Post by Edmond KAMENI

When we speak about “#agentic AI,” we should first be clear about what we mean. I do not see it as a mere extension of information systems, but as a shift in paradigm: from instrumental intelligence to distributed intelligence—embodied in autonomous, situated agents capable of initiative, coordination, and learning at the very core of work. From that perspective, the analysis proposed by Gildas Bouteiller and ☁️Jean-Christophe LAISSY☁️ rightly identifies that the bottleneck is organizational. Yet the method, in my view, remains fundamentally misaligned. It still relies on a legacy transformation mindset: budget allocation, change management, scaling programs. But agentic AI does not “scale.” It reconfigures. It does not transform organizations top-down. It reshapes them through a fine-grained redefinition of every decision point. The only viable method is therefore radically different. It starts by turning every role into an augmented role, equipping each individual—regardless of hierarchy—with the AI agent best suited to their constraints, decisions, and operational logic. You do not transform the organization first. You transform the decision layer. Only then does a second step become meaningful: building a global AI agent, not designed upfront, but emerging from the coherent integration of all these local intelligences—an approach that resonates with the challenges addressed by Vincent Delamarre and Jean-Charles VERDIER in their respective environments. This global agent is not another tool. It is the emergent form of the company’s collective intelligence. Third, the system must become self-adaptive: every role change, every new hire should trigger the creation or adjustment of a dedicated agent, with a corresponding update of the global agent. You no longer manage headcount. You orchestrate a living cognitive system. Finally, this requires a new kind of AI function—not purely technical, but strategic—responsible for regulating the interplay between local agents and the global agent, the latter clearly positioned at the executive level. In complex projects, as in large-scale transformations, one principle consistently holds: it is not the most ambitious architectures that succeed, but those that make decision-making possible at every level while maintaining overall coherence. Agentic AI is no exception. It does not deploy. It embeds, connects, and then self-organizes. The real question is no longer: “How do we scale it?” But: Are we ready to transform every decision point within the organization? I would be particularly interested in your perspective on this “role-first” approach versus traditional program-driven transformation models.