hard part is the "messy middle" — payments, trust, verification, liability — and a major European paytech just said so out loud. Decoding what Nexi's Raja Saggi is actually saying, why Europe is a year behind, and where our own data shows the middle breaks.
· 7 min read
Google Opened UCP for Lodging. We Already Tested What It Needs to Handle.
Google opened the UCP for Lodging waitlist — but the spec is still "coming soon." So here is the prescriptive read: what the lodging spec will have to handle for agentic hotel booking to work, grounded in our own tests of five frontier models against expiring travel inventory (none checked an offer's clock), plus the exact checklist hotels should run before they integrate.
· 7 min read
Agentic AI in Travel: Why UCP Isn't Travel-Ready Yet — and What We Measured
Amadeus says agentic travel needs protocols, not chatbots — and that UCP is still "retail-native." It is right, and we have the data: we ran five frontier models at expiring travel inventory and none respected an offer's TTL; only one survived a three-second expiry. Travel's supply side just standardized on UCP at GML 2026 — here is why the demand side is not ready, and what the protocol still needs.
· 7 min read
Google Universal Cart & UCP: What Marketing Live 2026 Means for Merchants
UCP shipped in January; four months on, the supply side has hit critical mass — 6,540 monitored domains, ~99% of verified stores already running the cart, checkout and Google Pay rails. Google Marketing Live 2026 is the moment the demand side switches on: Universal Cart, AI Mode checkout, YouTube, BNPL. The data read on the inflection — and the six checks every merchant and demand-side player should run now.
· 9 min read
Shopify Just Shipped a UCP CLI. It Buys Anywhere — But Only Finds Shopify.
On May 18, Shopify published @shopify/ucp-cli — an agent shopping skill that checks out against any UCP merchant on any platform. But its product discovery is hardcoded to catalog.shopify.com: it can buy anywhere, but it can only find Shopify. The architectural split, and what it means for everyone not on Shopify.
· 7 min read
UCP Variant Data: The #1 Reason Agent Checkouts Fail
Variant mismatches cause more agent checkout failures than any other issue we observe across 4,500+ monitored UCP stores. How to structure your variant data so agents reliably pick the right size, colour, and configuration — with concrete examples of what breaks and what doesn't.
· 12 min read
How to Test Your UCP Implementation with AI Agents
Schema validators tell you if your markup is correct. They tell you nothing about whether an agent can buy from your store. Here's how to test UCP with real frontier agents across multiple models — and why static validation alone leaves you blind.
· 7 min read
The UCP Technical Council Just Shipped Attribution into Core. Here's What That Means.
Yesterday the UCP Technical Council merged a top-level `attribution` field into UCP main — the first piece of explicit advertising and measurement infrastructure to land in the spec's core schema. Cross-vendor design between TC members at Shopify, Google, and Meta. The architectural decision behind it, who shaped it, and why the governance shape matters more than the field itself.
· 8 min read
UCP Playground at 1,000+ Agent Sessions: What 16 Models and 97 Real Stores Reveal About AI Shopping
Two and a half months after we shipped UCP Playground with 114 agent sessions, the dataset has crossed the 1,000-session mark — across 16 frontier models, 97 real UCP-enabled stores, and roughly $96K of agent-driven cart value. Here are the patterns the larger sample is now strong enough to anchor.
· 16 min read
AI Commerce Needs MLPerf — and Here's an Early Attempt
Agentic commerce has consolidated around UCP, but there's no shared way to evaluate how well any of these implementations actually work for real frontier agents. Today we're publishing UCP Playground Evals — standardised sequences, cross-store comparability, multi-model coverage, and replayable sessions for every result.
· 12 min read
Introducing the UCP Score: A 0–100 Agent-Readiness Grade for Every UCP Store
The most common follow-up to a UCP status check has always been 'OK, but is it any good?' — so today we're shipping the answer: a 0–100 score with category breakdowns, copy-paste fixes, and platform-aware percentiles for any domain.
· 9 min read