We’ve been experimenting with something that could change how banking apps handle payments: an AI agent that doesn’t just answer questions, it acts, calling directly into PaymentsHub, the same payments engine already running behind the app you use every day.
Most banking apps are rigid by design. They show you a fixed set of screens, a fixed set of filters, and a handful of predefined ways to look at your data. An AI agent isn’t bound by that. It can pull together whatever view, filter, or analysis the user actually asks for, on the spot.
The demo below shows a concept banking app split into two parts. The classic side handles cards and payments the way you’d expect, built on PaymentsHub. The AI agent side runs an LLM connected to a set of tools through an MCP server, and those tools call directly into the same PaymentsHub APIs, so data and behavior stay consistent no matter which side of the app you’re using. Every tool call the agent makes is logged for audit and forensic purposes, and before it executes anything that changes data, like a payment, it shows the user the exact parameters and waits for approval. Access is authorized through the same OAuth mechanism as the rest of the app: one security model, not two.
In the demo, the agent handles three scenarios: reviewing recent transactions and flagging irregularities such as unpaid bills, reading a photo of a payment slip and preparing (then executing, once approved) the payment order, and turning transaction history into a real picture of personal spending.
This is an early, unbranded proof of concept, sped up for the demo. But the direction feels right to us: agents that don’t replace the existing payments stack, they sit on top of PaymentsHub, using the same rails, the same security, the same payments data.