Kushal Khatri
Image courtesy of Kushal Khatri

A new kind of shopper has arrived, and it is not human. AI agents are starting to search, compare, and buy on people's behalf, reading product data and deciding, sometimes with no person in the loop, which items to surface or purchase outright.

For merchants, that changes what their product data must do. It no longer only has to look right to a person, but it has to be comprehensible to a machine that is making the choice. Getting merchants ready for that shift is the problem Kushal Khatri has been working on for the last few years.

A Different Buyer, Different Stakes

Kushal is the Chief Technology Officer of Feedonomics, the AI-powered feed management arm of Commerce.com, Inc. (Nasdaq: CMRC), formerly BigCommerce Holdings, Inc., and an industry-leading product in feed management. He is also one of the leaders steering the parent company's engineering strategy for AI and agentic commerce.

Feedonomics is the system merchants use to harmonise, optimise, enrich, and syndicate product data across sales channels. Increasingly, Kushal's attention is on a new question: making that data ready for channels where the audience is an agent rather than a browsing shopper.

The stakes change when the buyer is software. A person who lands on a poorly listed product might still scroll, compare, and forgive a missing detail, and may even bring a bias of their own. An agent usually will not.

An agent works from both structured and unstructured data, and if a product is miscategorised, missing an attribute, or unclear on price or availability, the agent tends to skip it. In that setting, Kushal argues, weak data does not just lower conversion; it takes the product out of consideration before a sale is even possible.

He puts it in a line that captures how Commerce and Feedonomics think about the role of data in the new age of e-commerce: 'Data is the new storefront.'

The product listing on a merchant's website, he means, has given way to the title, the attributes, the taxonomy that informs enrichment, and the structured and unstructured information an agent reads to decide whether a product exists as far as it is concerned.

Building for the Modern Agentic Protocol

That shift is why an emerging set of standards suddenly matters. In January 2026, Google launched the Universal Commerce Protocol (UCP), an open standard built in collaboration with a group of industry leaders, including Commerce among them, that gives agents and merchant systems a shared language for the whole shopping journey.

Commerce endorsed it at launch and has been building towards the protocol ever since, so that merchants can enable buying across Google's agentic surfaces and other agentic systems that adopt UCP. In its own announcement, the company pointed to Feedonomics as the part that sets its offer apart, describing the data enrichment it provides as the key differentiator in an ongoing partnership with Google.

An agent cannot surface or buy what it cannot read, and most catalogues are not clean enough for a machine to act on with confidence. Feedonomics is the system many large retailers use to close that gap, conditioning and enriching their product data, getting it into the shape the protocol expects, and syndicating it to the surfaces where agents operate.

Roughly 30 percent of the top 1,000 internet retailers, along with many Fortune 500 businesses, rely on Feedonomics to keep their data accurate, compliant, and ready for channels like UCP.

Kushal sits close to the centre of that work. He runs the engineering teams that build the enrichment and syndication systems merchants use to get ready for UCP and to move their catalogues onto agentic as well as non-agentic surfaces. According to the company, Kushal and members of his team have worked directly with Google's UCP group as the protocol took shape and as merchants began onboarding to it.

The Enrichment Problem

Underneath the standard lies the harder problem of turning messy catalogues into something an automated system can trust. Feedonomics leans on advanced solutions it has built using leading GenAI models alongside its own proprietary machine learning, including a categorisation engine called FeedAI, to classify and enrich products so software, not only people, can make sense of them.

The company says FeedAI classifies products with up to 98 percent accuracy across all key categories, work that once relied on slow manual tagging and that matters far more once software, and especially an agent, is reading the result: a miscategorised product may never be discovered at all.

The company says it has also built Agentic Catalog Exports and connections to AI answer engines beyond Google, part of an effort to keep merchant products visible wherever agents are beginning to shop. The aim, in Kushal's framing, is simple to state and hard to deliver: a product should be understandable to any agent that encounters it, on any surface, without a person stepping in to fix it first.

For merchants, the payoff Kushal points to is growth: products that surface and get chosen as more buying moves through agents, rather than quietly dropping out of a channel they may not even know they have lost. Most of the attention on AI in commerce still goes to the chatbots and the demos. His wager is that the businesses that win the agentic shift will be the ones whose data was ready for it, and that readiness is built long before an agent ever runs.