Bit For Bite 11
AI Doesn’t Fail in the Models
Dear readers,
Over the past few months, the debate around Agentic AI has entered a more mature phase. We are no longer talking only about systems capable of generating content or supporting decision-making, but about agents able to reason, plan, and execute actions within the core processes of financial institutions.
This topic was recently explored by Alessandro De Leonardis, CIO of Armundia Group, in an in-depth article on the evolution of AI in the FinTech landscape. His analysis starts from a point that is both simple and too often overlooked: the real limitation to effective AI adoption is not the quality of the models, but the technological architecture on which they are deployed.
The context is clear. According to the McKinsey Global Banking Annual Review 2025, the banking sector risks up to USD 170 billion in profit erosion over the next decade if it fails to translate AI into real operational value. At the same time, early adopters of Agentic AI may achieve significant gains in profitability. The difference, however, will not be determined by who experiments first, but by who is structurally ready to bring AI into production.
As De Leonardis points out, many institutions are attempting to apply advanced AI paradigms on monolithic systems designed for a different era. Rigid architectures, long release cycles, and inflexible governance models make it difficult to orchestrate specialised agents, continuously update models, and manage different levels of autonomy. The result is evident: initiatives that remain stuck in the pilot phase.
The response, according to this perspective, requires a shift in architectural thinking. Modular, service-based architectures make it possible to integrate AI selectively, activating it on specific processes – from compliance to credit, from reporting to customer interaction – measuring its impact and scaling only what works. Deloitte already reports cost reductions between 30% and 50% in targeted workflows, while MIT Technology Review highlights the growing adoption of agentic systems in production environments.
The message is clear: being AI-ready does not mean adding a layer of intelligence on top of existing platforms. It means designing systems built to host, govern, and evolve AI over time.
In the era of Agentic AI, the question is not whether to adopt artificial intelligence.
It is whether our architecture is ready to support it.
Best regards,
Stavri Pici