Artificial intelligence (AI) is moving closer to the point at which it can do more than analyse a treasury problem or recommend what a user should do next. The harder question for transaction banks is when a corporate client will be prepared to delegate authority to an agent to act on its behalf. That transition is particularly consequential in payments. ISO 20022 is giving banks richer, more structured transaction data at the same time that real-time and tokenised payment rails are making money increasingly programmable and always on. Standard Chartered is trying to bring those developments together without treating autonomous execution as the starting point. Sunday Domingo, Global Head, Digital Channel Solutions at Standard Chartered, said the bank is adopting AI internally in areas ranging from sales request-for-proposal preparation to its financial crime transformation programme. For clients, however, she sees the more immediate opportunity in using AI and richer payments data to solve practical treasury problems, including cash-flow forecasting, liquidity optimisation and payment and foreign-exchange efficiency. Rich data creates the foundation for useful AI Domingo said many banks initially approached ISO 20022 as a compliance obligation, but Standard Chartered has viewed the richer structured data created by the standard as a foundation for AI-driven solutions. That matters because many corporate treasurers are still gathering and consolidating information from multiple sources before they can perform tasks such as cashflow forecasting. For Domingo, the immediate value of AI in transaction banking lies in solving practical client problems before moving to autonomous execution. Better data can give AI more context with which to identify exceptions, generate insights and help treasurers make liquidity, payment and foreign-exchange decisions more efficiently. Standard Chartered has also been encouraging clients to look beyond the migration to ISO 20022 and consider how its richer structured data can be used. Domingo said the data gives banks and corporate treasurers a stronger foundation for applying AI to operational efficiency, risk management and more informed treasury decisions. Trust has to precede delegation Domingo sees the path to agentic payments as a progression. The first stage is visibility: giving corporate clients better insight into cash flow and treasury operations. The next is recommendation, such as identifying payments that require repair because information is missing or counterparty details are incorrect. Only after clients trust those recommendations does the discussion move to delegation. At that point, an agent could be given permission not merely to suggest the next action but to execute it within defined limits. "At the heart of it, we have to think about the people that use the AI," Domingo said, adding that organisations also have to invest in upskilling so employees know how to use the technology safely. Client readiness varies widely. Domingo said some companies are still at the stage of building the structured-data foundations required to use AI effectively. More sophisticated digital-native and technology companies are already developing their own agents and considering how those agents could interact with a bank's agents to execute treasury, payment and operational workflows. Standard Chartered's strategy is therefore not to impose a single level of automation. Domingo said the bank wants to meet clients where they are, while preparing for those that are already thinking seriously about agentic workflows. Existing controls can become the governance layer The difficult part of agentic payments is not simply whether an AI model can initiate an instruction. It is whether the bank and its client can establish how the agent is authorised, what it is permitted to do and how every action can subsequently be reviewed. Domingo said agentic workflows need a strong governance foundation, including delegated permissions, access rights and auditability. Standard Chartered plans to build on controls that already exist in its Straight2Bank digital platform, where corporate users and authorisers are onboarded and assigned permissions. The same foundations can be extended to agents. The bank can set up an agent's permissions and guardrails so that the financial transactions it is allowed to execute remain consistent with the client's internal policies and procedures. This approach treats an agent less as an unconstrained autonomous system and more as another authorised actor operating inside an established control framework. The distinction is important because Domingo sees technology advancing faster than governance. The ability to execute autonomously may arrive before many companies are comfortable delegating financial authority to software. Some clients are already preparing their own treasury agents The shift is not entirely theoretical. Domingo said Standard Chartered is in active conversations with clients that are building their own treasury agents and want those agents to connect to the bank, access data and use its digital solutions through Model Context Protocol (MCP), a standard that allows AI applications to connect with external data and tools. She expects those forms of connectivity to become more tangible over the coming year, potentially sooner, as clients move from building treasury agents to connecting them with bank data and digital services. Programmable money and agentic workflows are converging The longer-term proposition becomes more significant as payment infrastructure itself changes. Domingo sees programmable money and agentic workflows as two macro trends that are beginning to converge. Tokenised deposits and other emerging rails can allow money to move without conventional cut-off times. Today, a human may still trigger a transfer between tokenised accounts. Domingo expects programmability eventually to combine with agentic workflows that provide the authority, permissions and controls needed for those movements to occur with less human intervention. The convergence points to the larger question Standard Chartered is trying to solve: not whether AI can technically initiate a payment, but how a bank can let it act while preserving the control, auditability and trust that corporate treasury requires. For Domingo, the path to agentic payments therefore depends on clients becoming comfortable delegating defined actions within clear permissions and guardrails.
