Some of the largest technology and payments companies are building the plumbing for a future in which AI agents, not people, initiate and complete purchases. Coinbase, Stripe, Google and Visa have each put forward their own protocol for what the industry now calls agentic commerce, and the competition between them is shaping up as a contest over who controls the payment rails of a machine-driven economy.
Four approaches
Coinbase’s x402 protocol revives a long-unused HTTP status code, 402 “Payment Required,” to let software agents pay for web resources and services with stablecoins. Transactions settle on Coinbase’s Base blockchain, and the company pitches the design as a way for agents to make very small payments without accounts or card checkouts.
Stripe’s Machine Payments Protocol takes a related approach from the side of a traditional payments processor. The framework is meant to let a single HTTP flow accept cards, stablecoins and Lightning Network payments, so developers can expose paid APIs and services to agents without building separate billing systems for each method.
Google’s Agent Payments Protocol, or AP2, was announced in September 2025 with more than 60 partners, including Mastercard, PayPal, American Express and Coinbase itself. AP2 is built around signed “mandates,” digital records of what a user authorized an agent to buy, so merchants and banks can verify that a purchase reflects the person’s intent. It is designed to work alongside Google’s Agent2Agent protocol and the Model Context Protocol.
Visa’s entry, Intelligent Commerce Connect, extends the card network’s tokenization and authorization systems to agent-initiated transactions. The service has been in pilot with selected partners, and reports suggest Visa is preparing a broader rollout later in 2026.
What is at stake
The companies are solving the same underlying problem from different starting points. Card networks and processors want agents to use existing accounts and fraud controls. Crypto-native firms argue that stablecoins and open protocols are better suited to high volumes of tiny, automated payments between machines.
The outcome matters because whoever sets the dominant standard will influence how subscriptions are managed, how merchants verify that an agent is acting on a real customer’s instructions, and who bears liability when an automated purchase goes wrong. For now, the protocols are largely complementary and several companies participate in more than one, but the positioning suggests that the next phase of competition in AI may be less about building a better chatbot than about owning the checkout.
