Analysis · 4 October 2026 · Payments, AI
On 29 September, Federal Reserve Governor Christopher Waller spoke about "Payments in the Age of AI Agents" at Sibos in Miami. His summary was careful: "Market participants broadly agree that agentic commerce is in an early phase, but it could significantly reshape commerce and payments if adoption scales" (Federal Reserve).
The important words in that sentence are the last three. If adoption scales.
A day later, Mastercard expanded Agent Pay with a score that estimates how likely it is that a transaction was started by an AI agent. It is rolling out for testing in the US (Mastercard). It capped a month in which almost everyone shipped something. Meta's Muse agent launched with Stripe's Link checkout, usable at more than 1 million businesses (Stripe). Stripe made every hosted checkout page agent-ready, covering more than 7.8 million businesses (Stripe announcement, via WebMCP). Shopify switched agent checkout on by default for eligible merchants (Forkast). And six banks, among them Bank of America, Capital One, ING and NatWest, published shared principles for "trusted agentic commerce" (ING).
The rails are being laid at remarkable speed. The question is who is riding on them.
The 3% problem
The most sobering number in agentic commerce comes from Checkout.com's Agentic Commerce 2026 report. According to the merchants it surveyed in the UK and US, only 3% of their transactions involve AI agents, even though 89% say they are actively preparing for agentic commerce (Checkout.com). 42% are already testing it (Checkout.com). It is a self-reported figure from a merchant sample, not a market statistic, but it is the clearest view we have of the checkout.
Consumer data points in the same direction. PYMNTS Intelligence, surveying more than 2,000 US consumers and 60 merchants, finds that around 132 million American adults have used AI to help with a retail purchase. But 59% of those AI-assisted purchases still end on Amazon (Yahoo Finance). People let the AI research. Then they go back to the checkout they already trust.
Trust is the missing piece. In Forrester's March 2025 Consumer Pulse Survey, only 24% of US online adults said they trust AI agents to make routine purchases on their behalf (Forrester).
Why the numbers disagree
Readers of this magazine will remember a much bigger figure. In March, we reported Salesforce's estimate that AI and agents were behind $262 billion in holiday sales. Both numbers can be true, because they measure different things.
Waller's speech offers the most useful distinction. In the agent-assisted model, the buyer makes the decisions and handles the payment. In the agent-delegated model, the buyer authorises the agent to shop and pay on its own (Federal Reserve). Most of what is counted as "AI commerce" today is the first kind: influence. Almost all of the new infrastructure is built for the second kind: execution.
"Agentic commerce is not a reality yet," Hedera AI Studio product manager Lindsay Walker told The AI Conference in San Francisco. What exists today, she argued, is agent-assisted shopping, in which humans still make the decision at the point of purchase (The Register).
Liability, not technology, is the bottleneck
Waller named the core barrier directly: "The biggest barrier to scaling agentic commerce, particularly the agent-delegated model, is building sufficient trust among buyers and sellers" (Federal Reserve).
Merchants have a clear view of who should carry that risk. In the PYMNTS merchant survey, 93% said the AI or agent provider should bear the loss when an agent makes a wrong purchase, 80% expect providers to verify an agent's authority, and only 28% are willing to offer agents their full product range on the same terms as other channels (Yahoo Finance). The AI providers, unsurprisingly, have not volunteered.
Platforms are drawing their own lines. In September, Amazon began blocking purchases through Meta's Muse agent. "We think it's fairly straightforward that third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions about whether or not to participate," an Amazon spokesperson said (Quartz via Yahoo Finance). An open agent economy and a walled marketplace are now in direct conflict, and the marketplace still holds most of the checkouts.
This is why the most interesting recent launches are not really about agents. Mastercard's new score is fraud tooling, and the banks' principles are about transparency, safety and control. The hard problem is not getting an agent to pay. It is deciding who is responsible when it pays wrongly.
Where agents are already at work
There is one place where agents are not a rounding error. Stripe reported in June that 70% of requests for API resources through its command-line tool now come from agents (Stripe). Those are developer requests, not payments. But they show agents already acting on behalf of businesses inside financial infrastructure. Gartner expects AI agents to intermediate more than $15 trillion in B2B spending by 2028 (Digital Commerce 360), a forecast rather than a measurement.
The first agent economy is therefore unlikely to be a consumer handing a chatbot their credit card for groceries. It is more likely to be software buying compute, data and services from other software, with predefined budgets and clear mandates. That is also the world in which BlackRock expects agents to run on stablecoins.
Slow, not failed
Mastercard's own forecast, based on futurist predictions and consumer research, is that more than one in ten online shoppers could routinely use agents to shop and pay by 2030 (Mastercard). That is a meaningful market, and it also leaves the large majority who will not.
For banks, PSPs and merchants, the conclusion is less exciting than the launch month suggested, and more useful. Build for agent-assisted commerce now, because that is where the traffic is. Treat agent-delegated payments as a trust and liability product, not a checkout feature. And watch B2B, where agents are already doing work on behalf of their owners.
The infrastructure is ready. The rules of responsibility are not. Until they are, the agent will keep doing the research, and the human will keep pressing "buy."
· · ·
This article was researched and written with AI assistance for FinTech Weekly. All facts are linked to their sources in the text.
