Building AI is no longer a challenge in itself for businesses today. The real challenge is to make it last over time on information systems that are not designed to absorb it.
Companies are accelerating on artificial intelligence. Use cases are multiplying, tools are becoming more widely available, and the promise of rapid gains is well established. However, on the ground, another reality imposes itself. In most organizations, the information system is not structured to absorb this new complexity.
The discrepancy is rarely visible at first glance. The system works, supports operations, generates revenue. But as soon as it comes to going further, connecting data, industrializing uses or evolving what already exists, limits appear. Dependencies are revealed, flows become difficult to control, and each development turns into a sensitive project.
This observation changes the nature of the subject. The challenge is no longer just to create AI. It is to know whether the information system is able to support it over time.
The illusion of a technological subject
Faced with the rise of artificial intelligence, the reflex is often the same. We look for the right tools, the right models, the right use cases. Attention is focused on the ability to quickly produce value, to test, to deploy. AI is approached as a new brick to be integrated into an existing system supposedly capable of absorbing it.
This approach allows you to move forward quickly. It makes the first experiments possible, opens up concrete perspectives, and gives the feeling of transformation. But it is based on a rarely questioned hypothesis. That of an information system sufficiently controlled to support these new uses over time.
Because very quickly, the stakes shift. Moving from a prototype to reliable use, connected to business processes, requires handling consistent data, orchestrating flows, exposing services in a stable manner. So many prerequisites which directly depend on the structure of the existing system.
Added to this is a point that I see systematically underestimated. AI introduces new, sometimes significant, costs. Model consumption, data processing, infrastructure. Without detailed control of the system that supports them, these costs quickly become difficult to anticipate, and even more difficult to rationalize.
What then appears is not a limit of the tools. This is a boundary of the base. And it directly conditions the ability to make AI something other than a succession of isolated initiatives.
A system that we think we have mastered… but that we misunderstand
In many organizations, the information system gives an impression of stability. The tools are in place, the processes are running, and the activity is well supported. This mastery is based above all on daily use, more than on a real understanding of the whole.
In reality, this foundation is often much more fragile than it seems. Many systems were built over several years, sometimes several decades, by successive stacking of solutions. A central block concentrates the critical flows, around which business tools, marketing platforms, data layers and specific interfaces revolve.
Data circulates from one system to another, is duplicated and transformed. Mapping logic compensates for gaps between tools, while intermediate files or manual processing take over when the systems are no longer sufficient. Behind structured interfaces, part of the operation still relies on extractions, CSV files or Excel processing.
Each evolution then involves dealing with multiple dependencies, flows that are difficult to read, and bricks at the end of their life or that are not very scalable. What seemed controlled on a daily basis becomes a point of tension when it comes to changing the system as a whole.
In this context, wanting to integrate new uses, particularly around AI, often amounts to adding complexity to a system that already is. Without a clear vision of the information system, initiatives remain isolated. They are difficult to make reliable, and even more difficult to industrialize.
Urbanize to regain control
Faced with this complexity, the most common reaction is to act quickly. Replace an aging brick, add a new tool, launch a redesign. On paper, these decisions seem logical. In fact, they do not resolve the basic problem. They add to a system that is already difficult to read, and sometimes reinforce existing dependencies. This is a pattern that we often find: each decision taken without a global vision shifts the problem more than it addresses it.
Regaining control requires a change in posture. It is no longer a question of stacking solutions, but of restoring coherence to the whole. This begins with a precise reading of what exists. Understand real flows, identify dependencies, clarify the role of each brick. A technical map then becomes a central tool, not for documenting, but for making the system readable.
From this foundation, a target architecture vision can emerge. A projection of the system which takes into account business constraints, data issues and development needs. A form of IS land use plan, which makes it possible to structure responsibilities, limit redundancies and better organize interactions between components.
In fact, this transformation does not occur in rupture. It takes place over time, through a progressive transformation plan towards this target. This often involves encapsulating existing building blocks, setting up decoupling layers, stabilizing exchanges via controlled interfaces. Gradually, the system becomes more modular, more observable, more controllable.
This type of approach relies less on tool choices than on the ability to understand the system as a whole and prioritize developments.
Urbanizing ultimately means restoring coherence to a system which has gradually lost it. And without this consistency, transformations, including those linked to artificial intelligence, remain fragile by construction.
The real challenge is not to create AI, but to make it stick
Artificial intelligence will not redistribute the cards solely on the ability to innovate. Above all, it will reveal the gaps between organizations.
Some will be able to connect their systems, make their data reliable and sustain their uses over time. Others will remain stuck at the experimental stage, held back by a system that is too opaque to really be managed.
This is already what we are observing. The subject is therefore no longer just about creating AI. It is to know whether the information system is able to support it.
This shift puts back at the center subjects that we thought we had mastered. Understanding of the IS, consistency of architecture, mastery of flows and data. Not as technical issues, but as conditions of transformation.
For CIOs, the issue is clear. It is no longer just a question of supporting innovation, but of structuring a system capable of evolving over time, without creating new dependencies.
It is on this capacity that the success of future transformations will largely depend.
