May 11, 2026

On April 29 the global payment rails finished their first lap. Visa announced the global expansion of its Agentic Ready program to Asia Pacific and Latin America, adding more than 85 partners across those regions on top of the 20 already running in the United Kingdom and continental Europe. The same day, Stripe shipped its Link CLI agent wallet and OKX published the Agent Payments Protocol for autonomous agent-to-agent settlement. Three rails closed the trust loop on a single calendar date. The question for May is no longer whether agents will have somewhere to pay. It is whether anyone is ready to operate them.

The connective tissue arrived earlier this spring and was easy to miss inside the protocol-launch noise. On March 5, Mastercard and Google co-released Verifiable Intent, an open-source cryptographic standard that produces a tamper-resistant record of what a consumer authorized when an agent acts on their behalf. The specification is protocol agnostic and interoperates with Google's Agent Payments Protocol (AP2) and Universal Commerce Protocol. Fiserv, IBM, Checkout.com, Basis Theory, and Getnet signed on at launch. Pablo Fourez, Mastercard's chief digital officer, framed it tightly. "As autonomy increases, trust cannot be implied. It must be proven." Verifiable Intent answers the dispute-resolution question Stripe's Link CLI raised by hand and OKX's protocol raised on chain. A buyer agent that pays through any of these rails now leaves an authorization receipt consumers, merchants, and issuers can all read.

Rails are global. Production is unevenly distributed by sector. Cleverbridge's Radu Immenroth makes the B2B SaaS case directly. The high-friction parts of B2B procurement (RFPs, email chains, legal reviews) are exactly what agents do well, and tools like Pactum and ZipHQ already run AI-to-AI negotiation in production for enterprise procurement teams. Immenroth's framing of the second-order risk is sharper than most of the year's analysis. "As AI agents manage end-to-end purchasing, they bypass websites, creating massive disintermediation risk. With agents as the new customer entry point, traditional direct traffic, merchant visibility, and access to customer data will deteriorate." Selling to humans has been the merchant playbook for three decades. Selling to agents is the playbook for the next one, and most B2B merchants are running it before their consumer counterparts have built it.

Customer service is the other sector quietly past the production gap. Xictron's autonomous-processes brief cites Gartner's forecast that 80% of routine customer interactions will be fully handled by AI by 2026, with contact-center labor savings reaching $80 billion. The same brief surfaces a more credible production signal from Salesforce. Its agents currently run roughly 32,000 customer conversations per week with an 83% resolution rate. That is not a pilot metric. Xictron also notes that 89% of consumers prefer hybrid AI-plus-human support over either alone, which suggests this stack is converging on the right shape rather than chasing autonomy for its own sake. Returns processing, exchange suggestions, shipping anomaly detection, and tier-one escalation are now routine agent work in production environments at retail scale.

Even the marketing side of B2B has shifted. COSEOM's analysis of agentic AI in B2B marketing cites 6sense's 2025 Buyer Experience Report showing 94% of B2B buyers now use large language models during their buying journey. That number reframes the discoverability problem. If almost every B2B buyer is asking an LLM for vendor evaluations, then content strategy is no longer about ranking on Google. It is about being legibly summarized by a model that is doing the shortlisting. COSEOM identifies 11 marketing workflows already being absorbed by agents, including lead scoring, PPC bid management, analytics assembly, email personalization, and competitive monitoring. Strategy, brand direction, and cross-functional judgment remain human. Most of the daily mechanics do not.

Consumer agentic commerce is where the production gap is loudest. Amit Vadhera's LinkedIn synthesis of 2026 retail cites Gartner's projection that 20% of e-commerce transactions will be initiated by AI agents this year. Vadhera also cites McKinsey's expectation of 15 to 20% EBIT uplift from agentic retail operations and a projection that 35% of global retail AI spend by 2030 will flow into autonomous agent technologies. Walmart's super-agents are Vadhera's concrete example. They "compile themed product bundles, predict replenishments, and offer suggestions based on user behavior, weather patterns, or calendar events." The friction is on the consumer side, not the merchant side. Vadhera cites 66% of US consumers as uncomfortable with AI making purchases for them and 58% concerned about data privacy. Mastercard's Verifiable Intent and Stripe's synchronous-approval model are direct responses to those numbers. Whether they move them is a 2026 measurement question, not a 2026 architecture question.

What changed this fortnight is that the question "are the rails ready" closed. What did not change is which use cases earn their production budget. B2B procurement and SaaS purchasing are running agent-to-agent today, with Pactum and ZipHQ already in customer environments. Customer service is operating at Salesforce's 32,000-conversations-per-week scale. B2B marketing has 94% LLM penetration on the buyer side and content libraries are restructuring around that fact. Consumer retail has the rails it needs and the trust scaffolding to sit on top of them, but the conversion gap from previous weeks is still open and the consumer trust numbers are still moving slowly. The merchants who win 2026 already know which of their workflows belong to agents and which still belong to humans. The ones who do not will spend the rest of the year discovering it by accident.