Industry: plants, quality, maintenance

In Lyon, a Boilermaking TPE led an AI tool on its production history to estimate phases, time, material, margin and price from the 3D files of the parts.

At the ArcelorMittal site in Fos-sur-Mer, the "Final check" device — two lasers, seven cameras and the AI — identifies the coils and signals defects so that operators can intervene more quickly.

Danone has automated demand planning in 38 countries with SAP IBP and machine learning models: forecasts, which require 160 hours, are obtained between 20 minutes and 4 hours, for more than 230 forecasters.

Lucille Roger, head of MOA IT at the Compagnie Nationale du Rhône, moved an AI agent from a POC 2025 to a production start in March 2026: faced with a defect on an alternator, the operator interrogates notices, diagrams and industrialNGOs reports.

Gaëtan Goguet, maintenance manager at Saga Décor (Verallia group, 45 people, glass bottle decor), exports the GMAO in CSV and asks Copilot Premium the Pareto for faults: a "big study" comes out in less than ten minutes.

Paul F. Minssieux, DSI de Clauger (Industrial ETI, 3,000 employees, cold and air treatment), opposes a project of €80,000 to an assistant created in one hour that saves 5 minutes a day; 200 makers have already fed a library of more than 250 assistants, some 50 of whom are used daily.

Frédéric Minssieux, President of Clauger, worked with the industrial ETI lawyers on an agent able to improve supplier contracts by linking vigilance points in existing documents — momentum given at the same time as the 300 Copilot licenses.

At Laitik (Lait-sprit d'Ethique, Tremorel, 15 employees), the milk and cream industry, the quality service uses ChatGPT to write the processes — milk sterilization, wearing EPI — and then decline them into instructions, a work that represented hours; The technical team re-reads for consistency for IFS certification.

Setic Pourtier, an SME from Roanne (85 people) who manufactures machines for the cable industry, first built a predictive maintenance AI, then reoriented it to monitor the quality of the cables produced, more expected by its customers.

Aldes, an air quality solution manufacturer, co-developed with the start-up Catalog a tool that classifies and describes e-mails as prospects and customers, and then pre-writes the quotes.

Air Liquide has put online, as the first product of its Hub LLM, a multilingual chatbot that answers questions from employees of its shared financial services about their processes and procedures.

Suez has built "GenAI Core", a central platform that frames its 50+ generation AI projects and displays consumption, cost and CO2 emission estimates for each request.

Alstom has put procurement and finance agents in production: order management of low-cost components, assistance to the financial services centre and compliance check between purchase request and purchase order.

Babynov, a company of the Materne group that manufactures infant cereals in Courmelles (Aisne), automated the inspection of its metal boxes with an industrial vision system equipped with automatic learning, connected to an ejector.

Veolia has linked Mistral AI models to its data and knowledge base to help its teams drive the facility through conversation.

STADLER, a family-owned company that designs automated waste sorting facilities, has equipped its 650 employees with an assistant from AIA and has over 125 customized GPTs, especially for translation and e-mails.

TotalEnergies deployed MARGE, an agent that analyzes and ranks maintenance response reports, a task previously done by hand.

At TotalEnergies, Agent BuyerCompanion helps employees on purchases not exceeding EUR 50 000, from the drafting of the need to the identification of the supplier, respecting the rules of conformity.

Danone uses an agent to analyze customer orders, regardless of the format, and cross-checks prices with its internal systems to correct deviations before billing.

Valeo has developed with Gemini Enterprise its first engineering agents: creation of test cases from the design, continuous review in the integration chain, assistance in the drafting of system specifications.

Sepalumic, an aluminium manufacturer, has re-launched a project that has been blocked for six years thanks to the AI: identify in a few seconds some 3,500 similar references the profile that enters into lacquering.

Romain Méridoux presents the "score factory", customer scoring platform with advisor interface: filtering prospects and generating sales arguments. The Datalab also industrializes marketing campaigns with customized arguments by AI generic.

Fouad Maach, Head of Architecture, Engineering and Expertise at the Veolia Group DSI, indicates that the agents are widely exploited in support and up to the commercial services, for example to search for technical proposals in the internal catalogue. 250 agents are in production.

Fouad Maach, DSI architecture group of Veolia, describes a no-code faculty open to all internal teams: design an agent alone or an agent who orchestrates others. "An agent is rarely isolated; He'll call and consume other agents. 250 agents are in production, wide adoption.