The current landscape of agentic commerce is defined by a fragmented trust stack. As AI agents move from simple information retrieval to transactional execution, the industry has struggled to reconcile disparate standards like x402, Google UCP, and Mastercard AP4M. Into this architectural divide, Stripe introduced its Shared Payment Token (SPT) in October 2025, positioning the primitive as a necessary abstraction layer to standardize how agents interact with financial rails.
At its core, the SPT functions as a programmable, revocable token that prevents the exposure of underlying credentials during agent-initiated transactions. Mechanically, it operates similarly to traditional card-on-file network tokenization, where the network maps the token to the latest Funding Primary Account Number (FPAN). However, Stripe adds a layer of agent-specific metadata, including merchant scoping, time-bound constraints, and transaction caps. By integrating Stripe Radar for real-time fraud and risk signaling, the system attempts to solve the problem of runtime enforcement — ensuring that an agent’s authority is verified at the exact moment of execution rather than relying on static permissions.
Stripe is currently the only provider supporting both agentic network tokens and BNPL tokens within a single primitive. This is a notable consolidation, given that BNPL accounts for over $300 billion in global volume, and businesses on Stripe report up to a 14% revenue increase on BNPL-eligible sessions. As of March 3, 2026, the platform supports Mastercard Agent Pay, Visa Intelligent Commerce, and BNPL providers like Affirm and Klarna. By aggregating these diverse rails into a single interface, Stripe is effectively commoditizing the underlying payment method. For merchants already integrated with Stripe, this means supporting agentic transactions requires no additional development work. The payment method becomes secondary to the agent’s ability to execute, a shift that David Sykes, CCO of Klarna, frames as the definitive infrastructure for online checkout over the next decade.
Despite the technical utility, a substantial gap remains between the infrastructure being built and actual market adoption. While Mastercard projects that there will be 300 million AI agent shoppers by 2030, current data suggests a more cautious reality. Only 14% of consumers express trust in AI to execute purchases, a figure that drops significantly for transactions exceeding $50. Currently, only 3% of total transactions involve agents, even as 42% of merchants report testing the technology. The industry is betting that by automating the integration process for existing sellers, Stripe can lower the barrier to entry, but the fundamental consumer trust deficit remains an unaddressed variable.
Stripe’s strategy is not merely about facilitating payments; it is an attempt to build a governance and enforcement layer for the broader trust stack. By co-developing the open-source Agentic Commerce Protocol (ACP) with OpenAI, the company is pushing for a standardized runtime authority. This is a critical distinction: while other players focus on the rails, Stripe is focusing on the rules of engagement. The company’s $7.5 billion acquisition of OpenRouter — as detailed in our Dreamforce coverage — further signals an intent to control the routing and execution environment where these agents operate.
Stripe is attempting to unify a fractured ecosystem by acting as the middleware between agent intent and financial settlement. The SPT excels at the mechanics of runtime authority and enforcement, providing a robust framework for secure, scoped transactions. However, the long-term viability of this model as a universal standard depends on whether the industry can resolve the governance specification problem. If the consumer trust deficit persists, even the most efficient infrastructure will struggle to reach the scale projected by network providers. For now, Stripe has built the most compelling candidate for an agentic payment primitive, but the transition from a testing environment to a dominant market standard remains a work in progress.