While 89% of merchants are actively preparing for agentic commerce, only 3% of transactions currently involve autonomous agents, according to the Checkout.com 2026 report. This gap between preparation and execution is not merely a technical hurdle; it is a structural collision. As the industry attempts to scale, the control points of digital commerce are fragmenting across infrastructure layers, with the fundamental question of who controls the gate splitting rather than consolidating.
Merchants are increasingly reclaiming protocol-level control to manage how agents interact with their storefronts. Shopify recently implemented robots.txt restrictions to block AI crawlers from checkout and cart paths — the first major eCommerce platform to draw the line at the protocol level. Amazon has blocked Meta Muse agent checkouts due to concerns over unauthorized access and credential capture, as GeekWire reported. Only 11% of SMBs are currently agent-ready, per PYMNTS/Visa data, and just 28% of merchants are willing to offer agents their full product range. These moves signal a defensive posture against unvetted agent traffic, forcing a confrontation between open-access AI models and the walled gardens of established marketplaces.
Simultaneously, the underlying settlement infrastructure is rapidly maturing. The institutional stablecoin stack reached a critical milestone this week, integrating reserve management, enterprise payment, and credit assessment. Moody’s issued its first-ever stablecoin protocol rating, providing a baseline for risk, while the Citi-Coinbase collaboration wired its $6 trillion payment network to stablecoins. Furthermore, SAP Pay stablecoin settlement is embedded directly in its ERP software, effectively turning enterprise resource planning into a payment rail. The plumbing is shipping. The gate is not.
Despite these infrastructure advancements, the liability framework remains unallocated. PYMNTS Intelligence data shows that 93% of merchants believe AI providers should bear the financial loss for agent errors, yet no clear mechanism for this exists. While six major banks recently published voluntary guardrails, these principles lack enforcement and fail to address the critical liability gap. The party best positioned to prevent the error — the AI provider — has no financial incentive to do so.
Federal Reserve Governor Christopher Waller, speaking at Sibos, identified the core market-structure pivot: the choice between open and closed systems. “The biggest barrier to scaling agentic commerce, particularly the agent-delegated model, is building sufficient trust among buyers and sellers,” he said, naming authentication, liability, and fraud as the three binding challenges. Waller’s full remarks on market structure framed the open-versus-closed question as the variable that “could significantly influence how the market structure evolves.” Merchants are currently buying into infrastructure that is actively shipping, but they are doing so without the binding standards necessary to mitigate the risks of autonomous purchasing.
The market is currently defined by a proliferation of competing protocols — from Google and OpenAI to Mastercard and Visa — each vying for dominance in the agent-delegated model. Yet Walmart pilot data showed conversion rates roughly three times lower for direct agent checkouts than for click-out flows, according to Walmart executive Daniel Danker, first reported by Wired. The industry is left with a fundamental question: will the gate be governed by the merchant, the platform, or the protocol? Until liability is clearly assigned, the gate will remain a point of friction rather than a gateway to scale.