W3C and GS1 Put Product Data at the Center of Agentic Commerce

The most useful agentic-commerce conversation this week was not a product launch. It was a standards workshop.

On September 8-9, 2026, W3C and GS1 held a hybrid workshop in Zurich titled "E-commerce for Humans and AI Agents."¹ The event brought together web standards people, retailers, brands, payment and identity specialists, and AI platform providers to discuss a practical question: if AI agents are going to search, compare, choose, and sometimes buy on behalf of people, what does the open web need to expose so those actions remain accurate and trustworthy?

The workshop overview frames the shift clearly. Web content has historically been designed for humans, even when pages also expose structured data through APIs or embedded JSON-LD using vocabularies such as schema.org and GS1 Web Vocabulary.² LLMs and AI agents are becoming an intermediary between that content and end users: they summarize results, follow links, and may support online activity before a person visits a page directly.²

That matters because agentic commerce is not only a checkout problem. A payment flow can be technically correct while the agent still buys the wrong item. The harder problem is product truth: which product is this, which variant is active, who is authorized to describe it, and can an agent verify that before taking action?

The agenda shows that the workshop treated this as an infrastructure problem rather than a branding problem. Google presented the Universal Commerce Protocol, described in the agenda as an open source protocol intended to standardize commerce interactions across businesses, platforms, and payment providers.¹ Shopify discussed UCP and WebMCP in production contexts across millions of merchants, including protocol layering, interoperability, delegation, and attestation gaps.¹ OpenAI presented on progressive enhancement for agentic commerce journeys, including where standards can help and where custom extensions or fallbacks may be needed.¹

But the most concrete thread was product identity. Mars and K10X presented work on brand-authoritative product data repositories using GS1 Digital Link and GS1 Web Vocabulary, including the problem of distinguishing consumer pack variants such as reformulations, seasonal artwork, or promotional packaging under a single GTIN.¹ OriginTrail presented verifiable product knowledge and agent identity using open W3C and GS1 standards.¹ Digital Link's Paula Rivero focused on what the company called the "last meter" problem: an agent can search, compare, check out, and pay, yet still fail if it cannot identify the exact physical product the user meant.¹

Digital Link's own September 9 write-up makes the same point from a vendor perspective: agents can already buy, but the unresolved question is whether they know what they buy.³ The company describes GS1 Digital Link as a way to combine established product identifiers such as GTINs with URLs, creating online product identities that can be resolved by people, apps, and agents.³

The second major thread was authorization. The agenda included sessions on trusted agentic commerce, payment credentials, runtime behavioral assurance, verified agent identity, delegation, accountability, and the "last reversible moment" before an agent's proposed action creates an external effect.¹ FIDO's Payments Technical Working Group presented emerging work around agent-mediated commerce, user authorization, delegation, payment initiation, and verifiable evidence of intent.¹

That is the right level of boring. Agentic commerce only becomes durable when merchants can distinguish a human-authorized agent from a bot, when brands can publish product data that agents can verify, and when users can delegate narrowly enough that "buy batteries for my camera" does not become an open-ended spending permission.

For developers, the takeaway is simple: agent-ready commerce is going to depend on ordinary web infrastructure becoming more explicit. Structured data, canonical product URLs, verifiable credentials, scoped authorization, audit trails, and protocol discovery are not side quests. They are the substrate an agent needs before it can act usefully on the open web.

The Zurich workshop did not settle the standards landscape. It did something more useful: it exposed the shape of the work. The next phase of e-commerce will not be defined only by who adds a buy button to a chatbot. It will be defined by whether the web can make products, permissions, and intent legible enough for agents without locking merchants and users inside a single platform.

Sources