AI shopping agents are coming. No one is ready for them

Panelists at Fortune Brainstorm Tech said AI shopping agents are not yet ready for broad commerce use because of security protocols, missing agentic-commerce standards, retailer resistance, and unresolved liability for fraud, refunds, and returns. OpenAI’s commerce push was cited as having faltered, including the pullback of its Instant Checkout feature and retailer exits such as Walmart, while ChatGPT was said to return specific product recommendations only 9% of the time for shopping queries. The most immediate implications are for retailers, fintech/payment infrastructure, and cybersecurity vendors rather than the broader market.

Analysis

The near-term investable signal is not “AI commerce” itself, but who owns the choke points if agentic shopping scales before standards do. That setup favors platform and infrastructure names with authentication, fraud scoring, checkout orchestration, and identity layers; it is structurally awkward for retailers that want to preserve control of customer data, margins, and returns economics. If agent traffic grows without clean liability rules, merchants will likely respond by tightening access, increasing friction, or charging higher fees for agent-initiated transactions, which pushes the value pool upstream to the toll collectors rather than the sellers.

For WMT and GAP, the first-order revenue risk is modest, but the second-order effect is worse: agents compress brand differentiation and can make product selection more price-driven, which is margin-negative for discretionary retail. The bigger issue is returns/fraud provisioning; even a small rise in disputed transactions can offset a lot of the incremental conversion benefit, especially in categories with high AOV or fit risk. Over 6–18 months, retailers that are most exposed to digital traffic but least able to authenticate intent may see a creeping drag on operating expense and customer-service load before it shows up in reported comps.

The contrarian view is that this transition may be slower than the market assumes, and that the “winner-take-most” economics could be underappreciated once standards finally emerge. If a few identity or checkout protocols become default, the market may rerate a narrow set of enablers very quickly, while merchants who resisted early integration could be forced into less favorable economics later. In other words, the biggest alpha may come from owning the picks-and-shovels now, not from betting on headline AI commerce adoption.

Key Decisions for Investors

• Short WMT vs long a basket of identity/fraud-enablement names over 3-6 months: WMT faces rising friction if it blocks agentic checkout, while the enablers capture the economic rent from authentication and dispute handling. Use a 1:1 dollar-neutral structure; stop if Walmart announces a materially positive agent commerce partnership.
• Reduce long exposure to GAP on any strength; use 2-4 month downside puts or put spreads. The risk/reward skews negative because discretionary apparel is vulnerable to agent-driven price transparency and return-cost inflation; downside accelerates if online return rates tick higher.
• Long payment security / identity infrastructure exposure for 6-12 months, preferably via a basket rather than a single name. The trade works if fraud concerns delay broad rollout but force enterprise spend into monitoring, verification, and authorization tooling before consumer adoption becomes seamless.
• Consider a pair trade: long cybersecurity/fraud-prevention tech vs short retail discretionary basket. This isolates the second-order winner from the adoption bottleneck; expect the spread to perform if merchants harden defenses faster than they open checkout rails.
• Do not chase pure-play AI shopping agent names until liability standards are clearer; if you want optionality, use small call spreads with 9-12 month maturities. The path dependency is high, and premature adoption claims are more likely to disappoint than surprise positively in the next 1-2 quarters.