Agentic Commerce Runs on Machine-Readable Product Data — and Who Maintains It Is the Open Question

For most of the last two years, the agentic-commerce conversation was about the agents: which model, which protocol, whose assistant wins the buyer. That framing is quietly being abandoned by the companies actually building the infrastructure, because they’ve hit the same wall. The agents work. The bottleneck is underneath them — in the product data the agents are supposed to act on. And on that point, an unusually wide set of players now agrees.

The numbers explain the urgency. McKinsey has estimated that agentic commerce could orchestrate $3 trillion to $5 trillion globally by 2030, while Bain estimates the U.S. agentic commerce market could reach $300 billion to $500 billion by 2030. Accenture expects more than 30% of online commerce — close to $3.1 trillion in transactions — to run through AI agents by then. That’s a large enough prize that the infrastructure question stops being academic. If a meaningful share of commerce is mediated by agents, then the data those agents read has to be machine-readable and trustworthy by default, not on a good day.

The market is converging on the same diagnosis

What’s notable is who’s saying it. When Adyen — a payments platform processing trillions a year — launched Adyen Agentic in June 2026, it built the first layer around product data, not payments: a structured product and inventory layer that distributes real-time catalog, pricing, and availability data across conversational commerce environments. Adyen’s product-feed guide makes the operational point explicit: good product data has to stay up to date, move across AI platforms, and adapt as requirements change.

Adobe’s retail research lands in the same place from the merchant side. In Q1 2026, Adobe reported AI traffic to U.S. retail sites grew 393% year over year, while warning that many product pages are not machine-readable enough for AI systems to interpret consistently. Independent commentary has started naming machine-readable product data infrastructure outright as the whitespace worth investing in. Different vantage points — a payments company, retail analytics, and market analysts — are converging on one diagnosis: the durable layer in agentic commerce is the trusted product-data layer, and most of it doesn’t exist yet.

That’s a meaningful shift. The hard problem was assumed to be the model. It turns out the hard problem is the data structure the model has to act on — obsolete catalogs, inconsistent attributes, duplicates, and silent gaps that a human used to paper over and an agent can’t. If you want the practical version of that problem, start with our guide to AI-ready product data and the broader checklist for product data AI search visibility.

”Machine-readable” is the easy half of the requirement

Here’s where most of the discussion stops short. It’s straightforward to make a catalog machine-readable — expose it as structured fields, ship a tidy feed, add some schema. That clears the syntactic bar. But an agent acting on a buyer’s behalf doesn’t just need to read your data; it needs to trust it.

A machine-readable record that says a product is in stock when it isn’t, lists two part numbers for one item, or carries a spec in the wrong units is readable and wrong — and an agent will act on it confidently, at scale, before anyone catches it.

So the real requirement has two parts:

The first part is a formatting exercise. The second is the genuinely hard, ongoing one — and it’s the part that decides whether agentic commerce actually works for you or quietly recommends against you. This is why product matching, entity resolution, and a maintained canonical product record are becoming front-office infrastructure rather than back-office hygiene.

The open question is maintenance, not creation

You can build a trusted product-data layer once. The problem is that a multi-supplier catalog never holds still. New suppliers, new files, price changes, spec revisions, and discontinued lines arrive continuously, and a layer that was trustworthy in January is stale by March.

This is why the durable answer isn’t a one-time data project or a static feed. It’s a continuous operation:

1. 1
   Detect each change

   Watch supplier feeds, ERP updates, PIM edits, price files, inventory changes, and catalog imports as they arrive.

2. 2
   Resolve product identity

   Decide whether each new or changed row represents an existing real-world product, a duplicate, a substitute, or a net-new item.

3. 3
   Validate and enrich

   Normalize units, fill missing attributes from known sources, flag conflicts, and attach confidence and provenance to every important value.

4. 4
   Write back to existing systems

   Push the trusted result into the ERP, PIM, commerce platform, or procurement workflow your team already runs.

5. 5
   Monitor for the next change

   Keep the layer current as suppliers revise files, internal teams edit records, and downstream agent requirements evolve.

That continuous loop is the layer Claro is built to be. We don’t replace your ERP, PIM, or commerce platform — we sit on top of them and keep the product and supplier data inside them resolved, validated, and current. Whatever reads your catalog next, whether a procurement agent, a shopping assistant, a search index, or your own team, is acting on data it can trust. For platform teams, the same pattern is the practical version of an AI layer on ERP and catalog systems: leave the system of record in place, but maintain a trusted product-data layer above it.

The bet is simple: the durable layer in agentic commerce isn’t a better model or a one-off enrichment. It’s a continuously maintained, trusted product identity underneath the whole stack.

Where this leaves you

If a real share of your future demand will be mediated by agents, the strategic move isn’t to pick an agent platform — those will keep changing — it’s to get your product data to a trusted, machine-readable state and keep it there. That’s the foundation every agent, protocol, and channel reads from, no matter which ones win.

Start with your highest-value SKUs and your most volatile supplier feeds. Get them canonical and current, prove an agent can read and verify them, then expand from a foundation you can trust.

Sources and article inspiration

This article was shaped by the same market signals catalog and commerce teams are now reacting to:

Related Claro resources

FAQ

What is agentic commerce infrastructure? +

The systems that let AI agents discover, evaluate, and transact on products on a buyer’s behalf — across payments, carts, and product data. The product-data part is the layer that exposes a catalog to agents in a structured, trustworthy form, and it’s increasingly seen as the bottleneck rather than the models themselves.

Isn't making my catalog machine-readable enough? +

Machine-readable is necessary but not sufficient. A structured feed that’s incomplete, duplicated, stale, or wrong is readable and unreliable — and an agent will act on it at scale. The requirement is machine-readable and trusted: one canonical record per product, validated, current, with confidence and provenance.

Who maintains trusted product data as the catalog changes? +

That’s the open question. A one-time cleanup or static feed decays as suppliers send updates. A continuous loop — detect change, resolve identity, validate, write back, monitor — keeps the data trustworthy over time. That ongoing maintenance, on top of your existing systems, is what Claro provides.

Claro keeps multi-supplier product and supplier data clean, matched, enriched, and validated across your existing ERP, PIM, and procurement systems. Your catalog should get smarter every day.