Agentic AI has the potential to reshape the payments landscape. It refers to autonomous systems that can work together toward a goal. They learn to make decisions that spark actions and adapt to situations based on their programming and data gathering. This is different from generative AI, which pulls in content based on prompts but may not execute workflows independently. Some of the biggest payment firms have already rolled out agentic AI payment solutions. So, is this technology a significant improvement for the US payment system, or is it accelerating an already complex system of payments innovation?
Deloitte’s Tech Trends 2025 
report describes agentic AI as a new pillar of AI, arming consumers and organizations with co-pilots capable of transforming how we work and live. The benefit comes from allowing applications and interaction layers to act in concert. For payment systems, this could include gateway optimization or intelligent routing. The routing may entail choosing workflows based on predefined objectives, like cost, speed, liquidity, or fraud risk. Agentic AI can choose between rails like FedNow, real-time payments, or automated clearing house and initiate the transaction based on analysis of its objectives. For example, agentic AI can trigger recurring settlements when liquidity thresholds are met. Optimizing payment workflows reduces manual reconciliation and batch scheduling for improved settlement and liquidity management.
In October 2025, PayPal launched agentic AI commerce services that can modernize merchant platforms by allowing integrations across AI systems. One of the services mentioned is increased product discovery, which improves the checkout conversion rate. The shopping search may have been initiated through AI platforms such as generative chat or browsers, where payment can be made directly. On the consumer-facing side, agentic AI can remember prior payment histories, account behaviors, counterparties, and compliance profiles. In real time, agentic AI can maintain a behavioral profile of each customer or merchant to detect anomalies or retain the context of past disputes. These enhancements can improve the accuracy of fraud prevention techniques with adaptive risk scoring—a dynamic system that can automatically trigger actions like blocking access, requiring extra authentication, or escalating an alert.
Mastercard and Visa are optimistic that their new agentic AI offerings will quickly cultivate trust in consumers and merchants. In spring of this year, Mastercard offered Agent Pay , which is an agentic AI-based program designed to provide personalized experiences, such as programmable payments (read this blog by my colleague Chris Colson about Programmable Payments). The service also uses tokenization and passkeys for added security. Meanwhile, Visa rolled out developer tools such as Trusted Agent Protocol that lets AI agents securely connect with merchants. The solution uses cryptographic signatures to confirm authenticity of the agent and includes account and payment information for the merchant. This means AI bots can handle everything from finding products to checking out on merchant platforms, supporting the world of agentic commerce.
As exciting as the possibilities of agentic AI innovations are, the autonomous decision-making may also open the door to illicit activities by bad actors. Account takeovers are one example and are typically tied to identity fraud. Criminals may breach systems to reconfigure agent objectives such as purchases or payment information. As agentic AI rises, so do these ethical concerns. Some organizations maintaining robust risk management programs when implementing agentic AI in consideration of the potential cybersecurity aspects of their processes.
Leave a Reply