The first serious framing of agents in B2B payments
Sunrate and Mastercard have released a joint white paper, "Beyond Automation: Defining Agentic Global Payments," at the 2026 World Artificial Intelligence Conference in Shanghai. It is among the first payments-industry reports to examine how agentic AI reshapes B2B cross-border payments specifically, a corner of the stack that usually trails consumer payments in attention. The timing matters. Agent frameworks are proliferating across every enterprise function, and finance leaders need a reference model for what autonomy means when the workflow in question moves real corporate money across borders.
The paper's core claim is that payments are moving past the digitization and automation eras into a new autonomy stage. In this stage, AI agents reason, plan, and execute end-to-end payment and treasury workflows within governance frameworks, rather than simply executing pre-scripted rules faster. That distinction is the whole argument. Automation follows a fixed path a human designed; autonomy means the agent decides the path within defined guardrails. For treasury teams, that shift changes the control question from validating outputs to constraining behavior.
Sixteen pain points, thirteen use cases, one map
The report maps 16 major pain points across the B2B payment lifecycle and pairs them with 13 high-value agent use cases. That structure is the most immediately useful part for a finance or treasury leader. Instead of an abstract argument about AI, it offers an inventory: here are the recurring frictions in cross-border B2B payments, and here are the specific points where an agent could plausibly earn its keep. It turns a broad technology question into a prioritization exercise you can actually run against your own payment operations.
The lifecycle framing also disciplines the hype. Cross-border B2B payments carry real friction: FX exposure, compliance screening, counterparty onboarding, reconciliation, and settlement delays that tie up working capital. A map that names 16 of these and attaches candidate use cases lets a team ask which frictions are worth automating first and which need a human in the loop for the foreseeable future. That is the kind of concrete triage that separates a pilot with a business case from a science project.
Five agents, decomposed by role
The paper references Sunrate.AI's five-agent portfolio as a working illustration of how the workflow decomposes: a Payment Agent, an FX Agent, a Compliance Agent, an Onboarding Agent, and a Chat Agent. The design choice worth noting is the decomposition itself. Rather than a single monolithic system trusted to do everything, the workflow splits into agents with narrow, nameable responsibilities. That structure makes each agent's authority easier to scope, its actions easier to audit, and its failures easier to contain to a bounded part of the pipeline.
For anyone evaluating agentic systems, this role-based split is a pattern to borrow regardless of vendor. An agent with a single clear job, a Compliance Agent that screens counterparties or an FX Agent that manages currency exposure, is far easier to govern than a general-purpose assistant given broad access to treasury functions. The decomposition maps naturally onto existing separation-of-duties controls that finance organizations already understand. It lets you reason about permissions per agent instead of per system, which is the granularity that matters when money is at stake.
Governance is the load-bearing idea
The most important thread in the paper is its governance-first emphasis. It argues that autonomous payment decisions require auditable identity, intent, and action. Anouska Ladds, EVP Commercial and New Payment Flows, Asia Pacific at Mastercard, framed it directly: "As AI starts to act on behalf of businesses, autonomous payment decisions need a clear, auditable chain of identity, intent and action." That triad is the practical spec. You must be able to prove which agent acted, what it was trying to accomplish, and exactly what it did, for every transaction it touched.
This is where the paper earns attention from a CxO audience. The barrier to putting agents in production is rarely raw capability; it is the inability to demonstrate control to a risk committee, an auditor, or a regulator. An identity, intent, and action chain is the audit trail that makes an autonomous payment defensible after the fact. Any organization considering agents anywhere near its treasury should treat this triad as a minimum requirement and ask every vendor how they capture and retain it.
Where this sits against the rest of the market
This framework is distinct from the agentic-commerce sandboxes aimed at consumer checkout, such as Mastercard's Proto work. Those efforts target how a shopper's agent buys goods. The Sunrate collaboration points at the finance and treasury stack instead, framing what identity, intent, and audit controls must wrap any agent authorized to move corporate money. The two problems share vocabulary but differ in stakes. A consumer agent overspending is a customer-service issue; a treasury agent mispaying a counterparty is a material financial and compliance event.
That separation is useful for leaders trying to place these announcements. Agentic payments is not one market but several, segmented by who the agent acts for and how much authority it holds. B2B treasury sits at the high-consequence end, where governance requirements are strictest and the tolerance for error is lowest. A white paper written specifically for that segment, rather than repurposed consumer thinking, is a better starting reference for anyone whose agents would touch enterprise cash flows.
What finance and technology leaders should do now
Paul Meng, Sunrate's co-founder and CEO, argued that "AI agents will fundamentally reshape how enterprises manage global payments, enabling smoother capital flows, reducing operational friction, and embedding real-time intelligence into every payment decision." That is a vendor's optimistic read, and the honest response is to treat it as a hypothesis to test. Take the paper's inventory of pain points and use cases and score them against your own payment operations. Identify one high-friction, low-blast-radius workflow where an agent could run under tight supervision.
Then build the governance spine before the agent, not after. Decide how you will capture identity, intent, and action for every autonomous decision, and confirm your payment and treasury vendors can supply that audit chain. The organizations that will move agents from pilot to production in treasury are the ones that solve auditability first and capability second. For a CIO or CFO, the takeaway is to start the governance design now, so that when the capability is ready, the controls are already in place to trust it with real money.
