Bringing AI to Energy Management with Sam Cox of Mountain Vector Energy
Ep. 01

Bringing AI to Energy Management with Sam Cox of Mountain Vector Energy

•

Episode description

Somebody has to build the road before you can drive on it.

Sam Cox, CTO of Mountain Vector, joins Ryo Koyama to talk energy data, legacy utilities, and the unglamorous infrastructure behind Physical AI — from dot-matrix-era utility bills to running hundreds of remote devices with a 12-person team.

Bringing AI to an industry stuck in the past

Mountain Vector does energy management: utility bill analysis, solar and battery evaluations, and analytics that show customers where their consumption and spend actually go.

The hard part isn’t the AI. It’s the data. Utilities are among the most conservative industries in existence, and many municipalities still issue bills in formats that look like dot matrix printouts — poor-quality account numbers, inconsistent layouts, barely legible to a human. Before any insight is possible, someone has to make that readable to a machine.

Operating at national scale

Mountain Vector serves customers coast to coast:

  • School districts
  • Industrial facilities
  • Utility providers
  • Public sector organizations

One industrial customer tracks locational marginal pricing on 15-minute and five-minute intervals, deciding in near-real-time whether to curtail operations when costs spike. Mountain Vector is building predictive capability on top of that, so customers can plan ahead instead of reacting.

Physical AI needs infrastructure first

The models get the attention. The roads don’t. Factories, sensors, meters, and edge devices have to be connected and reliably reporting before AI can do anything useful with them.

With hundreds of devices deployed nationwide and a small team in New Mexico, Mountain Vector can’t send a technician every time something breaks. Remote access is what makes the model work:

  • Configure deployed devices
  • Troubleshoot and diagnose failures
  • Push software updates
  • Collect operational data into central analytics

Sam had scoped building this himself — reverse SSH tunnels, custom infrastructure — before finding Remote.It and concluding it was cheaper and faster not to. He also recounts a support case where the root cause turned out to be an MTU issue on his own end, and Remote.It tracked it down anyway.

Declarative, not imperative

Sam’s sharpest observation: software has always required you to specify how. AI shifts that to what. Describe the outcome — pull these fields from ten thousand inconsistent utility bills — and let the system work out the method.

AI won’t take your job, it’ll fill your queue

Sam’s experience runs against the prevailing anxiety. AI hasn’t reduced his workload; it’s multiplied his throughput. Where he shipped one feature a week, he now ships six. The queue empties faster and refills faster.

His analogy: Excel didn’t eliminate accountants. It let them serve far more clients and do far more sophisticated work.

The catch

Language models are very good at producing exactly what you asked for — which is not always what you needed. In production energy systems, that gap matters. Human expertise becomes more valuable as the tools improve, because someone has to judge whether the output actually survives contact with reality.

Video version: https://www.youtube.com/watch?v=EeT41BRmi4I

0:00Introduction
0:19Mountain Vector and Energy Management
0:56Utilities Stuck in the Past
2:25Operating at National Scale
3:52Physical AI Needs Infrastructure First
5:48Why Remote Access Matters
8:34Declarative, Not Imperative
9:35AI Won’t Take Your Job
11:45The Catch: What You Asked vs. What You Need
13:32Closing Thoughts