Agentic AI vs. Generative AI – Practical Impact for Banking Processes

Spotlight | 5. March 2026

As banks invest heavily in AI, the focus is shifting from experimentation to execution. Philippe Santraine outlines how agentic and generative AI can enhance operational efficiency while meeting regulatory demands.

Artificial intelligence in banking is often discussed in ambitious, future-oriented terms. In practice, the key question is much more concrete: how does AI actually improve everyday banking processes?

According to our Head of AI Product Management, Philippe Santraine, the discussion needs to move away from hype and toward measurable value.

“This is not about replacing systems or introducing experimental technology,” he says. “It’s about improving banking processes in a controlled and reliable way.”

Understanding the difference between Generative AI, AI Agents, and Agentic AI is central to that shift.

From Consumer AI to Enterprise AI

Consumer-facing AI tools are designed to answer questions, generate text, and support creative tasks. In banking, however, AI operates in a very different environment.

Enterprise AI is embedded in regulated processes and supports daily operations such as document management, incident investigation, application processing, and regulatory interpretation. These systems work within existing workflows and under strict governance.

“This is enterprise AI,” Philippe explains. “It is structured, supervised, and designed for real operational use.”

The objective is not experimentation but reliability.

Several persistent myths continue to shape how AI is perceived in financial institutions. One is the belief that AI requires constant real-time data. In reality, many effective use cases rely on batch-based information flows, particularly in compliance and reporting.

Another common assumption is that legacy systems, especially mainframes, are incompatible with modern AI. Philippe challenges this view.

“The mainframe combined with cloud-based AI capabilities creates a very strong foundation for secure and scalable operations,” he notes.

Concerns about hallucinations and loss of control are also frequently raised. While poorly configured systems can produce unreliable results, properly customized enterprise AI can achieve high accuracy, maintain traceability, and operate within defined limits.

“Autonomy does not mean lack of oversight,” Philippe emphasizes. “Human professionals remain in control.”

From Strategy to Trusted Enterprise AI

Philippe argues that strong AI strategies rarely begin with technology. Instead, they begin with business objectives.

Banks pursue AI to improve productivity, reduce risk, support regulatory compliance, strengthen existing automation, and focus on measurable outcomes. AI is therefore a means, not an end.

“This is not a competition between technologies,” Philippe says. “It’s about combining automation and AI in a way that supports business goals.”

Contrary to popular belief, successful AI adoption is not led primarily by data scientists. In banking, the most important drivers are often compliance specialists, operations managers, and business leaders. They understand how systems function in practice and where improvements are genuinely needed.

These professionals define requirements, validate outputs, and ensure accountability. Their involvement ensures that AI remains aligned with regulatory and operational realities.

“I can write, therefore I can prompt,” Philippe summarizes, highlighting how domain expertise has become central to effective AI use.

Turning AI Capabilities into Business Value

While Generative AI, AI Agents, and Agentic AI are often discussed together, they serve different roles. Generative AI supports users by producing content and summaries. AI Agents perform predefined tasks under supervision. Agentic AI goes further by learning, adapting, and maintaining context, while remaining within governance frameworks.

Industry research shows that many AI pilots fail to deliver measurable returns. Only a small share reaches full-scale production, and internally built solutions struggle more often than those developed with experienced partners. The strongest results are typically achieved in structured back-office processes, where governance and repeatability already exist.

Philippe sees this as confirmation of what banks experience in practice: narrow, well-defined use cases supported by strong partnerships are far more likely to succeed than broad, experimental programs.

At Samlink, enterprise AI is developed as part of long-term banking infrastructure, not as a standalone innovation project. The focus is on integrating AI into regulated processes, working closely with business owners, prioritizing accuracy, and leveraging Kyndryl’s global expertise alongside decades of core banking experience.

“In the end, enterprise AI must be trusted AI,” Philippe says. “That is what creates real value for banks.”

Rather than pursuing rapid transformation, the focus remains on sustainable progress: improving productivity, strengthening resilience, and ensuring that technology supports business objectives over the long term.