Banks are increasingly using artificial intelligence (AI) to enhance their services, but many are not completely overhauling their traditional processes to fully integrate AI. Historically, banking systems have relied on a model where a person makes a financial decision, and the bank’s systems carry it out. AI has the potential to shift this by interpreting a person’s goals and taking appropriate action. According to McKinsey, generative AI could generate between $200 billion and $340 billion in annual value for the banking industry. However, many banks are simply layering AI onto their existing systems instead of rethinking their entire approach. The most obvious AI applications in banking—like improving customer service, automating fraud detection, personalizing recommendations, and speeding up credit decisions—are easier to implement. For example, Lloyds Banking Group introduced over 50 AI use cases in 2025, which generated about £50 million in value, with even more expected in 2026. However, McKinsey has warned that merely adding AI to old systems without a fundamental redesign may not lead to major transformations and could instead create new technical challenges. A key difference between a truly AI-driven system and a simple chatbot is how flexible and integrated the AI is. An AI-native system should be able to choose and coordinate actions within system limits, not just follow fixed instructions. It should also allow AI to act with human oversight and proper documentation, while incorporating built-in controls for critical areas like compliance and fraud prevention. These features ensure the AI operates within legal and ethical boundaries. A more profound shift happens when banks focus on the customer’s goals rather than the banking products themselves. For instance, an intelligent system could monitor a customer’s financial situation and suggest whether to keep money liquid, invest it, or reduce debt. This concept is already being tested in some areas, such as Deutsche Bank’s use of an agentic AI system for managing third-party risks. In this setup, AI agents gather and analyze information, then propose decisions, which are still reviewed and approved by humans. As AI becomes more capable of not just recommending but executing actions, banks must ensure these actions align with the customer’s intent and risk tolerance. Institutions remain responsible for keeping AI agents within these boundaries. Deloitte’s research highlights that integrating AI with existing systems is the biggest challenge for 77% of banking executives, surpassing concerns about security and compliance. Traditional banking systems are often fragmented, with different applications handling payments, lending, and compliance separately. AI agents need to work across these systems, combining data and actions to make decisions. To address this, banks need an orchestration layer that allows AI to access various systems without requiring custom integrations for each use case. This setup lets core platforms remain as systems of record, while more decision-making occurs above them. This approach gives newer fintech companies, like Revolut or Ramp, an edge over traditional banks that must adapt legacy systems. If regulations remain stable, the first major financial institution designed around continuous decision-making could emerge within five years.