# habert-wenvision-proj-ai-repo-agent-ide-doctrine-2026-05-05

## Veille

Methodology article by Antoine HABERT (WEnvision) that formalizes **PROJ-AI**: a lightweight methodological layer so that collective projects become transferable rather than dying with their deliverable. Structuring triad: a **version-controlled git repo** (single source), an **AI agent** (Claude Code, Cursor) that reads the doctrine at every session, and a **markdown doctrine** specifying decision protocols and agent behaviors. Six directory zones (DOCS/, IDEAS/, DR/, OUT/, DOCTRINE/, AGENT/), operational **DPEV** cycle (Decide → Promise → Execute → Verify), Decision Records scored across 7 dimensions, dual interface (business Studio + tech CLI/IDE), five agent directives, and a shared **proj-ai-commons** library that bootstraps a project in 30 minutes vs. 1 week. Metrics across 3 engagements: onboarding **3 weeks → 2 days**, structural decisions tracked **30% → 100%**, architecture doc compilation **6 weeks → continuous**. Central aphorism: ***"The project is not a byproduct of the deliverable. The project IS the deliverable."*** Explicit stance: technology 20%, **team discipline 80%**.

## Titre Article

PROJ-AI — pour que vos projets ne s'arrêtent plus au livrable (Un repo, un agent, un IDE — pourquoi PROJ-AI ?)

## Date

2026-05-05

## URL

https://www.wenvision.com/fr/articles/un-repo-un-agent-un-ide-pourquoi-proj-ai/

## Keywords

Antoine HABERT, WEnvision, PROJ-AI, repo agent IDE doctrine, agentic project methodology, transferable project memory, deliverable as project, DPEV cycle, Decide Promise Execute Verify, Decision Records, 7-dimension scoring, six repo zones, DOCS IDEAS DR OUT DOCTRINE AGENT, slash-commands /dr-create /livrable-update, PROJ-AI Studio cockpit, dual interface business tech, five agent directives, systematic doctrine ingestion, internal source citation, end-of-session summaries, proj-ai-commons, 30-minute bootstrap, 2-day onboarding, 100% decisions tracked, negligible PM archaeology, technology 20% team discipline 80%, agent-agnostic markdown, Claude Code Cursor, guided methodology, 6-18 month engagements, organizational memory, versioned doctrine

## Authors

Antoine HABERT (WEnvision — cabinet français de conseil en stratégie et IA agentique).

## Ton

**Profile**: Methodology article signed by a senior consultant at a French agentic AI consulting firm (WEnvision), published in French on the corporate site. Hybrid format between an expertise showcase and a product/method pitch (PROJ-AI being a methodology packaged by the firm). Target audience: CIOs, project directors, IT executive committees, *project managers* at companies engaged in AI transformation missions, engagement partners; secondarily, Francophone agentic-watch readers looking for operational frameworks.

**Style**: Measured methodological voice, Frenchified strategy-consulting register. HABERT alternates between **pain-point diagnosis** (projects die when the deliverable ships), **mechanics** (the repo+agent+doctrine triad, the six zones, the DPEV cycle), and **cautious quantification** (3 active engagements, 31 decisions recorded, before/after metrics). No hype, no exorbitant promise: the tone is that of a firm that did the work and presents a framework proven across 3 engagements. Tight metaphors (*"PM archaeology"*, *"30-min bootstrap vs. 1 week"*).

**Central aphorism**: ***"The project is not a byproduct of the deliverable. The project IS the deliverable."*** This phrase serves as the rhetorical pivot and inverts the canonical relationship between project and output — it is **memory and transferability** that constitute the value, not the final artifact.

**Epistemic stance**: balanced and honest. Two explicit *caveats*: (1) ***"technology 20%, team discipline 80%"*** — rejection of technical solutionism; (2) ***"agent-agnostic"*** — no exclusive Claude lock-in, the markdown overlay supports multiple LLMs. Assumed commercial posture: *guided methodology, not self-service software*. WEnvision explicitly positions itself as an advisory firm, not a SaaS vendor.

**Authority**: built through (a) the coherence of the proposed framework (six zones / DPEV cycle / five directives), (b) quantified metrics across 3 active client engagements, (c) the mention of *proj-ai-commons*, which suggests consolidated methodological capital, (d) HABERT's previous 2025-10-29 publication (*IA agentique en production : leçons de deux ans*) — already in the watch file. Francophone authority built through consulting practice.

## Pense-betes

- **Date / source**: May 5, 2026, WEnvision (French agentic AI consulting firm). Author: **Antoine HABERT** — already in the watch file with his fiche *"IA agentique en production"* (2025-10-29). This article is a formalization of an in-house methodology.
- **URL slug**: *un-repo-un-agent-un-ide-pourquoi-proj-ai*. Editorial title: *"PROJ-AI — so your projects no longer stop at the deliverable"*.
- **Problem addressed**: *"Projects stop at their deliverable, leaving no organizational memory. Critical decisions untraceable, new-member onboarding ~3 weeks, executives unable to reconstruct the rationale months later."* This is the classic *project amnesia problem* of consulting and IT project work.
- **Central aphorism**: ***"The project is not a byproduct of the deliverable. The project IS the deliverable."*** An inversion that positions memory/transferability as the central value.
- **PROJ-AI triad**: 1. **Repository (git-versioned)** — single source of truth containing raw materials, decisions, deliverables, doctrine. 2. **Agent** — Claude Code, Cursor or equivalent that *reads the repo's doctrine at every session*. 3. **Doctrine** — governance markdown files specifying decision protocols, agent behaviors, learned patterns.
- **Six directory zones**: | Zone | Role | |------|------| | `DOCS/` | Raw inputs (interviews, audits, data) | | `IDEAS/` | Hypotheses, draft thinking | | `DR/` | **Decision Records** scored across 7 dimensions | | `OUT/` | Compiled deliverables | | `DOCTRINE/` | Governance and practices | | `AGENT/` | Slash-commands and session traces |
- **Operational DPEV cycle**: ***Decide → Promise → Execute → Verify***. Four traceable steps that turn vague ideas → defensible deliverables.
- **Dual interface**:
- **Business side**: *PROJ-AI Studio* — desktop application with Cockpit, Guide, Weekly, Monthly views.
- **Tech**: CLI/IDE integration with slash-commands: `/dr-create`, `/livrable-update`.
- **Single underlying source**: one repo only, no data fragmentation between interfaces.
- **Five agent directives (PROJ-AI)**: 1. **Ingest doctrine** before answering. 2. **Cite explicitly** internal sources. 3. **Propose Decision Records** for emerging choices. 4. **Never override doctrine** unilaterally. 5. **Produce end-of-session summaries**.
- **Metrics observed across 3 active engagements**: | Metric | Before | After | |----------|-------|-------| | New-member onboarding | ~3 weeks | **~2 days** | | Structural decisions tracked | ~30% | **100%** | | Architecture doc compilation | 6 weeks (final sprint) | **Continuous** | | PM archaeology time | ~30% | **Negligible** |
- **proj-ai-commons** (shared library): each project contributes anonymized/sanitized patterns (slash-commands, Decision Record templates, doctrine fragments) → **bootstraps new projects in 30 minutes vs. 1 week**.
- **Status, May 2026**:
- **3 active client engagements**
- **31 decisions recorded**
- **Studio in internal preview**, not yet public on GitHub
- **Positioned as a guided methodology**, not self-service software (WEnvision's commercial model).
- **Target audience**: mixed tech/business teams on **6-to-18-month** engagements — internal projects, consulting missions, audits, new service builds.
- **Two explicit critical caveats**: 1. ***"Technology is 20%, team discipline is 80%"*** — adoption requires a cultural shift toward explicit decision-tracing. 2. ***"Not Claude-exclusive"*** — agent-agnostic overlay; markdown + file conventions support multiple LLM platforms.
- **Fit within the watch file**:
- **Logical follow-on** to HABERT's 2025-10-29 fiche *"IA agentique en production : leçons de deux ans"* — moving from principles (4 pillars) to a **packaged methodology** (PROJ-AI). WEnvision consolidates its offering.
- **French convergence** with **Wescale Usine Logicielle Augmentée** (2026-05-03):
- Wescale = software production chain, focused on the **build**.
- WEnvision PROJ-AI = **project/consulting** methodology, focused on **governance + memory**.
- Both emphasize *injected governance* (Wescale) / *versioned doctrine* (HABERT) as a barrier to entry.
- **Six repo zones + slash-commands + AGENT/** = a transposition to the **consulting/project** domain of the *AGENTS.md / SKILL.md / Skills* patterns documented by:
- Vincent *Superpowers* (2026-04-02)
- Anthropic *Skills* (2025-10-16, Willison 2025-10-16)
- Karpathy *Skills for Claude Code* (2026-01-27)
- Gao *Vercel AGENTS.md* (2026-01-27)
- Osmani *Agent Harness Engineering* (2026-04-19) — primitive *AGENTS.md "pilot's checklist, not style guide. Earn each line."*
- Curran *Skills-Based Plugin Architecture* (2026-04-16) — marketplace with 267 skills.
- **Decision Records (DR/)** = a transposition to project management of the **ADR** (Architecture Decision Records) pattern — already seen at Wescale (2026-05-03 *"Intention: PRD & ADR"*).
- **DPEV cycle** = the French project-management equivalent of **Plan-Work-Assess-Compound** (Shipper/Klaassen 2025-12-11), **CDLC** (Debois/Tessl 2026-02-19/26), the *Ralph Loop* (Trivedy 2026-03-10), and *Plan-Execute-Test-Reflect* (Rohit 2026-04-29).
- **proj-ai-commons** as a shared library = the equivalent of Intercom's *private plugin marketplace* (Curran 2026-04-16, 153 contributors / 267 skills), at the scale of a multi-client consulting firm.
- **Onboarding metrics 3 weeks → 2 days** align with Stripe Minions (2026-02-09/19), Curran Intercom (2026-04-16), and StrongDM Software Factory (2026-02-06) on productivity amplification through agentic orchestration.
- **"Technology 20% team discipline 80%"** corroborates HBR Anand & Wu's *Gen AI Playbook* (2025-11) *"70% people/processes"* + BCG's *Brain Fry* (2026-03-05) *"70% people/processes"* — a convergence among consulting firms on the ratio.
- **Francophone target audience** rounds out the French file: **Caseau** (2025-11-05), **Ensarguet** (2025-11-04 and 2026-03-11), **Wescale** (2026-05-03), **Mornati** (2026-03-14), **ALMIA AG2R** (2026-03), **Habert WEnvision** (2025-10-29 + 2026-05-05), **Dèbes Les Echos** (2026-04-22), **Geudin** (2026-01-26).
- **To be leveraged for**: designing an AI-native project methodology for a consulting firm or CIO office; *Decision Record templates* to standardize; a sales pitch on organizational memory; the *30-min bootstrap* benchmark to present to the executive committee; structuring an internal *proj-commons* for multi-project organizations.
- **Limitations to flag**: (a) metrics from only 3 engagements — small sample, firm bias; (b) Studio in internal preview, so not externally verifiable; (c) the 20/80 ratio is a cautious claim without its own empirical study; (d) the *proj-ai-commons library* is not (yet) public — no external validation possible.

## RésuméDe400mots

Antoine HABERT (WEnvision, a French agentic AI consulting firm) published a methodology article on May 5, 2026 that formalizes **PROJ-AI**, a lightweight methodological layer so that collective projects become **transferable rather than dying with their deliverable**. The diagnosis: projects stop at their deliverable, leave no organizational memory, critical decisions become untraceable, onboarding a new member takes three weeks. Central aphorism: *"The project is not a byproduct of the deliverable. The project IS the deliverable."*

The methodology rests on a **triad**: (1) a **version-controlled git repository** as the single source of truth; (2) an **AI agent** (Claude Code, Cursor or equivalent) that reads the doctrine at every session; (3) a **markdown doctrine** specifying decision protocols and agent behaviors. The repo is organized into **six zones**: `DOCS/` (raw inputs), `IDEAS/` (hypotheses), `DR/` (Decision Records scored across 7 dimensions), `OUT/` (deliverables), `DOCTRINE/` (governance), `AGENT/` (slash-commands and session traces).

The operational **DPEV** cycle — *Decide → Promise → Execute → Verify* — articulates four traceable steps that turn vague ideas into defensible deliverables. A **dual interface** combines PROJ-AI Studio (Cockpit/Guide/Weekly/Monthly views for the business side) and CLI/IDE with slash-commands `/dr-create`, `/livrable-update` (for tech), on top of the same single repo. Five **agent directives** structure behavior: ingest the doctrine before answering, explicitly cite internal sources, propose Decision Records for emerging choices, never override doctrine, produce end-of-session summaries.

Across **three active engagements** (31 decisions recorded), HABERT documents: new-member onboarding **3 weeks → 2 days**, structural decisions tracked **30% → 100%**, architecture doc compilation **6 weeks → continuous**, PM archaeology time **30% → negligible**. The shared **proj-ai-commons** library capitalizes on anonymized patterns (DR templates, slash-commands, doctrine fragments) and lets a new project bootstrap in **30 minutes instead of a week**.

Two honest *caveats*: (1) ***"technology 20%, team discipline 80%"*** — an explicit rejection of solutionism; (2) ***"agent-agnostic"*** — markdown supports multiple LLMs. Status: Studio in internal preview, methodology delivered as a guided engagement by the firm (not self-service).

This piece, converging with **Wescale Usine Logicielle Augmentée** (2026-05-03) and HABERT's previous fiche (2025-10-29), consolidates a French doctrine of AI industrialization. PROJ-AI transposes to project/consulting management the patterns of *Skills* (Vincent), *Plugin Marketplace* (Curran/Intercom), *AGENTS.md* (Anthropic, Vercel, Osmani) and *Decision Records* (Wescale). To be leveraged as an operational framework for consulting firms, CIOs, and project teams on 6-18 month engagements.

## GrapheDeConnaissance

- Antoine HABERT —a_créé→ PROJ-AI (METHODOLOGIE, 0.97)
- Antoine HABERT —travaille_chez→ WEnvision (ORGANISATION, 0.97)
- PROJ-AI —est_basé_sur→ triade repo + agent + doctrine (CONCEPT, 0.97)
- PROJ-AI —utilise→ six zones répertoires (CONCEPT, 0.96)
- Cycle DPEV —est_basé_sur→ Décider → Promettre → Exécuter → Vérifier (CONCEPT, 0.96)
- Decision Records 7 dimensions —utilise→ scoring sur 7 dimensions (CONCEPT, 0.94)
- Agent IA —utilise→ doctrine avant chaque réponse (CONCEPT, 0.96)
- PROJ-AI —permet→ double interface Studio + CLI/IDE (CONCEPT, 0.95)
- proj-ai-commons —permet→ bootstrap projet en 30 min vs 1 semaine (CONCEPT, 0.95)
- WEnvision —mesure→ 3 missions actives PROJ-AI, 31 décisions enregistrées (MESURE, 0.96)
- PROJ-AI —réduit→ onboarding de 3 semaines à 2 jours (CONCEPT, 0.95)
- PROJ-AI —améliore→ traçabilité décisions de 30% à 100% (CONCEPT, 0.94)
- Antoine HABERT —affirme_que→ "Le projet EST le livrable" (CITATION, 0.97)
- Antoine HABERT —affirme_que→ technologie 20% / team discipline 80% (AFFIRMATION, 0.96)
- PROJ-AI —s_applique_à→ multiples plateformes LLM (agent-agnostic) (CONCEPT, 0.94)
- PROJ-AI —est_instance_de→ méthodologie accompagnée pas self-service (CONCEPT, 0.94)
- PROJ-AI —s_applique_à→ équipes mixtes tech/métier sur 6-18 mois (CONCEPT, 0.93)
- Doctrine markdown —utilise→ repository git (versionnement) (TECHNOLOGIE, 0.96)
- PROJ-AI —est_basé_sur→ doctrine versionnée + slash-commands (CONCEPT, 0.93)

---
Canonical: https://www.thekb.eu/en/fiches/habert-wenvision-proj-ai-repo-agent-ide-doctrine-2026-05-05/