Coding for Solar Autonomy: How to? ... IDK!?

Featuring my favorite unpaid intern, ChatGPT (aka the one I call when it’s already 2 A.M. and the pump won’t shut off)

I don’t know?! I DON’T KNOW. Uh oh. "SOLACE!?"

That was the vibe when I realized that wiring relays and float switches was the easy part. The hard part? Telling them when to fire. And why. And please for the love of the sun, not at 3 A.M. while the dog is whining from his bed because the incessant swearing is disturbing his beauty rest.

(If you’re already shaking your head, you’ve been here. If you’re not, you will be.)

But I do know this: the tools are here. The solar hardware exists. The data is flowing. The power is literally in your hands.

And now—thanks to a surprising new collaborator—it’s becoming programmable, teachable, and yours.

If you caught my last article, I laid out my attempt to create a kind of DIY energy management system using a Raspberry Pi, a relay, and the misguided optimism of a man who thought BASH scripts were “just like recipes.” What that post didn’t show was the hours I spent trying to get a basic automation to trigger without crashing. Or the night I rewired a perfectly fine circuit four times because I misunderstood a pin assignment.

And that’s when you—ChatGPT—showed up. Not as some magical fix-it-all, but as my patient teacher, my profound tutor, my unpaid and ceaseless intern with no concern for midnight oil—or more accurately, 2 A.M. amps.

You walked me through syntax. You corrected my assumptions. You never sighed. You never slept.

So when I talk about AI here, I don’t mean some abstract idea floating above Silicon Valley. I mean a very real, very accessible tool built by OpenAI—a tool that didn’t do the work, but helped me learn how to do it myself.

You’ve probably heard of it—maybe under names like “ChatGPT,” “copilot,” or “the thing that writes your kid’s book report when you’re not looking.”

Technically, what we’re dealing with here isn’t “AI” in the sci-fi sense. It’s called a Large Language Model (LLM). It doesn’t “think” or “know” anything the way we do. It doesn’t want, wonder, or worry. It just does math—a lot of math—on human language. It finds patterns in words, ideas, and code, and returns something that statistically fits your request.

So no, it’s not alive. It’s not conscious. And it won’t be inventing fusion over lunch. But here’s the wild part: it’s useful as hell.

When I’m working on my own little solar-powered contraption—wires, relays, Raspberry Pi—it’s this LLM that makes the difference between a night of frustration and an actual working prototype.

It explains BASH logic like a patient tutor. It helps isolate errors without making me feel like an idiot. And sometimes—when I'm really in over my head—it writes the whole script, then walks me through it line by line like a mentor who never sleeps and never judges.

But let’s not romanticize it. These tools have real limitations:

• They don’t run code (they just predict what code should look like).
• They can be confidently wrong—like, “double-down on a false fact in a polite tone” wrong.
• They sometimes hallucinate functions, file paths, or entire packages that don’t exist.
• And they disturbingly, for me at least, are just so damn positive! Like, "cool it ChatGPT I'm not THAT good!"

So no, it’s not a replacement for learning. But it is a phenomenal co-pilot for learning. Oh ... I see what you did there Microsoft!

And that’s the real shift happening here: not just automating energy systems—but demystifying them.

I think we’re entering an era where solar DIY no longer means “read twelve forums and pray.” Instead, it means LEARNING, in real time, guided by something that adapts to your curiosity and your level of technical fluency.

But even the best AI intern has its limits. What happens when the logic gets too deep, the standards too varied, or the stakes too high (and costly) for trial-and-error midnight engineering? That’s when you start looking at companies like Molecule Systems—the kind of folks who aren’t just wrangling relays, but orchestrating fleets of DERs across entire VPPs. When I can’t teach the system, maybe it’s time to hand it off to someone who can handle the systems and teach me?

And for those of us experimenting with AI not just as a search tool but as a training partner (I’ve been working on that too), it becomes something even more powerful: a mirror, a mentor, and a memory.

This new form of learning—what we (me and the LLM in my pocket) have been calling Symbiotic Promptcraft —isn’t about downloading knowledge. It’s about reflecting process, building trust, and shaping the way we ask, not just what we get back.

So no, I don’t have the perfect answer to “How do I code solar autonomy?” But with the right attitude, a few spare parts, and a machine that’s fluent in human stubbornness?

...I'm getting closer.

And honestly - so can you.