Jean-Philippe Arné
AI Transformation Lead
I lead AI transformation: challenging existing workflows, then building the automation where it belongs.
I started as a software engineer, mobile apps, web platforms, drone firmware, a training tool for the British Olympic sailing team. I wrote code, shipped products, and learned what building actually means.
Then I spent five years learning how organisations break and how to fix them. At a European media group, I redesigned a department from 10 to 50 people. At an industrial electrical manufacturer, I built dashboards that turned a 500-item opaque backlog into a visible, measurable system. At a national energy utility, a global digital consultancy, and a French e-commerce leader, I coached teams through transformations that stuck because the teams owned them, not me. Today I lead the AI initiative inside that same national energy utility: a cross-functional team of about ten, mapping our processes end to end, from the macro value stream down to the exact step a person or an AI agent performs. The goal is a real AI strategy, deciding where AI belongs, where humans stay, and which areas are too sensitive to automate.
Since December 2025, I've been building AI agents intensively as a solo engineer. 26 projects in 5 months. When Claude Code shipped, I put my hands on it immediately. The operational knowledge tells me where to point the AI. The engineering background lets me build it.
This isn't agile coaching with new tools. It's AI transformation: challenge how the work is done, then build the automation where it belongs. The years of transformation work taught me which questions to ask. The engineering and AI let us act on the answers.
I map how work is really done, end to end and at every scale, from the value stream down to the step a person actually performs. Then I challenge it. Lean, value stream mapping, KPI and dashboard design, organisational redesign. Most workflows carry old decisions that nobody has tested for years.
Automation is not the objective. I take it to the experts who actually run the steps, and we decide together what should be automated, what should stay human, and what is too sensitive, regulated, or low-value to automate. That judgement is where many AI projects are weakest. It is also what makes the automation credible.
I build AI tooling, workflow automation and multi-agent systems, hands on. At a national energy utility, I lead a cross-functional AI initiative of around ten people across engineering, AI, design and product. The work is the same discipline in practice: challenge the workflow, decide where AI belongs, then build.
A self-hosted MCP bridge that gives my AI agents controlled access to my services without ever handing them a credential. Encrypted vault, per-capability switches, an audit log the agent cannot touch, and drop-in connectors for anything new. Built end to end by the Black Box App Factory: 1,053 tests, security-audited, MIT, alpha.
A 22-phase SDLC where Claude plays 7 roles (PM, Architect, Developer, QA, Security, Workflow Architect, User Tester). Configurable quality tiers, file-based state machine, agent failure recovery. My operational transformation knowledge encoded as executable software.
A Waybar fork that hosts a permanent Claude agent in my status bar. One keystroke drops the bar into a terminal pane running an always-on Claude session, stateful across reboots, accessible from every workspace. No Electron, no browser tab, just libvte inside GTK. Running on my daily driver. Alpha, MIT licensed.
My knowledge bases and my entire computer connected to Telegram via AI agents. Take notes, retrieve files, execute tasks on my machine, access my professional knowledge, from anywhere, all the time. Daily use since January 2026.
Autonomous Instagram agent: character-driven personas, 30-day rolling content plans, AI-generated 2K images, automated posting via Instagram Graph API. Ran one account at 3 posts/day fully autonomously.
AI photo transformation SaaS. Full production stack built solo: Next.js 15, Supabase, Stripe, Google OAuth. Shipped to production, ran for one month, failed to find market fit. Shut down. The build was a success; the market wasn't.
Nightly automated scalping bot factory with daily parameter tuning. Paper trading across 7 repos of progressive iteration, swing, scalp, LLM-assisted strategies. Running every night since April 2026.