An AI agent can navigate the entire digital funnel – searching, comparing, checking out, and executing payments – yet still fail at the point of purchase. This is the “last meter” problem, a structural friction point where the digital intent of an agent meets the physical reality of a product. If an agent cannot definitively identify the exact physical item a user intends to buy, the entire automated commerce stack collapses into a liability or a return-processing nightmare.
The W3C and GS1 workshop, “E-Commerce for Humans and AI Agents,” held in Zurich on September 8–9, 2026, and co-hosted by Google, centered on this specific failure mode. The agenda revealed that while the industry is racing to build agentic commerce protocols, the foundational data layer remains fragmented. The core challenge is not the transaction itself, but the authoritative identification of the goods being transacted.
The industry is coalescing around the Global Trade Item Number (GTIN) as the primary key for physical goods. Both Google’s Universal Commerce Protocol (UCP) and OpenAI’s Agentic Commerce Protocol (ACP) rely on this standard. Shopify, which has deployed UCP and WebMCP across millions of merchants, highlighted the necessity of protocol layering and the current gaps in attestation. Without a universal identifier, agents are left guessing between reformulations, seasonal artwork, or promotional packaging.
GS1 Digital Link is emerging as the primary mechanism to bridge this gap. By providing a canonical, brand-controlled web URL via a 2D barcode, it allows agents to resolve a physical item to trusted, brand-authoritative data. A pilot program by Mars and K10X demonstrated this in practice, using GS1 Digital Link and the GS1 Web Vocabulary to solve Consumer Pack Variant disambiguation. This ensures that the data an agent consumes is not just a scraped web snippet, but a verified source of truth.
However, a significant vocabulary interoperability gap persists. Paola Di Maio of the W3C AI KR CG noted the friction between existing standards like schema.org, GoodRelations, and the GS1 Web Vocabulary. For agent commerce to scale, the industry requires a vocabulary-neutral interoperability layer. Without it, product data remains siloed, forcing developers to build custom adapters for every ecosystem.
Trust infrastructure is also maturing alongside these data standards. Sessions on agent identity layers, including presentations from Skyfire on KYAPay and Vouched.id on KYA-OS, suggest that the market is moving toward signed-JWT protocols. These allow merchant security infrastructure to distinguish between human-authorized agents and malicious bots, effectively creating a verifiable identity layer for automated actors.
The broader commerce stack is also evolving to support this. Huawei’s TASP protocol introduces levels of payment autonomy, while Ericsson and Samsung are exploring how network capabilities and spatial commerce can feed into the agentic decision-making process. Yet, as Claudia Caluori of Studio Polimeni noted, the legal reality remains: liability for an automated actor rests with the business that deploys it. This creates a strong financial incentive for brands to control their product data.
The workshop concluded without a formal outcomes report, leaving the industry to grapple with the next steps. The structural question remains: how quickly can the ecosystem move from proprietary, fragmented data models to a unified, GS1-anchored standard that agents can reliably consume? Until that gap is closed, the “last meter” will continue to be the most expensive point of failure in the agentic commerce journey.