July 13, 2026

Visa built transaction controls last week. This week the market started building the systems that connect them. NPCI is reportedly developing a Unified Agent Protocol to register, verify, authorize, and log trusted agents above UPI without replacing the payment rails underneath. Swift launched a programmable shared ledger with 17 global banks that connects banks' in-house tokenized payment systems, with agentic commerce among its future applications. The proposals operate at different scales, coordinating trusted agents across a national payment network and tokenized money systems across global banks. Last week's trust stack already had plenty of primitives. Taken together, these efforts expose the next problem. No single network sees the complete agentic transaction. Bank of America's analysis cites Gartner's estimate that agentic commerce could account for roughly 20% of global ecommerce transactions by 2030. That projected volume turns interoperability from a technical detail into a systemic requirement.

That coordination has to span competing payment architectures. Mastercard said it completed Moldova's first agentic payment pilot with maib and Moldindconbank, using registered agents, verified intent, and payment tokens constrained by consumer permissions. Alipay says its Xiaoyu merchant agent is now available across a Tap network comprising 30 million devices, where it supports business analysis, marketing, membership management, and integration with Taobao Instant Commerce. Moldova's banks inherit Mastercard's trust framework. Alipay joins its own protocol, payment system, merchant devices, logistics, and outside software agents. Neither implementation is becoming a universal standard, but both add endpoints that must agree on what identity, permission, and intent mean.

An agent credential does not necessarily validate the business behind it. Proof and Enigma announced Business Certificates, which they say will combine maintained business records with credentials showing which people and systems may act for a company. Cequence introduced two different controls for the traffic those systems generate. Its Intent Graph classifies automation as legitimate or adversarial, while Biometric Check requires human approval on a registered device before high-stakes actions. Together, these systems validate represented businesses, classify behavior, and require human approval for high-stakes actions, but their sources do not describe how they exchange what each verified.

Authorization covers the happy path. Recourse begins when execution departs from the agreement. Internet Court packages contracts, payment, escrow, verification, and dispute resolution into an open agent skill, with parties selecting a dispute provider before they transact. Its published materials describe contractual and escrow-based enforcement but do not establish whether outcomes have independent legal enforceability. Pinsent Masons mapped the public-law questions that remain, from checking both the original mandate and the resulting transaction to reimbursement, reversal rights, and possible developer or merchant liability. Those are legal forecasts rather than proposed rules. They expose the missing bridge between a private protocol enforcing its terms and a public system deciding who owes the consumer when those terms fail.

Public institutions started designing coordination above the private stack. The UK Financial Conduct Authority's Mills Review, informed by commissioned research involving more than 5,000 financial-services consumers, found that one in five UK adults is open to AI making decisions for them. Its seven recommendations on AI's impact on retail finance include adapting the regulatory perimeter, strengthening system-wide oversight, monitoring autonomous models, and asking the FCA Board to consider an AI-enabled agentic supervisory model. A separate Global Agentic Regulator Hackathon, supported by the BIS Innovation Hub, GFIN, DRCF, and other organizations, is asking teams to prototype agentic-payment oversight, Know Your Agent, AI-fraud, and market-herding controls. Although neither initiative is an operating rulebook, both move regulation from warning about autonomous markets to designing ways to observe them. Supervision has to reconstruct transactions assembled across systems that report to different owners.

Commercial platforms are betting on specialization before the coordination problem is solved. Bank of America reinstated coverage of Shopify with a Buy rating on the thesis that value moves toward catalog, checkout, payment, and transaction infrastructure as discovery shifts to agents. Meta Chief Data Officer Alex Schultz called agentic commerce a potential next tier of business. CoinDesk framed Meta as the conversational interface with third-party stablecoins beneath it, while Schultz said stablecoins are a major part of the solution and that Meta would prefer to plug into an external identity service. More than one million businesses use Meta agents weekly, the company says, though it has not disclosed their activity or transaction volume. Shopify does not need to own discovery, and Meta does not need to own settlement. Each needs the surrounding networks to accept what the others authorize.

Last week asked whether private trust infrastructure could absorb systemic risk. This week supplied the beginning of an answer, and it is not a larger gate. An agent may arrive with a payment token from one network, a business certificate from another, settlement through a third, and a dispute process none of them governs. Every component can work exactly as designed while the complete transaction fails at the handoff. Coordination is now a product, a protocol, and a regulatory task at once. The next failure may not happen inside any rail. It may happen between two systems that each believe the other one checked.