Architect of AI-based operating models
AI is not a question of tools. It is a question of how the organization runs.
A leader's real question is not "which model" or "which demo". It is: how does my organization start to run differently once agents take on roles, responsibilities, and work alongside people.
What I do
I work one level above the technology: the level of the operating model— how a company or a public body organizes, decides, produces and keeps its commitments once agentic AI is part of it. It is the level executives speak at, and the one the market handles worst: saturated with demos and POCs at the bottom, empty of models that hold at the top.
Designing these models is a concrete craft: agent identity, the architecture of an organization that mixes humans and agents, industrialization, reliability and governance, sovereignty and regulatory constraints treated as design parameters. I designand I implement operational agents in production, with a mission, bounded authority, a human owner, and explicit controls.
The threshold. A tool assists an action. An operational agent receives a bounded mission, acts through its own identity and permissions, and remains accountable to a human owner. That shift requires a different operating model.
The five pillars
A single roof — AI-based operating models — held up by five faces.
Agent identity
Who the agent is within the organization: role, responsibility, boundary of autonomy, what it decides and what it hands back to a human.
Organizational architecture
Structuring an augmented organization of humans + agents: role mapping, orchestration, interfaces, governance.
Industrialization
Going from POC to a model that runs at scale: frugality, cost, deployment, scaling up.
Reliability & governance
What makes a model hold: guardrails, agent evaluation, observability, human-in-the-loop.
Sovereignty & regulation
GDPR, AI Act, DORA, sovereignty treated as design parameters. A pillar, not the cage.
Start with the organization
The series on the organizational architecture of agents starts with a concrete question: before connecting more agents, how do we define who does what, with which authority, under whose accountability, and with which stopping mechanism?