Resources
The book, a white paper, a classification of AI patterns, three articles on the approach, a demonstrator, and notes from a major AI event.
Methodology & Portfolio
The frame I work from, set out in full: what a large organization has to put in place to make AI hold, from the leadership down to the application estate. Forty-eight pages, eighteen diagrams - the three convictions, the five offers, and the argument that ties them together. Edition 2026, in English. 7.1 MB.
Download the PDF ↓The Silent Intelligence
A late-2025 state of the art on application management moving from break-fix to prediction and autonomy. Seven use cases, the technologies underneath, the business case, and the risks - including the one that names the paper: silent intelligence must not become opaque intelligence. Forty-three pages, v1.0. 3.8 MB.
AI Patterns Classification
One A0 sheet, twelve families of corporate use cases. For each: what the business actually does with it, the AI patterns underneath, and the market products that implement them, named. Built to situate a request before scoping it. November 2025.
Human-first
Augmenter ou automatiser : une question de priorité, pas de choix.
Why augmenting the craft comes before automating the process, and why the order matters more than the choice. Runs the 1990 reengineering promise against what it actually delivered, and puts the target back where the value is: tacit knowledge.
End-to-end-first
Cas d'usage et MVP : quand le découpage détruit la valeur.
The two ways of cutting a transformation down - a portfolio of use cases, or a minimal product you grow - and what both leave out: the links between steps, where the value sits. Cost made cutting necessary. It no longer does.
Knowledge-first
Connaissance métier : prérequis ou livrable de la transformation IA ?
Business knowledge is usually treated as a prerequisite, something to clean up before AI can start. The case for treating it as the deliverable instead, and why knowledge management failed in the 1990s: writing down what you know pays nothing.
MyPronote
A working demonstrator, built end-to-end: school data, transport and local conditions pulled into one daily view for a family. An overview blueprint, an 84-second audio briefing, and a console showing the orchestration underneath - agents, workflow logs, model status.
Adopt AI
Reading notes from the Adopt AI exhibition in Paris. The official portal and replay channel, then eleven NotebookLM notebooks built from the sessions: two stages, CEO and Tech, and nine sector reads from finance to sport.