What Is Agentic Commerce? How AI Agents Discover Products
Online shopping has traditionally started with a search bar.
A customer enters a few keywords, scrolls through links, opens several product pages, compares the details, and eventually decides what to buy. Every step depends on a human reading and interpreting the storefront.
Agentic commerce changes that process.
Instead of personally searching through dozens of pages, a shopper can describe what they need to an AI agent:
“Find me a waterproof carry-on backpack under $150 that will arrive before Friday.”
The agent can then search for suitable products, compare prices and availability, evaluate whether each option satisfies the request, and return a short list of recommendations.
In more advanced cases, the agent may also help assemble a cart, initiate checkout, or manage an order.
This emerging model is known as agentic commerce. It replaces part of the traditional browsing process with software that searches and reasons on the customer’s behalf.
For merchants, that creates a new question:
Can an AI agent actually understand your products well enough to recommend them?
How AI agents see an online store
A human can look at a product page and intuitively identify its most important details.
We can usually distinguish the product name from the navigation menu. We can recognize that a crossed-out number is an old price, that a row of circles represents color options, or that a disabled button means an item is unavailable.
An AI agent does not necessarily interpret the page the same way.
It may receive only the page’s underlying HTML, structured metadata, or text extracted by a crawler. Interactive components may not render. Important information may be distributed across scripts, images, variants, and application state.
That means a storefront can look excellent to a customer while remaining surprisingly difficult for an automated system to understand.
A product may become effectively invisible when an agent cannot confidently determine:
- What the product is
- Who sells it
- How much it costs
- Which variants are available
- Whether it is currently in stock
- Where the customer can purchase it
- When the information was last verified
Traditional search engines can often tolerate some ambiguity because they primarily return links for a human to evaluate. An agent attempting to compare or recommend products needs much more certainty.
Agentic discovery is not the same as SEO
Search engine optimization remains important. Product pages should still have useful titles, descriptions, canonical URLs, internal links, and crawlable content.
However, ranking a webpage and supplying a reliable product record are different problems.
A search result might tell an agent that a page appears relevant. It does not always give the agent enough structured information to determine whether the product actually satisfies the shopper’s request.
Agentic discovery depends on three layers working together:
- Discovery: The agent must be able to find the product.
- Understanding: The agent must be able to interpret its attributes.
- Trust: The agent must be able to determine whether those attributes are current and authoritative.
A product page can succeed at the first layer while failing at the other two.
The information agents need
A useful machine-readable product record should contain more than a title and URL.
At a minimum, it should communicate:
- Product name and description
- Merchant identity
- Canonical product URL
- Product images
- Price and currency
- Availability
- Variants or offers
- Brand and category
- Product identifiers, when available
- Observation and refresh timestamps
The origin of that information matters as well.
There is a meaningful difference between a price supplied by a verified merchant and one copied from an unknown third-party page. There is also a difference between inventory checked five minutes ago and inventory observed several weeks ago.
For an agent to make a dependable recommendation, product data needs both structure and provenance.
The emerging protocols behind agentic commerce
There will not be one universal method through which every AI system discovers products. Different agents and platforms require different interfaces.
Several technologies are already becoming important.
Structured data
Schema.org markup, including Product and Offer data, helps crawlers identify the meaning of information embedded in a webpage.
This remains one of the most accessible ways for merchants to improve machine understanding. However, implementation quality varies significantly, and structured data does not automatically guarantee that every value is current or complete.
Product feeds
CSV, XML, and JSON feeds allow platforms to ingest large catalogs without crawling every product page individually.
Feeds are efficient for bulk distribution, but different platforms frequently expect different formats. They may also become unreliable when exports are generated infrequently or disconnected from the live storefront.
REST APIs
A conventional API gives applications a direct way to search products and retrieve structured records.
REST remains valuable because it is widely supported, predictable, and easy to integrate into existing software. An agentic application can translate a shopper’s request into an API query and receive ranked product candidates.
Model Context Protocol
The Model Context Protocol allows AI applications to discover and invoke tools exposed by external services.
Instead of teaching an agent the implementation details of a product database, an MCP server can expose clearly described actions such as searching a catalog or retrieving a product record.
This makes MCP particularly useful for tool-using agents.
Universal Commerce Protocol
The Universal Commerce Protocol defines commerce capabilities that agents and businesses can negotiate through a shared standard.
Its scope includes catalog search and lookup as well as later stages of the commerce journey, including carts, identity, checkout, and order management.
These protocols are complementary rather than interchangeable. A serious product-discovery layer should be available through multiple channels instead of assuming every agent will use the same one.
Why freshness matters
Product information changes constantly.
Inventory can disappear in minutes. Prices change during promotions. New variants are introduced. Product pages move. Entire listings are discontinued.
A static product export can remain technically readable while becoming commercially misleading.
Agentic commerce therefore requires more than publishing data once. Product records need to be observed, refreshed, and marked when their current state can no longer be confirmed.
A useful system should be able to answer questions such as:
- When was this price last observed?
- When will the availability be checked again?
- Has the merchant verified control of the source?
- Which page supplied this field?
- Is the product active, stale, unpublished, or withdrawn?
This is the difference between providing product data and maintaining product infrastructure.
What merchants should do now
Agentic commerce is still developing, but merchants do not need to wait for every protocol and platform to settle before preparing.
The most valuable steps are practical ones:
- Keep product pages publicly accessible and crawlable.
- Use consistent canonical URLs.
- Add accurate Product and Offer structured data.
- Maintain complete prices, availability, images, and descriptions.
- Avoid hiding essential product information exclusively inside client-side interactions.
- Publish machine-readable feeds when possible.
- Establish a reliable process for refreshing product data.
- Make it clear which merchant is responsible for each product.
These improvements help conventional search engines today while making the catalog easier for future agents to use.
Where Nexus fits
ULIX Nexus is a maintained discovery layer between merchant storefronts and AI agents.
A merchant connects and verifies a store domain. Nexus discovers product pages, extracts their commerce data, and converts valid products into structured canonical records. Those records are refreshed over time and distributed through public search, REST, MCP, UCP, structured representations, and product feeds.
Nexus does not replace the storefront. Customers still purchase from the merchant’s canonical product page.
Its purpose is to make the catalog easier for machines to discover, understand, and trust.
Agents can access the public infrastructure through the Nexus agent gateway, while merchants manage their catalogs through the Nexus console.
The storefront is no longer the only audience
For years, merchants have designed product pages for two audiences: customers and search engines.
AI agents introduce a third.
This audience does not care how polished a page looks. It cares whether the underlying product information is complete, current, unambiguous, and available through an interface it can reliably use.
The merchants who prepare for that shift will not abandon traditional search. They will extend their catalogs beyond it.
The next generation of product discovery will not always begin with someone typing into a search bar.
Sometimes, it will begin with a person describing what they need and an agent deciding which products deserve to be considered.