Across June 8 and 9 I was in a room in Los Angeles, talking to a few hundred utility people about something I care about a lot and say out loud surprisingly rarely: how applied AI actually gets from a slide deck to a production system at a US utility. The session was called "Applied AI for Grid Modernization: From Pilot to Production at US Utilities," twenty five minutes, part of Smart Grids USA 2026.
I will be honest about the part that made me grin backstage. The program had NERC, Schneider Electric, Siemens Energy, Dominion Energy, Lawrence Berkeley National Lab. Serious names who keep the lights on for tens of millions of people. And there I was, a founder who, the night before, had been arguing with an AI model about a database migration. The gap between "AI for the grid" as a panel topic and "AI as the thing I actually build with at 11pm" is the whole story, so let me tell it.
The unglamorous version of applied AI
Most talks about AI and infrastructure live at thirty thousand feet. Resilience. Optimization. Digital twins. All real. But the work I do is two feet off the ground. At AvanSaber we ship an unusually broad set of AI-native products, and a lot of that gets built shoulder to shoulder with Claude. Not a chatbot bolted onto an old app. The model in the loop while the thing gets made.
So when I tell a utility audience that the leap is "pilot to production," I am not speaking in the abstract. The leap is the boring middle: the validation, the audit trail, the reversibility, the part where a probabilistic system has to live next to deterministic controls that absolutely cannot be wrong. That is the same discipline whether you are posting a journal entry in an ERP or switching load on a feeder. The demo is easy. The Tuesday after the demo is the job.
Building with Fable
Which brings me to the co-builder. The current model I lean on is Claude, in its Fable generation, and the honest experience of building with it is not the one in the keynote sizzle reels. It is quieter and better. It is less "AI writes your company for you" and more "a fast, tireless collaborator who never gets bored of the unglamorous part."
A concrete example: the kind of project that used to be a quarter of engineering now fits in a week. I built a working Laravel Forge alternative in seven days with AI-assisted coding, the server provisioning, the deploys, the unglamorous plumbing, because the model never tired of the thousandth edge case and I stayed the one accountable for the result. That is the speed I was really describing on stage.
The thing I keep noticing is that the experience rewards the same instinct I was on stage advocating: treat the AI as a capable junior who is brilliant and occasionally confidently wrong, put real guardrails around it, and keep a human accountable for the result. That is not a limitation. That is the operating manual. The teams who win with this are not the ones who trust it the most. They are the ones who verify the fastest.
Why I bother saying it on a stage
Utilities are exactly the kind of place where "move fast" is the wrong instinct and "move deliberately, with proof" is the right one. That is a feature, not a frustration. The most useful thing I think I said in those twenty five minutes was not a prediction. It was a permission slip: you do not have to choose between AI and control. The whole point of building the boring scaffolding is that you get to keep both.
Then I flew home, opened the laptop, and got back to the boring scaffolding with my fast tireless collaborator. The grid and good software want the same thing in the end. Reliable, auditable, a little dull, and quietly doing their job while everyone argues about the future on a stage.