Diop Daily #128 — October 2026

A Product Must Carry Its Own Passport

An agent can search, compare, check out, and pay while still buying the wrong physical item. The failure can occur after every interface has behaved correctly: the offer was discoverable, the payment was authorized, and the receipt was generated. The missing layer is product truth, a durable identity that ties an object to its exact variant, claims, origin, authority, freshness, and permitted uses.

This is the next boundary in agentic commerce. The machine needs more than a catalog entry and more than a merchant endpoint. It needs a product that can answer for itself when the agent asks: Which thing is this? Who speaks for its data? Which version is on the shelf? Which claim is current? What evidence supports the answer? A product passport is the compact name for that bundle of identity and evidence.

Payment can authorize a purchase; only product truth can tell us what was purchased.

The last meter is where identity becomes economic

Much of the public discussion about agentic commerce concentrates on the visible journey: discovery, comparison, checkout, payment, and order management. Google’s Universal Commerce Protocol describes this journey as a standardized set of commerce capabilities that can connect consumer surfaces, business backends, and payment providers. The protocol’s architecture addresses a genuine integration problem. A business needs a common way to expose what it can sell and how an agent can transact.

Yet a common transaction language does not automatically create a common object of reference. A product title can remain ambiguous across languages, package sizes, formulas, seasonal designs, geographic versions, or batches. An agent may complete the correct protocol exchange around the wrong interpretation. That is why the W3C/GS1 workshop held in September placed product identity, structured data, provenance, agent identity, and delegated authority beside the commerce protocol itself.

One workshop session described the exact physical item as the last-mile problem: an agent may search and pay successfully while selecting the wrong product. Another described consumer-pack-variant disambiguation, where a single trade-item identifier can cover reformulations, promotional packaging, or seasonal artwork that still require distinct machine-readable treatment. These are workshop contributions and implementation perspectives, not evidence that the market has solved the problem. They do show the standards community naming the boundary that a payment success screen hides.

The economic consequence is direct. If the agent cannot distinguish the item, the merchant bears the cost of returns, disputes, unsafe recommendations, inventory errors, and lost confidence. If the product carries a stable identity that resolves to current, authoritative information, the same transaction can support a chain of later actions: replenishment, warranty, safety instruction, recall, repair, resale, or provenance review.

Identity is separate from information about the thing

GS1’s Conformant Resolver standard makes a useful distinction. A GS1 Digital Link URI identifies the entity itself, while the information about that entity may change over time. The identifier can remain persistent as product pages, instructions, certifications, or other linked resources are updated. A resolver can expose several machine-discoverable and machine-actionable links without forcing every related asset into one database.

This separation is stronger than a static product page. It treats identity as a durable address and information as a governed, changing projection. The same object can therefore have different views for a consumer, a retailer, a customs process, a service technician, or an authorized agent, while the underlying identity remains stable enough to connect those views.

That pattern has particular value in multilingual and fragmented markets. A product’s name, instructions, and commercial description may travel through several languages and distribution channels. The identity should travel with it, while each authorized information source can provide the language and level of detail required by its audience. This is an architectural inference, not a claim about any specific market. It points toward a practical form of sovereignty: local institutions can publish the meanings and evidence attached to an identifier without surrendering the identifier to every platform that displays the product.

A product passport should therefore contain, or point to, at least these fields:

  • Canonical identity: the identifier for the product, location, asset, batch, lot, or serial instance being discussed.
  • Variant and time: the size, formulation, language, market, expiry, revision, or packaging state that limits the claim.
  • Authority: the organization or person authorized to publish, update, revoke, or correct each important assertion.
  • Evidence and provenance: the source, transformation, certification, measurement, or signature that supports the assertion.
  • Resolution policy: the links, permissions, freshness rules, and refusal states that determine what an agent may learn or do next.

The list is deliberately operational. “Organic,” “safe,” “available,” “compatible,” and “the customer’s requested item” are different kinds of statements. They need different authorities, evidence, time windows, and correction paths. A machine-readable product surface becomes trustworthy when it preserves those distinctions instead of flattening them into a persuasive description.

Protocols move messages; passports preserve referents

Recent protocol work can make commerce more interoperable while leaving the referent unstable. UCP describes discovery and capability schemas across consumer surfaces, businesses, and payment providers. W3C’s workshop agenda places this beside questions about product truth, trusted identity, agent reputation, and authority to speak for a product identifier. NIST’s AI Agent Standards Initiative adds the identity and authentication vocabulary required for secure human-agent and multi-agent interaction.

These layers should be kept distinct. The commerce protocol governs how a request travels. The agent identity layer helps establish who is acting and under whose authority. The product passport governs what object the request concerns and which claims may be relied upon. A payment credential answers whether a transaction may proceed. It does not answer whether the agent and merchant agreed on the same physical item.

Collapsing these layers creates predictable confusion. A merchant may prove that it authorized a checkout while the buyer disputes the variant. A platform may sign a product description while lacking authority over the brand’s underlying identifier. An agent may present a trusted identity while carrying stale product data. A system that preserves the transaction but loses the referent leaves the hardest question outside the record.

For African commerce and cultural production, the implication is larger than retail. Books, medicines, agricultural inputs, craft goods, educational materials, and creative licenses all require machine-readable distinctions that survive distribution. A local publisher should be able to state which edition an agent is recommending. A cooperative should be able to distinguish a product lot from a generic category. A rights holder should be able to identify the exact asset and the conditions under which it may be quoted, sold, or transformed. The infrastructure should support local terminology and local authority without making every institution rebuild a separate integration for every agent.

Where the investable surface is widening

The commercial object is the product-truth layer: infrastructure that keeps identity, variant, evidence, authority, freshness, and machine-actionable resolution attached to the thing being offered.

  • Identity registries: systems that issue, resolve, and maintain canonical identifiers across products, variants, lots, editions, and service assets.
  • Authority registries: records that distinguish who may speak for an identifier, who may update a claim, and how a disputed assertion is corrected or revoked.
  • Variant resolution: services that prevent a general identifier from hiding a formulation, language, market, batch, or packaging change that matters to the decision.
  • Evidence services: provenance-aware stores that bind claims to source documents, measurements, certifications, signatures, and validity windows.
  • Agent-readable distribution: interfaces that expose product truth through APIs, UCP, A2A, MCP, Web data, and human-readable pages without creating contradictory copies.

Underwriting can begin with measurable questions. How often does a transaction resolve to the wrong variant? Can an agent retrieve the same product identity across two languages and two channels? How quickly does a revoked claim disappear from authorized surfaces? Can the merchant prove which evidence was current when the recommendation was made? Can a buyer export its identifiers, claims, provenance, and correction history when it changes platform?

The stronger companies in this layer will sell fewer promises and more durable references. Their customers will include brands, retailers, logistics networks, payment providers, publishers, manufacturers, regulators, and agents that need a shared answer to “what is this?” Revenue may arrive through verification, resolution, compliance, data infrastructure, or transaction support. The defensible asset is the network of maintained identities and authoritative relationships, not a prettier catalog.

Build the passport before automating the purchase

A serious institution can begin with one product family and one high-cost ambiguity. Choose the item that creates the most returns, disputes, unsafe confusion, or manual clarification. Assign it a canonical identity. Separate its variants. Name the authority for each claim. Attach evidence and validity dates. Publish a resolver that returns both human-readable and machine-actionable information. Then test an agent in English and French, with a refusal path for uncertainty.

The test should ask more than whether the agent completes checkout. It should ask whether the agent can explain which exact item it selected, which claims it relied on, who issued those claims, and what would cause the answer to change. A successful purchase with an untraceable referent is operationally incomplete.

Standards will help agents cross boundaries, but standards alone cannot decide who owns a product’s truth. That decision belongs to institutions that maintain the identifier, the evidence, the language, and the right of correction. The machine may carry the request across the market. The product must carry the passport that tells the market what it is.

Sources