Generative AI for Energy: Operations Knowledge and Customer Intelligence

A utility deploys a commercial AI tool for control room operators. Within weeks, an operator uses it to look up a switching procedure. The AI generates a confident answer that's close to the actual procedure but omits a critical safety step — the grounding verification that's required before live-line work can proceed on the affected section. The operator catches the omission because he's experienced. A less experienced operator might not have. The switching procedure exists in the utility's operating manual, was updated after an incident three years ago, and the AI's training data predates the update. This is the fundamental problem with generic AI in utility operations: the consequences of a wrong answer aren't a customer complaint — they're a safety incident on energized equipment.

Generative AI that works at utilities requires grounded retrieval against the utility's actual documents — current operating procedures, current NERC standards, current equipment manuals, current tariff schedules. Cited sources on every answer so the operator can verify. Explicit refusal patterns for safety-critical questions that the agent can't answer from the document set. NERC CIP boundary compliance for any agent that touches BES data. And the training that makes every user understand what the agent can and cannot be trusted for. Done with this discipline, generative AI becomes a real productivity tool for operations and customer service. Done casually, it creates safety risk.