Cognite makes trust the battleground for industrial AI

BANGKOK — Cognite has reduced its Gastech pitch to four words across its Bangkok stand: “Industrial AI You Can Trust.” Behind the slogan is one of the central technical problems now facing energy companies: an AI agent is only as dependable as the industrial context beneath it.

Cognite’s approach starts with Cognite Data Fusion, which brings operational, engineering and enterprise information together and maps relationships through an Industrial Knowledge Graph. Its Atlas AI layer then gives industrial AI agents access to that contextualised information.

The model is not the starting point

That distinction matters in plants where a single asset can appear under different identifiers in maintenance systems, engineering drawings, sensor histories and enterprise software. Cognite argues that simply centralising those sources does not give an AI system enough context to reason reliably about operations.

The company’s September 2026 product release pushes further into agentic workflows. Cognite says Atlas AI can retain operational context as users move between workspaces, while citations allow engineers and operators to trace generated insights back to underlying industrial information before acting on them.

Opening the industrial context to other agents

A particularly significant development is Industrial MCP, currently described by Cognite as a public preview. It is intended to let third-party AI tools and agents connect securely to Cognite Data Fusion while using the same governed knowledge graph and permissions as Cognite’s own Atlas AI environment.

That points toward an industrial AI architecture in which the value is not simply the language model. The defensible layer may increasingly be the governed representation of the plant: equipment relationships, telemetry, documents, work history, engineering models and the permissions determining what an agent is allowed to see and do.

Energy is already the proving ground

Cognite lists energy deployments including TotalEnergies, Aker BP, OMV and Equinor among its customer examples. The company says TotalEnergies is using its platform across 36 upstream assets, integrating information from multiple operational and enterprise sources.

Those examples and performance figures remain company-reported rather than independent Stratawire validation. But the direction is clear on the Gastech floor: industrial AI vendors are increasingly competing on whether their systems can provide traceable context and controlled access, not merely produce fluent answers.

Seen alongside today’s discussions about agentic AI and engineering intelligence at Gastech, Cognite’s stand captures the emerging question neatly. The challenge is moving beyond an AI that can talk about a plant toward one that can understand enough of the plant’s real operational state to be trusted with consequential work.