"Let robots do the robots' work, and give humans the humans' work." That's the philosophy Philippe Leonhart, Smart Automation Manager (Michelin) detailed at the Tech For Industry Show, alongside Jean-Philippe Watzky, Enterprise Sales Executive (UiPath). A rare case study spanning seven years of automation — 170 robots in production — and the ongoing shift toward agentic automation.

UiPath — a European vendor, 4,000 people, €2 billion in revenue — started out with RPA (automating repetitive, time-consuming tasks) before moving up to process automation, document management (extracting information from documents to push into applications), then application and performance testing. AI has since spread throughout, all the way to agents: "the best of both worlds — an agent that makes decisions, and a robot capable of executing them."

Faced with the proliferation of agents across every vendor — ServiceNow, SAP, Salesforce, and more — UiPath positions itself around end-to-end orchestration: giving business teams a complete, auditable view of their processes, with history, by orchestrating its own robots and agents alongside those of third-party tools. With a governance and security layer ensuring that LLM access is handled in a controlled manner.

At Michelin, the automation question arose in 2018 within Corporate Business Services — the group's Global Business Services model — which handles order to cash, procurement (from supplier identification to invoice payment), part of logistics and finance, "personnel" processes (at Michelin, "people are not a resource": no HR department, a Personnel department instead), and master data management.

The founding choice: not to create three siloed centers of excellence (RPA, process automation, orchestration), but to answer a single question — process automation — by classifying digital levers into two families: insight capabilities (BI, data exposure and transformation, knowledge base) that support decisions, and execution capabilities that carry out the process end to end. With a telling distinction: the big core systems (ERP) already had their teams and organizational setup; it was the process-agnostic levers — hyperautomation: RPA, low-code, chatbots, digital adoption platforms — that needed structuring.

The CoE trains its teams around four moves: Ignite (inspire through use cases, without technology jargon — getting business teams to look at their processes, not at RPA); Advise (direct each problem toward the right service offering within the group — its own, or another's); Build (business analysis and solution creation: UiPath RPA, low-code, workflows, and the group's first chatbots to free up experts from repetitive questions); and Drive & Sustain the Value: the mission isn't to deliver robots but to create value — every RPA logs the time it saves, "hours back to business," a consolidatable value proxy ahead of its accounting translation.

The organizational journey will resonate with many: first outsourcing (consultants to learn), then a centralized CoE in Pune, India — "real expertise, but too far from the business" — and finally the federated model: a hub that explores new capabilities (including agentic ones), owns the platform, best practices and training paths, and delivery units organized by time zone, close to business cultures and contexts. "It's this model that makes us ready for the next stage."

The footprint: around 170 RPA products — counted as "products" in the business sense, a robot serving a process, regardless of how many technical artifacts sit behind it. A notable fact for an industrial trade show: R&D is the second-biggest user — freed from ancillary tasks (orders, trade show organization, bookings) to focus a bottleneck resource on value creation, following the motto "think big, start small." Examples: orchestrating tire homologation tests — each homologation authority having its own specifics, hard to industrialize, ideal to automate — and, on the manufacturing side, supporting mold management, a unit too small to justify large dedicated developments.

Three structuring lessons: build the CoE around a mission, not a technology — "just because you have a hammer doesn't mean everything is a nail"; embed value capture into solution design from the start; and rehabilitate the process approach — out of fashion since the business process redesign of the 1990s-2000s, it provides the framework for knowing where to focus effort, holistically rather than incrementally.

Three drivers are pushing Michelin toward agentic automation: demand from the CoE teams themselves, top-down process rationalization pressures, and above all a technical reality: automation remained fragmented because certain steps required cognitive intelligence that wasn't available. Take procurement: automated validation, then document understanding on PDFs — but decisions still required a human eye. GenAI now makes it possible to cover these repetitive but cognitive links.

The warning, backed by a Harvard Business Review reading cited during the session: extracting value from agentic automation requires rethinking the process — where RPA merely changed who performed a task. The target model: a chain orchestrating humans (where their decision is worth it), robots (strict business rules), and agents (which decide or propose) — a vision championed notably by UiPath CEO Daniel Dines, and consistent with the challenges encountered.

Three concrete developments: integrating agentic use cases into the Ignite offering to inspire process owners; training teams to build agents and orchestrations; and, above all, learning to work compositionally — orchestrating agents residing in Salesforce, Oracle or elsewhere requires project managers capable of bringing together contributors from across Michelin's entire ecosystem and building genuine collaboration agreements, where RPA could previously "work on the surface" without much buy-in from anyone.

This seven-year case study shows how successful enterprise automation depends more on a clear mission and organization than on the technology itself. Michelin's shift toward agentic automation confirms this logic: it's not adding new tools that creates value, but the ability to redesign processes around structured collaboration between humans, robots and agents.

Watch the full session, along with the other Tech For Industry Show replays, on our dedicated page.