Shell uses C3 AI agents to automate predictive maintenance

Shell will deploy autonomous agents from C3 AI across its upstream and downstream operations to automate the predictive maintenance lifecycle and reduce unplanned downtime.

The company is expanding its use of the C3 AI Reliability Suite, which already monitors more than 30,000 critical assets at production and processing sites. The next phase adds agentic AI on top of existing machine learning models.

The platform ingests high-frequency operational technology sensor data and links it with business records from enterprise systems such as SAP. Operators configure agents for specific equipment types, including pumps, turbines and compressors, by setting objectives and permitted actions.

When a machine-learning model detects a deviation from normal baselines, an agent gathers contextual information such as recent maintenance logs, environmental conditions and upstream process variables. The agent then performs root-cause analysis, drafts a work order, verifies parts availability in inventory and can generate procurement requests within the company’s ERP workflows.

Human planners can approve or override agent recommendations. Where procedures and outcomes are repeatable, Shell can enable automated responses without continuous human oversight.

C3 AI’s software provides a model-driven integration layer that joins real-time OT data with structured financial and maintenance records. Connectivity to systems such as SAP allows agents to operate inside existing planning and execution channels used by Shell’s teams.

Shell and C3 AI say the automated lifecycle shortens the interval between a predicted issue and completed repair, which reduces unplanned downtime and limits unnecessary maintenance. The companies also note potential safety and environmental benefits from earlier intervention on developing faults.

Stephen Ehikian, president of C3 AI, described the expanded work with Shell as proof that enterprise AI can operate at global scale and deliver large economic returns, saying it can reduce unplanned downtime and deliver “hundreds of millions of dollars in economic value.” Sandy Gupta, vice president at Microsoft, called the rollout an example of “real applications, running in production,” and said the systems on Azure link AI to operational workflows.

Shell’s deployment follows several years of collaboration with C3 AI and Microsoft. Company officials say the program aims to reduce the manual steps that often delay repairs after a failure prediction, allowing engineers to focus on approvals and exceptions while agents handle investigation and planning.