Walmart has published terms that allow external AI shopping agents to interact with its catalogue, subject to explicit guardrails (Walmart agent terms). The move formalizes something that was already happening informally, because agents were scraping and transacting regardless of whether a retailer permitted it. What changed is that the retailer now defines the rules of access, and those rules will shape how product data has to be structured across the industry.
What the terms actually govern
The terms govern access, rate, representation and liability. Access defines what an agent may read, rate defines how often, representation defines whether the agent may act on a customer behalf, and liability defines who pays when the agent gets it wrong. Each of these is a design choice, and each choice determines whether agents become a distribution channel or a cost centre.
The guardrails matter more than the permission. An open catalogue with no rate limit invites agents to hammer endpoints, which degrades performance for human shoppers. A closed catalogue with strict limits pushes agents toward third-party data, which reintroduces the accuracy problem the retailer was trying to control. The interesting engineering sits between those two failure modes.
Why product data becomes the constraint
Readability before richness
Agents select products from structured attributes, not from marketing copy. A listing with rich prose but missing dimensions, materials or compatibility fields is effectively invisible to an agent, no matter how well it ranks for human search. Readability now precedes richness in the priority order, which reverses a decade of content practice.
This is a measurable gap. Research shows AI improves discovery while reviews still determine conversion (AI discovery versus reviews), which means agents can widen the consideration set but cannot close the sale on their own. Retailers that structure attributes well will be included in that wider set, and retailers that do not will be filtered out before any human sees the listing.
The fulfilment promise behind the agent
Agents recommend what they believe will arrive on time. Walmart has opened a next generation fulfilment facility in California to strengthen west coast coverage (Walmart California facility), and Amazon plans roughly 300,000 seasonal hires for the holiday period (Amazon holiday hiring). Both moves are about delivery certainty, which is the input an agent uses to decide whether to surface a product at all.
Fulfilment as a ranking input
This links catalogue work directly to logistics investment. An agent that recommends an item which then arrives late has damaged its own accuracy score, so agents will systematically favour retailers with tighter delivery windows. Fulfilment performance is therefore becoming a discoverability variable rather than a purely operational one.
How marketplaces fragment the same shelf
Brands are simultaneously managing an expanding set of marketplaces this holiday season (marketplace expansion), which multiplies the number of catalogues that must stay consistent. Each marketplace has its own attribute schema, and an agent querying three of them will surface whichever one has the cleanest data. Consistency across channels is now a competitive input, not a hygiene factor.
The practical consequence is that product information management has moved from a back-office function to a growth function. Teams that treat attribute completeness as a data quality chore will lose visibility in exactly the channel that is growing fastest, and they will lose it without any error appearing in their own analytics.
What to build first
A three-layer readiness check
The first layer is attribute completeness against the schemas agents actually read, measured as a percentage rather than a checklist. The second is delivery promise accuracy, measured as the gap between quoted and actual arrival dates. The third is machine-readable policy, meaning returns and warranty terms expressed as structured fields rather than paragraphs of prose.
These three layers can be audited in a quarter and they compound. Complete attributes improve inclusion, accurate promises improve ranking, and structured policies reduce post-purchase friction that would otherwise appear as negative reviews. None of them require new technology, which is why the gap between leaders and laggards will widen quickly.
The broader point is that agentic commerce rewards specification over persuasion. When the buyer is a machine, the arguments that work are dimensions, stock status and delivery windows, and the retailers that write those down cleanly will be the ones the agents keep recommending.
