Twenty Years of Cleaning Bugs
The Sunday-morning call has not changed in twenty years. Only the bug has: from a crash, to a configuration, to a sentence that is fluent, confident and wrong.
AI
2026
Standing Watch
A daily security watch that runs itself: a swarm of local agents, a Kanban board, and a background task wired into a one-person SOC. A Vibe Orchestration field report.
Hermes Agent on Mac: a local-first setup with two profiles, A (local) and B (cloud)
A sovereign, local-first AI assistant on Apple Silicon for sensitive data, plus a high-performance cloud mode when the task earns it. Two isolated Hermes profiles, one toggle.
Deploying GenAI Without Breaking GDPR
A leader freezes a working AI project out of fear of GDPR and the AI Act. The three texts are not three walls. They ask, in three vocabularies, the same four operational questions.
Prompts Are Dead, Long Live Direction
The prompt engineering era is over. What replaces it isn't a better prompt. It's the room the prompt sits inside.
The Expert's Paradox
If you need to be an expert to verify what the AI says, you don't need the AI. The paradox that quietly defines AI use in 2026.
When the Room Empties
The share of genuine human content on the live web is shrinking faster than most organisations realise. What a CIO actually sees in logs, pipelines, and API traces, and why it matters for everything downstream.
Algorithmic Slop
Confident, well-formatted, quietly wrong. Five real cases from a fractional-CIO practice, and the discipline that catches them.
Vibe Orchestration
Most people chat with AI. A few direct it. What changes when you stop being an operator and start conducting.
In French
2026
Qui formera les experts de demain ?
FR
Le travail confié aux juniors ne servait pas qu’à produire, il les formait. Si l’IA le reprend sans qu’on y pense, on leur retire de quoi apprendre le métier.
RAG sans illusions
FR
Ce que coûte vraiment un RAG en PME, et pourquoi la plupart des projets qui en demandent un ont besoin de cinq documents et d’un prompt.
Acculturer une équipe à l’IA générative
FR
Payer des licences ne suffit pas à installer l’usage. L’adoption de l’IA générative bute surtout sur la peur et la permission floue, bien plus que sur le choix du modèle.
Où héberger son IA ?
FR
Le pilote fonctionne. Reste la question que personne n’a posée avant de commencer : où vit le modèle, et de quoi accepte-t-on de dépendre ?