Agentic AI is attracting growing attention, but scaling it across organisations remains difficult. For Pierre-Olivier Patin, the issue is not primarily one of models or use cases. “AI agents are mostly a topic of platformisation,” he argued, calling on companies to create common foundations rather than multiplying isolated initiatives.
The message comes at a time when, according to figures presented by the Sogeti executive, 85% of AI initiatives fail to deliver the expected outcomes. His recommendation is to focus on a model that can “unify solutions and tools” and provide reusable capabilities across the organisation. Without that layer of coordination, businesses risk creating a patchwork of disconnected AI projects that are difficult to govern and scale.
If there is no orchestrator, you are the orchestrator as a human.
The VP Global CTO also described a profound shift in the way users interact with technology. Today, employees often move between multiple applications and interfaces to complete a task. In the emerging agentic environment, however, “if there is no orchestrator, you are the orchestrator as a human.” The objective is therefore to create systems in which agents coordinate tasks on behalf of users.
This change goes beyond interface design. “It’s mostly a topic of intent,” he said. Rather than navigating long workflows and sequences of clicks, users increasingly focus on the outcome they want to achieve while agents handle the underlying processes.
Context, sovereignty and trust
For the Sogeti executive, context is becoming the critical ingredient of the next phase of AI adoption. Organisations are not starting from scratch: they already operate data platforms, integration layers, SaaS applications and systems of record. The challenge is connecting those assets into a coherent foundation capable of supporting agents.
“We are in a tech convergence era,” Pierre-Olivier Patin said, noting that companies are simultaneously dealing with AI adoption, software development transformation and sovereignty requirements. His advice was to create “this backbone” through a combination of APIs, data-sharing mechanisms and emerging protocols such as MCP.
Data alone, however, is not enough. He argued that organisations must revisit enterprise architecture and build richer representations of their business operations. “On top of data, we need to provide the right object of the company,” he explained. Those objects – and the relationships between them – allow agents to understand context and produce more meaningful results. According to figures presented during the session, more than 30% of organisations see legacy systems as a limitation in their AI adoption journey.
If you are not addressing the right foundation, the right platform, it will be really difficult to stay in control.
Security and sovereignty are emerging as equally important priorities. Particularly in regulated sectors, organisations need the flexibility to operate across public cloud, sovereign cloud and on-premises environments while maintaining control over their data and processes. At the same time, agents themselves must be governed. “We are defining identities to represent agents and to put these agents under control,” the speaker said.
The challenge is becoming increasingly urgent. Between 50% and 70% of organisations now identify security and privacy concerns as a major obstacle to AI agent initiatives. To address that, he advocated stronger governance, observability and audit capabilities across agentic systems. “We need to put all of this under control,” he added.
Building before scaling
The Sogeti executive closed with a call for pragmatism. Organisations should experiment, identify the right partners and technologies, and accept that some initiatives will fail. “Eventually fail fast,” he said.
But experimentation should not come at the expense of long-term foundations. “If you are not addressing the right foundation, the right platform, it will be really difficult to stay in control,” Pierre-Olivier Patin warned.
As businesses previously built data, application and developer platforms, he believes the next step will be the emergence of agentic platforms capable of bringing order, governance and scalability to an increasingly complex AI landscape.
