What is agentic commerce — and why the number is staggering
In 2026, artificial intelligence agents conduct approximately 620 million commercial transactions per year on behalf of human users — purchasing tickets, renewing subscriptions, restocking corporate inventories. That number seems large until placed beside the projection Juniper Research published on June 29, 2026: 120 billion annual transactions by 2031. Growth of more than 19,000% in five years. Agentic commerce has moved beyond the laboratory. It has become one of the highest-stakes bets in the digital economy.
Definition and market scale
The Juniper Research report "Agentic Commerce Market 2026-2031" defines agentic commerce as any transaction in which an AI agent makes purchasing decisions and executes payments autonomously, without human intervention at the moment of transaction. The distinction is fundamental: this is not an assistant that suggests products — it is a system that researches, compares, decides, and pays. The user defines the boundaries and the agent operates within them. Juniper estimates the number of global users of this modality will jump from fewer than 300 million in 2026 to 1.3 billion by 2031 — 350% growth over the period. Total value transacted is expected to reach 3.5 trillion dollars over that horizon, rising from just 8 billion in 2026.
B2B: the segment with greatest acceleration potential
The B2B segment is identified by the report as the one with the greatest acceleration potential — and the reason is structural. Corporate purchasing already flows through highly standardized platforms: travel management systems, procure-to-pay solutions, ERPs with closed catalogs. An agent integrated into these platforms does not need to navigate millions of online stores — it only needs to master a finite number of corporate APIs. Juniper identifies this model as the path of least resistance to massive B2B scale. The caveat is equally clear: for companies to maintain trust in agents, each purchasing decision must be auditable and explainable. Systems that operate as black boxes will not survive compliance department scrutiny.
B2C: the obstacle is psychological, not technical
In B2C, the central obstacle is not technical — it is psychological. A PYMNTS survey published in 2026 found that 95% of consumers express at least one concern about purchases made by AI on their behalf. Privacy, control, and reversibility of transactions top the list of concerns. Juniper acknowledges this data in the report and argues that trust will grow as AI becomes a normalized part of daily life — the same cycle observed with contactless payments and digital wallets in the previous decade. Major retailers such as Etsy and URBN already accept agent transactions using Stripe's Shared Payment Tokens infrastructure, signaling that the merchant side is moving faster than the consumer side.
The war for the rails: Mastercard, Visa, and Stripe
The infrastructure that will sustain this volume sits at the center of an unprecedented strategic dispute. Mastercard launched Agent Pay in April 2025 — a framework based on "Agentic Tokens," an extension of the same tokenization stack that powers contactless payments. The token is bound to a specific agent, a merchant scope, and a user consent policy: the model cannot spend beyond what the account holder authorized. Visa responded with Intelligent Commerce, built on equivalent logic. In March 2026, Mastercard and Santander completed the first end-to-end payment executed by an AI agent in Europe using real banking infrastructure and Agent Pay in a controlled environment. Stripe positioned itself as the orchestration layer: in March 2026, it announced simultaneous support for Mastercard Agent Pay, Visa Intelligent Commerce, and BNPL methods such as Affirm and Klarna — becoming the first processor to unify agentic network tokens and BNPL tokens in a single integration.
The second front: stablecoins and Coinbase
There is a second front that headlines about shopping assistants frequently overlook. Coinbase and the stablecoin ecosystem represent an alternative bet: if AI agents continuously optimize costs, the 2-3% interchange fees charged by Visa and Mastercard on every transaction become an obvious target for elimination. A Citrini Research report widely cited in February 2026 modeled the scenario in which agents progressively migrate to stablecoin rails where transactions cost fractions of a cent. Mastercard responded by acquiring stablecoin infrastructure company BVNK for up to 1.8 billion dollars. Stripe had acquired Bridge, specialized in the same segment, for 1.1 billion dollars in 2025. The message of both moves is identical: card networks believe AI agents are about to start spending serious money soon — and they want to build the rails before someone else does.
What is at stake and who will control the rails
Morgan Stanley projects that agentic commerce could capture 385 billion dollars of U.S. e-commerce by 2030 — with only 1% of shoppers using the agentic modality today. The country readiness index published by Juniper covers 61 nations and points to significant differences in regulatory maturity and infrastructure. The battle for protocol standards — between Google's Universal Commerce Protocol, OpenAI's Agentic Commerce Protocol, and card network initiatives — has no defined winner yet. Nick Maynard, vice president of research at Juniper, was direct: agentic commerce is all about early mover advantage, and early participation in frameworks will be vital for success. The figure of 120 billion transactions by 2031 is a projection, not a certainty — but the movement of capital is already real: billions in strategic acquisitions, four major frameworks launched in less than twelve months, and the world's largest retailers accepting agent transactions today. Whoever controls the rails will control a growing share of how money moves in the world.
