AI agents have made code cheap, but bank modernisation still stalls on requirements. Legacy rules sit in old code and people’s heads, and an agent cannot ask what a vague requirement means. It builds its guess into both code and tests, so the error passes every check. This talk presents a business-technology integrated toolchain that turns new processes and legacy rules into one business capability model: an unambiguous, traceable source for agents to generate code and tests.

Users rarely need just one AI model. Different tasks call for different models, but every extra provider brings another integration, contract and bill to manage. Payments networks existingly had been building a shared infrastructure that many institutions connect through and had already trusted. This session explores how that approach could work for AI: a single gateway that give institutions access to a range of vetted models through one API, with flexible usage plans and consolidated billing. It also looks at a new opportunity for card issuers, offering AI credits as a cardholder benefit so tech-savvy customers earn AI tools access as they spent on their daily purchases.

Digital banking has largely been designed around customers visiting an app, comparing options and making a choice. If customers delegate those choices to agents, banks may have fewer direct opportunities to earn their attention and trust.

How will customers discover, select and remain loyal to financial providers when software mediates the relationship? Which interactions will people still want to handle themselves? This session examines how channels, service design and customer consent must evolve, and what institutions need to preserve a meaningful relationship with the person behind the agent.