The Agentic Discom x PEACon 2026

Discoms own the last mile of the grid, serve every consumer, collect every bill and absorb most of the losses. Over the last decade they have also built a sensing layer: smart meters, SCADA, GIS, outage systems and modern billing. The result is more data than any utility has ever had, and far fewer people than it would take to act on it. Most of the signals that could help sit unread in dashboards and data lakes.

What an agent does that a dashboard cannot

Most discoms already have analytics. A model scores consumers for theft risk, a report lands in an inbox each month, and the trail goes cold because acting on it depends on someone finding the time. An agent closes that gap:

• It works continuously, on data as it arrives, instead of waiting for a monthly run.

• It uses existing data: it pulls records from billing, checks asset history, drafts a work order or a consumer notice, ready for a person to approve.

• It learns from outcomes. Every confirmed or rejected case makes the next one sharper.

Analytics produces information. An agent produces a prepared decision.

Agents propose, people decide

Much of the AI story sold to utilities is automation: lights-out operations, fewer people. That story misreads the work. Is this anomaly theft or a faulty CT? Should this household be disconnected or offered a payment plan? Is this transformer worth replacing now or next year? These decisions carry legal, social and safety consequences, and they stay with people who know the network, the community and the regulator.

So the Agentic Discom has a key principle: Agents do the watching, sorting, drafting and suggesting. People do the judging, deciding and acting. Every consequential decision has a human sign-off, every recommendation shows its reasoning, and every correction from a veteran engineer (“that pattern is a faulty CT, not theft”) is captured and passed on.