Agentic AI is establishing itself in France and moving AI from generation to autonomous action, requiring the CDO to guarantee a solid trust framework. A strategic shift for businesses.
For eighteen months, discussions on artificial intelligence (AI) have mainly focused on its production capabilities: generating images, writing memos or even synthesizing data. Now, in French companies, a new stage is beginning and the use of AI is evolving: we are moving from simple content generation to putting it into action. The priority is not just the proposal, but the execution.
We are witnessing the emergence of agentic AI: systems that are no longer limited to generating content, but operate autonomously in order to achieve defined objectives. These agents can negotiate rates with suppliers, reorganize a supply chain in the event of weather-related disruptions, or even adjust a marketing campaign in real time based on sales data. For the Chief Data Officer (CDO), this evolution does not represent a simple extension of his responsibilities, but a profound and structuring transformation of his role.
The speed of this transformation in France is particularly remarkable: 55% of French companies have already deployed one form or another of agentic AI, and 16% plan to deploy it in the next six months*. This means that within a year, almost three-quarters of the organizations around you could rely on autonomous systems capable of making operational decisions. The question is no longer whether the CDO should be interested in this technology, but rather how quickly he can adapt to it. Indeed, agentic AI does not just use data; it exploits them, acts accordingly and depends on them to function effectively.
This new reality profoundly redefines the role of the CDO. He is no longer just the guardian of static assets, the one who ensures the cleanliness and compliance of data within the data warehouse. He now becomes the architect of a more dynamic organization, capable of integrating operational autonomy. At the heart of this transformation is what we call a “trust context”: a framework where autonomous agents can intervene in complete security, thanks to a precise definition of the limits, authorizations and meaning of the data. In this controlled environment, systems can act autonomously without constantly requiring human validation.
From the role of “data manager” to that of “action coordinator”
Traditionally, the performance of a CDO was evaluated through indicators focused on data quality, governance and compliance. The essential questions related to their accuracy, their protection, their traceability or even the ability to demonstrate to the regulatory authorities their origin and the access that had been made to it. These issues, although fundamental, were mainly retrospective: they aimed above all to secure and structure information assets.
The rise of agentic AI is profoundly transforming this approach. Data are no longer simple resources consulted occasionally; they become data mobilized continuously for decisions taken in milliseconds, without human intervention. An inventory management agent no longer waits for reporting: he analyzes sales, checks supplier deadlines and places orders. A marketing agent no longer depends on a brief: he adjusts campaign budgets based on observed performance, sometimes from one day to the next.
In this context, having “clean” data is no longer enough. They must be enriched, contextualized and immediately interpretable by systems. An agent must not only recognize a customer ID, but also understand their segment, their payment history or the pricing rules that apply to them. Data must be explicit, meaningful and ready to be activated.
This change marks a major transition. The CDO becomes the architect of the flow of information between agents, the coordinator of a distributed digital workforce. It must ensure that marketing, logistics and financial agents rely on a common definition of key concepts, whether it is a “customer”, a “pending order” or a “profitable transaction”.
This development also calls for a technical transformation. The era of massive data lakes, where we dump raw data and wait for humans to extract value, is coming to an end. CDOs now build real “inter-agent knowledge graphs”: structures that represent not only entities, but also their relationships, their rules of use and the operational intention that connects them. It is no longer the equivalent of a dictionary, it is the transmission of a context, of a native understanding of the business environment.
The success of this system is based on a foundation of trust: to allow agents to act autonomously and compliantly, the origin of the data must be verifiable, their meaning unambiguous and their access rights strictly regulated and enforced. When this context is firmly established, human intervention is no longer required for every micro-decision. Agents operate within their authorized perimeters, in complete safety and in full compliance.
Avoiding the pitfalls of autonomy
Entrusting operational responsibilities to autonomous agents naturally carries significantly greater risks than those associated with previous generations of AI. A hallucination produced by a chatbot can certainly be embarrassing, but an erroneous financial transaction initiated by an autonomous agent can cause a real crisis. This difference in impact illustrates the significant increase in the level of vigilance required regarding these new systems.
While 57% of French companies still consider data quality and recovery to be major challenges*, deploying agentic AI on fragile foundations does nothing to resolve these difficulties: it simply amounts to automating existing dysfunctions. An agent that makes decisions based on incorrect data does not improve performance; it amplifies errors with the speed specific to automated systems. The “context of trust” must therefore be based on data that is not only clean, but truly irreproachable.
The security scope is also evolving significantly: according to the Digital Trust Digest, 69% of cybersecurity professionals now consider the vulnerabilities inherent in AI agents to pose a more serious threat than their potentially malicious use by human actors. The risk model has transformed: it is no longer just a question of preventing the exploitation of a tool by a malicious third party, but of controlling the behavior of the tool itself, which is likely to become dangerous due to an error, a misconfiguration or an unforeseen interaction with another agent. The risk is therefore no longer limited to external attacks, it now includes internal failures that could be confused with an intrusion.
There is also the question of dependence: 60% of French companies rely, or plan to rely, on external service providers for their AI* agents. This raises a major governance issue. As soon as the data leaves the internal trust context to join the ecosystem of a third party, who assumes responsibility? How can you ensure that an agent operated by a provider located in another jurisdiction complies with your internal policies and French regulatory requirements? The CDO’s role now extends to managing and overseeing external data, necessary to interact securely and compliantly with partners and suppliers.
The modern CDO finds itself at a complex crossroads. On the one hand, the need for innovation, speed and efficiency that autonomous agents can offer. On the other, the need for trust, control and certainty regarding the relevance of the organization’s decisions and the respect of its obligations. The CDO is no longer just the guarantor of data security, he is also the facilitator of autonomy. It must create the conditions that allow agents to act freely, not by relaxing controls, but by integrating them so intrinsically with the data itself that freedom and compliance become one.
Succeeding in this transition requires governance integrated into the very heart of the data, impeccable quality and a technical framework where each relationship and each usage right are explicitly defined. The organizations that will benefit from agentic AI are those capable of delegating action without delegating responsibility. This delicate balance now becomes a strategic pillar of the company’s performance and transformation.
*CDO Insights Report, Informatica, 2026
