Abstract

Incumbent banks have invested in artificial intelligence while leaving much of the work that agents must perform embedded in fragmented data, documents, informal judgement and disconnected controls. This practitioner position paper calls the institutional foundation that makes such work executable the operating layer. It defines an AI-native bank as an organisation in which an authorised agent can act on verified business facts, codified policy, controlled workflows and callable capabilities. Each action must also be evaluated, attributable and open to human intervention. The argument combines the 2026 move towards enterprise deployment vehicles, public evidence on AI returns, established risk standards and professional observation in financial-services architecture. It applies the theory of constraints to distinguish model adoption from business value, then develops a working framework for operating-layer readiness. The framework links seven architectural properties to four phases of delivery: diagnostic, codification, build and bound, and handover. Regulatory binding and executive translation are integrated into that delivery discipline. The central claim is that codification creates a reusable institutional asset. AI can reduce the cost of producing the asset, while human owners preserve warrant through approval, versioning, audit and change control. Testable propositions define an agenda for future case research. The framework is conceptual and practice-informed; causal effects across banks remain to be tested.

Keywords: artificial intelligence, AI agents, banking, financial services, operating layer, operating model, enterprise architecture, theory of constraints, forward-deployed engineering, AI governance

JEL: G21, O32, O33, M15

Suggested citation

Antikatzidis, Georgios (2026). Becoming an AI-Native Bank: A Practitioner Framework for the Operating Layer. SSRN. https://doi.org/10.2139/ssrn.7350399

Written 13 May 2026