The narrative that artificial intelligence will decentralize retail is hitting a wall of established marketplace gravity. According to the PYMNTS Intelligence September 2026 report, “Will the 2026 Shopping Season Go Agentic?,” nearly 132 million U.S. adults have already used AI to assist in a retail purchase. This scale is significant, yet the data reveals a stark tension: rather than breaking the dominance of incumbent platforms, AI is currently acting as a force multiplier for them.
The report, which surveyed over 2,000 consumers and 60 merchants between September 2025 and August 2026, finds that 59% of AI-assisted purchases still conclude on Amazon. This figure suggests that agentic commerce is not yet the disruptive force many anticipated. Instead of enabling consumers to bypass traditional gatekeepers, AI tools are frequently funneling users back into the most familiar, high-trust ecosystems. For the retail sector, this indicates that the platform lock-in effect remains robust, even when the interface shifts from a search bar to an autonomous agent.
The structural reality of this market is best understood by the widening gap between discovery and execution. While 132 million Americans are using AI to find products or compare prices — with 46% of shoppers specifically using AI to hunt for the best deal — the actual volume of transactions completed by agents remains negligible. Data from Checkout.com indicates that only 3% of actual transactions currently involve agents. We are seeing a massive surge in AI-assisted research, but a profound hesitation when it comes to handing over the keys to the wallet.
This hesitation is mirrored on the merchant side, where the liability gap serves as a binding constraint. A staggering 93% of merchants surveyed by PYMNTS believe that the AI or agent provider should bear the financial loss when an agent makes an incorrect purchase or selection. Furthermore, 80% of merchants expect providers to verify an agent’s authority before it acts. Until there is a standardized framework for liability and authentication, merchants are unlikely to grant agents full, unfettered access to their inventory. Currently, only 28% of merchants are willing to offer agents their full product range on the same terms as other channels.
Consumer behavior reinforces this caution. While users are comfortable with AI as a research assistant, they remain deeply skeptical of its role as a financial proxy. Consumers place significantly more trust in established wallets, banks, and card networks than in AI platforms. This is supported by data from Worldpay, which shows that only 14% of consumers trust AI to execute purchases without manual verification. The infrastructure of trust is not being built by the AI models themselves, but by the financial institutions that have historically managed risk.
That structural gap explains why the industry is moving toward an infrastructure-first approach to bridge intent and payment. Mastercard’s projection of 300 million AI agent shoppers by 2030 relies on the assumption that these trust and liability hurdles will be cleared. Simultaneously, the launch of Stripe’s Shared Payment Token, which enables automatic agent checkout for any seller on the platform, represents the type of plumbing required to move from 3% transaction volume to something more substantial. These tools are designed to provide the security and verification that merchants and consumers currently lack.
Despite the current dominance of Amazon, the data offers a glimmer of hope for brands looking to reclaim their direct-to-consumer relationships. The PYMNTS report notes that 53% of consumers would be willing to leave a large marketplace if a brand’s website offered a lower price. This suggests that the platform lock-in is conditional rather than absolute. If agents can be programmed to prioritize price and value over platform convenience, the current dominance of the giants may prove vulnerable. The future of agentic commerce will likely be decided not by the sophistication of the AI, but by which infrastructure providers can successfully solve the liability and trust problems that currently keep the market in a state of perpetual research.