December 26, 2025

Putting AI on the Grid Data You Already Have

Utilities and generators sit on enormous operational data, AMI, EMS, SCADA, generation telemetry, and use a fraction of it. The opportunity isn't more sensors. It's making the data you already have answerable, without touching the control path.

  • pod-e
  • power
  • utility
  • ai-build

A modern utility collects staggering amounts of data. Smart meters report constantly, the energy management system tracks the grid in real time, generation telemetry streams off every turbine. Most of it is stored and queried only after something has already gone wrong. The data is right there. What's missing is a way to ask it forward-looking questions in time to act. Closing that gap, not adding sensors, is where the value sits, and it can be done without going near the control path.

The under-used asset

Power operators rarely have a data shortage. They have an access problem. The signal needed to anticipate an outage or catch a developing turbine issue is usually already in the historian or the AMI stream. What's missing is intelligence that turns it into a timely answer for the operator or engineer. That reframes the build: not a new platform, but a layer on top of the data you already collect.

Read widely, never write to the control path

The discipline that makes grid AI safe and deployable is a hard line between analysis and control. The system can read from AMI, EMS, SCADA, and telemetry to understand what's happening, and surface recommendations to a person, but it does not write to the operational control path. That keeps the deployment inside NERC CIP's comfort zone, keeps a qualified operator accountable, and keeps the systems of record exactly where they are. Advisory first is almost always the right starting point in a critical-infrastructure setting.

Building inside the security boundary

Power adds a constraint the other industries don't have to the same degree: the cybersecurity boundary. Grid data is sensitive and CIP-governed, so the build has to respect where data can live and who can reach it. The good news is that an advisory layer reading from existing systems, with access designed in and logged, fits that boundary far more easily than anything that reaches into control. Build for the boundary from the start and the security review is a conversation, not a wall.

What a working example looks like

Kaysee, Cortland's reliability platform, follows this pattern on generation equipment. It reads turbine and balance-of-plant telemetry from the systems already in place, surfaces developing problems to the engineer who owns the call, and never touches the control path. The EMS stays the EMS; the historian stays the historian. The operator gains an advisory layer that makes them answerable.

The fastest, safest way into grid AI isn't to automate the control center. It's to make the data the control center already has answer the questions the operators are already asking, inside the security boundary they already work within.

Keep reading

Take the next step.

If this is the kind of work you want Claude doing inside your own operation, Cortland scopes engagements in three tiers: Walk (strategy), Run (build), Sprint (ongoing). Start wherever the risk fits.