:quality(75))
Adobe Summit ran April 19-22, 2026 at The Venetian in Las Vegas. Strip out the keynote staging and the Commerce story was narrower than usual, and more interesting for it: nearly all of it was about making a storefront readable by something that isn't a person.
The Commerce MCP server
The headline is a Commerce Model Context Protocol server, which lets a brand expose its own structured commerce data — catalogue, pricing, inventory, promotions — directly to AI systems.
The significance is the direction of travel. Until now, getting an AI assistant to answer questions about your range meant hoping it had scraped your site correctly. An MCP server inverts that: you publish the data, in a shape agents can consume, and stop relying on a crawler's interpretation of your markup.
Alongside it came new B2B drop-in components and Brand Concierge for storefronts, which brings product detail, pricing and checkout into a conversational experience.
The number behind it
Adobe's own figure is the one worth writing down: AI traffic to US retail sites climbed 269% year-over-year in March 2026.
That is a growth rate, not a volume, and the base is small — nobody should reallocate a budget on it. But the trend is consistent with what we see in client analytics, and it is the reason the LLM Optimizer announcement exists at all: brands increasingly appear in answers they did not write and cannot see.
What we would actually do
Not much, yet, and we would be wary of anyone telling you otherwise. Announced is not shipped, and an agent-native storefront is not a Q3 project for most mid-market merchants.
What is worth doing now costs nothing: find out how your brand currently gets described by the major assistants, and whether the product data behind those answers is yours or a guess. That is a half-day exercise, and it tends to be a bracing one.