In December 2025, Visa reported that it had completed “hundreds” of secure, agent-initiated transactions in a closed beta. By the 2026 holiday season, the company projects that “millions” of consumers will use AI agents to complete purchases. This numerical gap — the distance between a few hundred test cases and a multi-million-transaction reality — is the adoption ceiling for agentic commerce.
The Structural Barriers
The transition from experimental beta to mass-market utility is currently obstructed by three primary structural barriers: a persistent consumer trust deficit, a standoff over merchant liability, and the friction of protocol proliferation. While 47% of U.S. shoppers use AI for at least one shopping task, the leap to authorizing an agent to execute a payment remains a significant hurdle.
Consumer trust is the most immediate bottleneck. According to the Visa Earning Trust Report, only 23% of U.S. consumers trust generative AI to handle payment transactions. While 61% express specific trust in Visa, the broader sentiment is one of caution; approximately 85% of consumers demand transparency and control over how their data is used, and half indicate they would cease using agents entirely if that control were removed. This aligns with data from PYMNTS Intelligence, which found that only 14% of consumers trust AI to execute purchases without manual verification.
The Liability Wall
Merchants are equally hesitant, primarily due to the unresolved question of financial risk. PYMNTS Intelligence reports that 93% of merchants believe the AI provider should bear the financial loss for incorrect purchases. Consequently, only 28% of merchants are currently willing to offer their full product range to AI agents. This risk aversion is the top-ranked barrier for acquirers, who are calling for a complete redesign of the liability framework before they can scale.
The OpenAI retreat from in-chat checkout in early 2026 serves as a case study in structural resistance. Data from Walmart indicated that in-chat checkout converted at only one-third the rate of on-site transactions, leading the industry to consolidate around a “discover in AI, buy on own site” model rather than full agent-led execution.
Protocol Proliferation
As noted in our previous analysis of the Protocol Proliferation Tax, the technical landscape is fragmented. There are at least five competing checkout protocols — including Visa Intelligent Commerce, Mastercard Agent Pay, Stripe ACP, Google UCP, and Meta Muse — each requiring distinct integration efforts. For merchants, the cost of supporting these disparate standards can range from $5,000 to $500,000 per protocol. This fragmentation acts as a binding constraint on adoption, forcing merchants to choose winners in an environment where no standard has yet achieved dominance.
Visa is attempting to bridge this gap with Intelligent Commerce Connect, a single integration platform designed to be network-, protocol-, and token-vault-agnostic. While it supports multiple standards, including the Trusted Agent Protocol and UCP, the platform remains unproven at the scale required to hit Visa’s holiday 2026 targets. Meanwhile, competitors like Mastercard are moving cautiously; their first live end-to-end agent payment in March 2026 was explicitly labeled as not constituting a commercial rollout.
The Ten-Week Horizon
With Black Friday 2026 approaching, the industry faces a compressed timeline. While Hypertrade reported a 4,700% year-over-year increase in AI-generated traffic to retail sites, the conversion of that traffic into actual payments remains minimal. Bernstein research notes that agentic commerce currently accounts for less than 1% of U.S. e-commerce, characterizing the current state of the market as the “boy who cried wolf.”
The industry is currently caught between ambitious projections and the reality of a fragmented, risk-averse ecosystem. Until the liability impasse is resolved and the integration landscape is simplified, the “millions” of transactions promised for the holiday season will likely remain a theoretical target rather than a commercial outcome.