The agentic commerce wars have a discovery problem.

A Modern Retail story published Friday catalogs the unhappy gap between how fast retailers are launching AI shopping apps and how few consumers are actually using them. The numbers are striking: nearly 900 apps now live inside ChatGPT and another 353 connectors inside Claude, with roughly 10% of the ChatGPT apps tied to shopping. Retail Dive's coverage of the launch wave noted the rapid build-out across Target, Walmart, Sephora, Instacart, Etsy, and dozens of others. And the candid quote in the piece — from Dimitri Ewald, chief of staff at Alpic, a French firm building these apps — was that "for the moment, to be honest, adoption and conversion are pretty low."

This is the first real reality check on the trend Endcap has been tracking for weeks, from Shopify's Agentic Storefronts launch to Google's UCP rollout to Etsy's bet on putting its catalog inside ChatGPT. The supply side is moving fast. The demand side hasn't shown up.

Why discovery is the real bottleneck

The single biggest finding in the Modern Retail piece is structural: ChatGPT app discovery is broken. The Sephora app, as Modern Retail noted, won't surface when a user asks ChatGPT for cosmetics advice. The shopper has to know the Sephora app exists, navigate to the ChatGPT app store, install it, and then explicitly invoke it. That breaks the entire promise of agentic shopping, which was supposed to be context-driven and frictionless.

Claude, according to RetailWire's analysis, takes a different approach. Claude decides which connector to use based on user intent, not user invocation — but it still requires the connector to be installed first. The discovery problem moves but doesn't disappear.

What's striking is that this is a self-inflicted problem. Both OpenAI and Anthropic could solve discoverability with platform-driven recommendations, native catalog browsing, or contextual prompts. They haven't. The likely reason: doing so would require them to take editorial responsibility for which retailer surfaces in which conversation — a curation problem with antitrust and trust implications neither company seems eager to absorb.

Why the adoption data is so muted

There are three plausible reasons why the consumer side hasn't shown up yet.

Habit. Shoppers spent 25 years learning Amazon search, Google search, and Instagram discovery. Asking them to install and launch app-by-app inside an LLM is an enormous behavior shift, and the value proposition has to be much better than "the same products you could buy through your usual flow."

Conversion friction. Modern Retail noted that integration depth varies wildly. Instacart lets you check out inside ChatGPT. Most other apps still kick you out to a retailer's site or app to complete the purchase. Consumers who try the experience once and get punted to a separate checkout flow are unlikely to return.

Pricing transparency. Agentic interfaces work best when users are looking for the right answer, not a deal. But shoppers in 2026 — squeezed by tariffs, gas prices, and inflation — are price-shopping more aggressively than they have in years. LLM-mediated discovery doesn't yet do head-to-head price comparison well, and that's a structural problem with how these models retrieve and weight information.

What the smart retailers are doing differently

The pattern that's emerging from retailers seeing higher engagement on ChatGPT and Claude apps tracks closely to retailers that already have content-driven brand affinity. Sephora's app sees more activity than commodity retailers' because Sephora customers actively ask for product advice — exactly the use case LLMs handle well. Etsy's app benefits because its long-tail catalog rewards conversational discovery.

Retailers selling commodities at scale — packaged goods, basic apparel, household staples — are getting effectively zero return on app development right now. That's not a temporary problem. It's a structural mismatch between LLM mediation and the kinds of purchases shoppers want to make quickly.

Modern Retail's takeaway — and the same one echoing through Modern Retail's analysis of ChatGPT's Instant Checkout misfire — is that the AI conversation is shifting from "build it" to "prove it." That's a healthy turn. It also means the next 12 months will separate retailers running real attribution against agentic channels from those treating these apps as marketing optics.

The bigger pattern

The agentic commerce wave is real. The infrastructure is being built. But Friday's Modern Retail piece is the first widely-reported acknowledgment that the user side of the equation isn't responding the way the platforms predicted.

For retail leaders deciding whether to fund a deeper ChatGPT or Claude integration: the right question isn't whether agentic shopping will be big. It's whether your specific catalog rewards conversational discovery — and whether you're willing to wait through 18-24 months of low adoption before the platforms fix discovery on their end.

Right now, building these apps is a hedge, not a growth channel. That's a fine reason to do it. It's a bad reason to skip the harder work on first-party search, recommendation, and checkout that's still where 95% of digital revenue actually lives.