In this talk, Emmanuel Camus of the Norwegian company Cognite explains why so many industrial AI projects stay stuck at the pilot stage, and how to move to scaled deployment. The starting point: language models rely on text, while industrial data consists of time series, 3D models, P&ID diagrams, and loosely structured documents — creating a high risk of hallucination when AI isn't connected to properly contextualized data.
The speaker outlines three recurring challenges (data access, use cases chosen for their "flashiness" rather than scalability, and organizational alignment across business, IT, and leadership), then presents Cognite's method: building a knowledge graph that links engineering, IT, and OT data around each piece of equipment, before developing high-value use cases through AI agents.
The talk includes a detailed client example: TotalEnergies, which chose a particularly ambitious deployment of its digital use cases across its upstream (exploration & production) assets, aiming to cover 36 sites — including several FPSOs — by 2028.
