When Your AI Agent Buys Without Asking You First
Commerce has always been a human act. A person decided to buy something, chose where to buy it, entered their payment details, and confirmed the transaction. That model is now fragmenting.
AI agents — software programmes that browse, compare options, and execute transactions on behalf of a human user — are beginning to handle the full purchase cycle autonomously. The user sets a preference (“buy me the cheapest flight under $400 with no stops”) and the agent does everything else, including completing payment. J.P. Morgan Payments describes this as agents that “anticipate needs, evaluate options, and execute transactions” on a shopper’s behalf — a shift from transactional commerce to delegated commerce.
The technology for AI agents to navigate e-commerce sites, compare products, and initiate checkouts is functional in 2026. Companies like Google Shopping, Amazon’s Rufus assistant, and a growing ecosystem of vertical shopping agents are placing real orders. The piece that hasn’t kept pace is the payment layer — specifically, the question of how payment networks handle a transaction initiated not by a human who confirms their own card details, but by a software agent acting under a delegated authority.
PaymentsDive’s 2026 analysis frames this directly: “bot payments lag in agentic commerce.” The shopping capability is ahead of the payment authorization infrastructure designed to support it safely.
The Stakes: A $1 Trillion Commerce Transition
The numbers attached to agentic commerce help explain why Visa and Mastercard are treating protocol development as a strategic priority.
Boston Consulting Group surveyed 2,532 consumers and found that approximately 81% are open to using agentic commerce tools — a level of consumer acceptance that is unusually high for an emerging technology. McKinsey estimates the US retail sales opportunity at $1 trillion by 2030, with up to $5 trillion globally. Retailers who have piloted agentic shopping integrations are reportedly achieving conversion rates 30% higher than through other sales channels.
These figures explain why the protocol question is not a technical footnote — it is a revenue and control question. The payment network that defines how agent-initiated transactions are authenticated, authorized, and settled will be embedded in every agentic purchase flow. That is an infrastructure position analogous to what Visa and Mastercard built for card-present and card-not-present transactions over the last 50 years.
Advertisement
The Protocol Landscape: Multiple Competing Standards
Unlike the early card era, where Visa and Mastercard built standards with limited competition, the agentic commerce protocol race has multiple credible participants.
PaymentsDive reports that Visa and Mastercard have each developed proprietary protocols for agentic commerce — but so have Google (promoting new technical standards for agent-based transactions), PayPal, and Walmart (which controls enough retail volume to impose its own agent payment rules on suppliers). A separate consortium led by OpenAI and Stripe is developing a competing standard.
This multi-standard environment is significant. In card-not-present transactions (online purchases), Visa and Mastercard’s 3DS (3D Secure) protocol became the de facto global standard because they controlled the issuing and acquiring relationships. In agentic commerce, they face meaningful competition from platform-level players — Google, Amazon, OpenAI — who control the agent software layer and can embed preferred payment infrastructure at the source rather than at the network level.
The distinction matters for how the standards battle resolves. If Google’s Shopping agent pre-selects Google Pay for agentic transactions, Visa and Mastercard’s protocol relevance depends on whether those Google Pay transactions route through their rails — which they currently do, but which is not guaranteed as payment infrastructure continues to evolve.
What This Means for Enterprise Commerce Teams
The agentic payment transition is not arriving uniformly across all commerce. It is arriving fastest in travel (flight booking, hotel reservation), subscription management, and replenishment commerce (reorder household goods when stock runs low). These are the use cases where agent delegation is clearest and where the spending ceiling per transaction is high enough to justify the integration investment.
1. Audit Your Checkout for Agent Compatibility Today
Enterprise e-commerce platforms that use non-standard checkout flows — pop-up authentication windows, custom CAPTCHA implementations, single-page applications with state-dependent rendering — are likely to be invisible to current-generation shopping agents. An agent that can’t complete your checkout without a human confirming a CAPTCHA doesn’t place the order elsewhere; it fails silently and the sale is lost.
The immediate action is a technical audit: can a headless browser (the technology most shopping agents use) complete a purchase from your product page to order confirmation without a human in the loop? If not, that is a conversion opportunity gap that grows as agentic shopping adoption scales. Commercetools has documented the integration requirements for agent-compatible commerce architectures, including transparent consent flows and granular payment authorization controls.
2. Understand the Fraud Liability Shift Before It Arrives
Today’s card-not-present fraud liability rules were built around a human cardholder. Chargeback dispute resolution assumes a human who either authorized or did not authorize a transaction. Agentic commerce introduces a new liability question: who is responsible when an AI agent is granted broad spending authority and makes a purchase the user later disputes?
J.P. Morgan Payments notes that “transactions remain intact within the existing authorization, authentication, and dispute frameworks” for now — which means current chargeback rules apply even to agent-initiated purchases. But as agent commerce scales, those rules will need to evolve. Merchants who understand the current framework and build clear consent logs — documented evidence of what spending authority a user granted to their agent — will be in a stronger position when disputes arise.
3. Monitor the Protocol Consolidation for Vendor Lock Risk
The multi-standard environment of 2026 will not persist permanently. Standards races in payment infrastructure tend to consolidate — 3DS won the card-not-present authentication standard, EMV chip won physical card-present. The agentic payment standard race will likely follow the same pattern: 2–3 years of competing standards, then consolidation around one or two dominant protocols.
The risk for enterprise merchants is vendor lock-in during the consolidation phase. A merchant who builds a deep integration with Visa’s agentic commerce protocol before the standards war resolves may face significant re-engineering costs if OpenAI/Stripe’s competing standard wins the agent software layer. The safer strategy in 2026 is to monitor, run pilots, and maintain flexibility — rather than making committed infrastructure investments in a single protocol.
What Comes Next
The agentic payment standards race will accelerate through 2026–2027 as the major technology platforms — Google, Amazon, Apple, Meta — embed shopping agents more deeply into their consumer surfaces. Each platform will initially favour its own payment infrastructure, but the network effects of the dominant card rails will create pressure toward Visa and Mastercard compatibility.
The resolution will likely be a hybrid: Visa and Mastercard define the authorization and settlement standard, while platform-level agents choose their preferred identity and consent management layer. PaymentsDive’s analysis suggests the distinction between shopping capability and autonomous payment will “persist in the near term” — which gives enterprise merchants a window to prepare their infrastructure before the transition accelerates.
Frequently Asked Questions
What is agentic commerce and how does it differ from normal online shopping?
Agentic commerce refers to purchases made by AI software agents acting autonomously on behalf of a human user. Instead of a person browsing a website, selecting a product, and entering payment details, an AI agent performs all these steps under delegated authority — “buy me a specific item under this price with these specifications.” The key difference is that no human reviews and confirms the transaction in real time. Current AI agents handle the browsing and product selection well; the payment authorization layer is still catching up.
Why are Visa and Mastercard developing separate agentic commerce protocols?
Each network sees agentic payment authentication as a strategic infrastructure position — equivalent to 3D Secure (3DS) for card-not-present transactions, which Visa and Mastercard developed in the early 2000s and which became the global standard for online card verification. Controlling the agentic payment standard means being embedded in every AI-initiated purchase flow. They face competition from platform players (Google, OpenAI/Stripe) who can embed payment infrastructure at the agent software layer, bypassing network-level authentication standards.
What is the current fraud liability framework for agentic purchases?
As of 2026, agent-initiated purchases use the same fraud liability and chargeback dispute rules as standard card-not-present transactions. J.P. Morgan Payments confirms that “transactions remain intact within the existing authorization, authentication, and dispute frameworks.” This means merchants and issuers use standard chargeback mechanisms for agent-initiated disputes. The liability framework will need to evolve as spending delegation authority becomes more complex — particularly distinguishing between purchases a user intended to delegate and purchases made outside the agent’s intended scope.
