As enthusiasm around agentic AI accelerates across the financial sector, the term is often presented as a decisive shift beyond traditional automation: intelligent agents that plan, decide, and act with increasing autonomy. Yet beneath this narrative lies a far more grounded reality. In an exclusive online article for the Journal of Financial Services, Udo Milkau examines agentic AI through a statistical and operational lens, questioning what these systems truly are today, and what their real impact may be for banks navigating data‑dense, risk‑sensitive environments.
Rather than exhibiting genuine autonomy, current AI agents are built on the same foundations as large language models: statistical estimators that predict the “next best token” and require rule‑based orchestration to operate within real‑world workflows. Their outputs may be fluent and convincing, but their behaviour is still governed by probabilities, error accumulation, and the quality of the data they ingest. This statistical constraint shapes both their capabilities and their limits.
The article stresses that if agentic AI were truly reshaping banking, we would already observe measurable shifts in productivity, cost‑income dynamics, or governance structures. Instead, early impacts remain incremental. Agentic AI currently supports institutions in more pragmatic ways: coordinating information, standardising routine work, and navigating complex rule sets, valuable enhancements, but far from a structural transformation.
A recurring theme across this JoFS edition is that advanced AI capabilities remain limited by data quality, governance, lineage, and observability the very foundations needed for consistent, reliable system behaviour. Without these, even sophisticated agent architectures cannot outperform established, well‑governed automation.
Milkau’s analysis offers a measured counterweight to the industry narrative, inviting leaders to rethink their expectations. Agentic AI is not yet a rule‑rewriting force but a statistically bounded tool whose capabilities are defined by data quality, risk management, and operational controls. What it delivers today is meaningful but modest. Better coordination, more consistent execution, and support for navigating complex information. These gains matter, even if they fall short of structural change. For organisations exploring agentic AI, the message is clear: progress depends on evidence, strong data foundations, and controlled deployment rather than assumptions of autonomy.
For deeper insights and the full analysis, read the online‑exclusive article in the Journal of Financial Services launching 25 March 2026. You can pre-register to receive the online edition first via the button below.
The Projective Group Institute’s Journal of Financial Services (JoFS) provides structured insights on developments in the European financial sector. Each edition brings together contributions from practitioners, academics and regulatory experts to help readers understand key changes in the industry.
This edition examines how data and Artificial Intelligence are influencing financial services. It looks at how modern analytics and evolving regulations such as GDPR, DORA and the EU AI Act are raising expectations for data quality, governance and oversight. Through concise, thoughtful articles, the edition highlights the practical implications for decision‑making, risk management and operational resilience.
