Post by Clement Bruneau

Yesterday we hosted our very first AI Breakfast at Hôtel de Crillon, A Rosewood Hotel with Yoann Bourges, Head of AI & Digital Experience at Eramet, to hear about their experience deploying AI internally. We talked through what worked, what didn't, and what they'd do differently. 3 things that stuck with me: GenAI as the fastest path to proving value — including for decisions that go beyond AI. Eramet uses Dust to move fast: test a hypothesis, surface a result, show leadership something real. In some cases, once the value is proven, they hand it off to more specialized tooling. That's not a limitation — that's exactly the point. The hard part was never RPA. It was knowing what was worth building. GenAI solves that. Every person in the room was fighting the same battle with their ExCom. Consultancy, industrial group, financial services - different industries, same problem: how do you build a case for a "go slow to go fast" approach when your leadership wants competitive benchmarks and ROI now? Eramet's answer: engineer one concrete win early. For instance, their procurement team ran a day-long hackathon with Dust and surfaced a few million in potential savings. The business case doesn't come first — it comes from the first real result. They started with zero governance. On purpose. The bet was: liberate first, structure later. It worked for adoption — and created 1,000 agents, many now unused. Governance, FinOps, an ethics committee: all being built retroactively. The lesson they'd apply if starting over? Governance earlier — not because it feels urgent at the start, but because retrofitting it is expensive. Big thanks to Yoann for accepting the invite and for the level of honesty - this kind of peer sharing is rare. To Amelie D. for the flawless organization. And to everyone who traded their Wednesday morning for croissants and AI governance: Sylvain DEVES, Chloé Rabanel-Clerc, Ky Lan Vu Tong, Alizé Santini, Valerie Bompard🎗️, and the rest of the room. PS: next time we'll get a proper photographer 🙂