Our Blog

Blog Index

AI Agents Get Wallets: How Agentic Payments Are Reshaping Online Commerce

Posted on 4th Aug 2026 06:03:54 in Artificial Intelligence, Machine Learning

Tagged as: agentic commerce, AI agents, agentic payments, Visa, OpenAI

For three decades, online commerce has ended the same way: a human clicks the buy button. That era is closing. In 2026, software agents are beginning to shop, compare, negotiate and pay on behalf of their owners, and the world's largest payment networks are racing to build the rails underneath them.

At its payments forum in San Francisco this June, Visa announced that it had embedded its payment network inside ChatGPT, letting an AI agent not only recommend a product but complete the purchase at any merchant that accepts Visa. Tell the chatbot you want wireless headphones under $150, Visa's chief product and strategy officer Jack Forestell explained, and it finds a pair and buys them for you. Mastercard moved on the same day, while Google, Stripe, Alibaba and Amazon are all building competing layers. Agentic commerce, the practice of machines transacting with wallets of their own, has crossed from experiment to infrastructure.

What Agentic Payments Are and How They Work

Agentic commerce is a form of online shopping in which an AI agent finds, compares and sometimes completes purchases for a customer. The customer states an intent, for example "book me the cheapest nonstop flight that lands before noon," and the agent evaluates the options and, once the human approves, finishes the checkout without the person ever seeing the payment page.

The security machinery is what makes this more than a party trick. Rather than handing a chatbot a raw card number, the payment networks issue what Visa calls an "agentic token": a scoped credential that ties a tokenized card to one specific agent, a defined merchant scope and a consent policy controlled by the consumer, complete with spending caps and instant revocation from a banking app. If a single token is compromised, exposure is limited, and the design prevents replay attacks where a captured credential is reused in a different context.

Control stays with the buyer. User-defined spending limits, merchant-category restrictions and required-approval thresholds are enforced in real time, before any agent-initiated payment is authorized. OpenAI's Agentic Commerce Protocol goes further with a delegated payment specification that issues one-time payment requests with spending caps and expiration, meaning a ChatGPT agent cannot reuse a credential or exceed the amount a user authorized for that session. Checkout state and payment processing remain on the merchant's systems, and the AI provider is never the merchant of record. On the infrastructure side, the Model Context Protocol gives agents structured, machine-readable access to inventory, prices and checkout logic instead of forcing them to scrape messy web pages, while Stripe Issuing can generate temporary virtual card numbers for an agent so real card details are never exposed.

The Payment Networks Place Their Bets

Visa entered the race early. Visa Intelligent Commerce launched on April 30, 2025 with nine founding AI and technology partners, including OpenAI. On June 10, 2026, at the Visa Payments Forum in San Francisco, the two companies announced a deeper integration: tokenized Visa credentials can now power agent-initiated checkout inside ChatGPT and OpenAI's Codex coding agent, with Visa's tokenization, agent identification and fraud-monitoring infrastructure routed into OpenAI's consumer and developer systems. Visa also unveiled three merchant tools, an Agent Score that assesses whether agents can navigate a merchant's site, an Agentic Directory of verified agents and merchants, and a Large Transaction Model fraud engine built for high-value agent purchases.

Mastercard answered within hours. Its Agent Pay framework, first announced in April 2025 with Microsoft, was extended the same day into Agent Pay for Machines, an open protocol designed for agents transacting autonomously at machine speed, including micropayments worth fractions of a cent. The clearest structural difference between the two networks is architectural: Agent Pay for Machines stores agent credentials and spending permissions on public blockchains such as Polygon, Solana and Base, while Visa relies on its own tokenization infrastructure without on-chain credential storage.

"AI will transform commerce more profoundly than the internet or mobile technology ever did," Forestell said at the June forum. "Visa's focus is to ensure transactions are trusted, secure and seamless." Visa says it has already run hundreds of controlled real-world agent transactions and predicts millions of consumers will be buying through agents by the 2026 holiday season. Analysts agree the prize is large: a joint McKinsey and ICSC report projects roughly $1 trillion in orchestrated US retail revenue by 2030 and $3 to $5 trillion globally.

The Reality Check: Trust and the Protocol Wars

For all the momentum, adoption is young, and Forrester's latest assessment is deliberately sober. US consumer adoption of OpenAI's original Instant Checkout was low and stagnant from debut to discontinuation, and OpenAI wound the initiative down in March 2026 to refocus on product discovery and ChatGPT apps, with most merchants redirecting customers to their own sites for checkout. Walmart is embedding its own shopping agent, Sparky, into ChatGPT first and Google Gemini later, and data from its pilot showed users converting at roughly 70 percent of the rate of shoppers using Walmart.com directly. Amazon, meanwhile, pushes its Buy For Me agent, which completes purchases on a customer's behalf using a stored address and card.

Trust remains the gating factor. PYMNTS Intelligence research found that while 45 percent of consumers are comfortable allowing AI agents to complete purchases on their behalf, 95 percent express at least one concern about agentic commerce, and half of US consumers would trust it more if fraud protections were clearly established. The same dynamic is playing out in the protocol layer: Google's Universal Commerce Protocol and OpenAI's Agentic Commerce Protocol are the frontrunners, backed by Gemini's 750 million monthly users and ChatGPT's 900 million weekly users, while Alibaba's Alipay has launched its own agentic payments protocol inside the Qwen app with roughly 300 million monthly users.

Analysts argue the strategic prize is not the assistant but the authorization layer underneath it. Token service providers, the firms that issue and manage the credentials, are becoming the connective tissue of agentic commerce, the place where identity, consent and payment authority converge before any agent is allowed to transact. Every autonomous transaction requires identity verification, every identity requires credentials, and every credential requires a framework for establishing trust. The companies that control that layer will not merely facilitate agentic commerce; they will govern how it operates.

What Comes Next: From Shopping Agents to Machine-Speed Commerce

The template of verified agents acting under bounded mandates will spread far beyond shopping, to booking, contracting and insurance claims. Stripe and Tempo, with Visa as a design partner, have already launched the Machine Payments Protocol for agent-to-agent payments, with an early use case of AI agents autonomously paying a per-API-call fee for web access. Forrester expects this kind of machine-to-machine commerce to take off in B2B first, where payments are highly repeatable and automation is already entrenched.

Eventually the checkout disappears into the background of daily life. When an assistant renews a household's insurance in 2029 after comparing forty quotes overnight, it will be running on rails being laid right now. Forestell framed the endgame with the question your agent will one day ask: "Do you want me to just not check?" For merchants, the practical imperative is closer. Just as search engine optimization made content legible to search engines, agentic commerce requires making a business legible to agents: machine-readable product catalogs, checkout flows that run without a human in the loop, transparent return policies and trust signals an agent can verify. Gartner predicts that 40 percent of enterprise applications will ship with task-specific AI agents built in by the end of 2026, up from less than 5 percent a year earlier.

Regulators are watching too. On August 2, 2026, the European Union's AI Act switched on continent-wide transparency rules requiring AI systems to identify themselves when they interact with people, and the European Commission gained enforcement powers over general-purpose AI models. In a world where agents negotiate with agents and occasionally with us, knowing who, or what, is on the other end of a conversation stops being a courtesy and becomes infrastructure. The checkout without you is arriving, and it is being built to be both trusted and auditable.

Sources

Recent Posts

Catgories

Archives

whatsapp me