Antoine Habert

Antoine Habert, consultant senior at WEnvision, published two pieces roughly six months apart, and the pair traces one argument from an incident-response system to a project governance method.

The first (2025-10-29) reports two years of deploying agents in production at a European fintech, starting with AI4Ops in 2023: incident resolution, diagnostic qualification, status communication and infrastructure monitoring, reaching 100% automation of level-1 support with cost cuts above 90% under banking compliance constraints. From that, Habert names what the 2024 framework wave left undone. He states plainly that LangGraph, CrewAI and AutoGen lack production foundations: they supply orchestration, not reasoning transparency, structured organizational memory, genuine cognitive cooperation between agents, or supervision that grades with system maturity. WEnvision formalized the four pillars in its RAISE platform.

The second (2026-05-05) formalizes PROJ-AI, built on a git repository as single source of truth, an agent (Claude Code, Cursor or equivalent) that reads a markdown doctrine each session, and a six-zone repo layout with Decision Records scored on seven dimensions. Across three active missions and 31 recorded decisions, Habert reports onboarding falling from three weeks to two days and structural decisions traced rising from 30% to 100%. His aphorism: "Le projet EST le livrable."

The constant across both is a refusal of technical solutionism, stated as technologie 20% / team discipline 80%. Habert keeps returning to the same claim: the missing layer is cognitive governance, not another framework.