The Doors Are Locked: How Retailer Postures Are Stalling the Agent Commerce Stack
The digital storefront is currently a house divided, where the architecture of commerce is being retrofitted for autonomous agents while the doors remain firmly bolted. When an AI agent attempts to navigate a merchant site, it is not merely encountering a technical interface; it is colliding with a set of proprietary gatekeeping strategies that define the current agentic commerce landscape. The friction here is not a failure of the underlying code, but a fundamental disagreement on who owns the transaction flow.
Retailers have settled into three distinct postures. Some are welcoming hosts, proactively feeding catalogs to OpenAI and Google to ensure visibility. Others are more cautious, allowing agents to browse aisles but slamming the door when it comes time to process a payment. Then there are the digital bouncers, like Amazon, which blocked Meta’s Muse agent on September 20, 2026, citing unauthorized access and data privacy concerns, as detailed in a Wall Street Journal report by Patrick Coffee on October 2, 2026. Amazon’s history here is instructive: it previously sued Perplexity over its Comet AI agent, winning a preliminary injunction in March 2026 only to have it reversed by the Ninth Circuit in August. The court ruled that the user, not the AI company, accessed Amazon’s computers under federal anti-hacking law. Consequently, Amazon’s current Muse block pivots to Conditions of Use rather than hacking claims, arguing the agent fails to identify itself and captures sensitive account data-a characterization Meta disputes. As Tapestry CIDO Yang Lu notes, “People are adopting agentic commerce much more slowly than other AI consumer tools; we’re comfortable using AI to browse and discover products, but aren’t yet ready to hand over our credit cards.”
eBay’s recent maneuvers reveal that these blocks are often about more than security. In January 2026, the company updated its user agreement to ban most third-party buy-for-me LLM agents, while carving out an exception for OpenAI Operator. This coincides with a new “any-click” attribution model that charges sellers ad fees if a user clicks a promoted listing and subsequently purchases within 30 days. By blocking external agents, eBay effectively obscures the audit trail, making it difficult for sellers to verify attribution claims and ensuring the platform retains total control over the transaction flow.
This institutional hesitation is mirrored in the data. A PartnerCentric survey from January 2026 found that only 3% of US consumers view chatbots as autonomous purchasing agents, while the Visa Trust Index 2026 confirms that only 14% of consumers trust agents to act on their behalf. VML Tomorrow’s Commerce 2026 adds that one-third of active AI users would never let an agent spend their money. Merchants are responding to this skepticism with extreme caution. The PYMNTS/Visa GDSI Merchant Edition reveals that only 11% of SMBs are currently agent-ready, with a mere 15% possessing the structured product data necessary for agents to function. Perhaps most tellingly, only 23% of merchants can even distinguish between AI and human traffic.
The technical plumbing is already finished. We have a fully functional agentic commerce protocol : Visa TAP, Mastercard Agent Pay, Stripe SPTs, and Google UCP are all live. Shopify enabled agent-readiness by default, driving an 8x year-over-year increase in AI traffic during the first quarter of 2026. Yet, a PYMNTS test from September 9, 2026, saw the Muse agent fail three consecutive errands-reordering toilet paper, ordering a pizza, and making a reservation-not because the AI lacked reasoning, but because of credential friction and app connection failures.
Federal Reserve Governor Christopher Waller addressed these structural hurdles in his September 29, 2026, speech at Sibos . He identified three primary barriers: authentication, liability, and fraud. Waller noted that current systems are calibrated for human patterns, making them ill-equipped to handle the speed and volume of autonomous agents. He framed trust as a market-structure problem rather than a regulatory one, suggesting that the market must build the trust layer itself, as the Fed will not mandate how a retailer handles an AI agent.
The agentic storefront is not a technical impossibility; it is a policy choice. We have the pipes, but the doors are locked, and the bouncers are still deciding who gets to come in. Until merchants can agree on a standard for how agents identify themselves and how liability is shared, the agentic economy will remain a collection of isolated, high-friction experiments rather than a cohesive, functional ecosystem.