​The AI Shopping Agent Is Easy; The Commerce Platform Behind It Is Hard

Vikas Mittal, Principal Architect building real-time AI platforms for retail enterprises at Walmart Global Tech.

​AI is making it easier to understand what customers want, but understanding intent is only one part of completing a commerce transaction. Behind every AI shopping experience sits a complex platform responsible for catalog data, pricing, inventory, eligibility, fulfillment and transactions. As shopping becomes more agentic, companies should focus as much on modernizing these underlying capabilities as they do on the AI experience itself.

Imagine telling an AI assistant, “Find me a 65-inch TV under $800 that can be delivered before the game this weekend.”

The AI can understand that request remarkably well. It can interpret the product category, budget, size and delivery constraint, compare specifications and narrow hundreds of products down to a few choices.

After more than two decades of building software products, including several years working on large-scale retail and e-commerce platforms, I have learned that the interface is rarely where the hardest engineering lives.

AI Understands Intent But Commerce Platforms Establish Truth

Large language models are particularly good at interpreting human intent. A customer doesn’t need to know a product taxonomy or construct the perfect search query. They can simply describe what they are trying to accomplish.

But understanding what the customer wants and determining what a retailer can actually sell and fulfill are different problems. The AI layer can handle intent, discovery, comparison and reasoning. The commerce platform still has to establish authoritative answers for product information, pricing, inventory, eligibility, fulfillment and transactions.

If those signals were inconsistent, the worst outcome was accepting an order only to discover later that it could not be fulfilled and had to be canceled.