Discovered how to program PLCs with AI
Any task that involves a keyboard can and will be automated
So here's the thing: I've basically decided to dedicate what's left of my career to making keyboards and mice as obsolete as fax machines. Not because I hate them (okay, maybe I hate mice a little), but because after decades of repetitive strain injuries masquerading as "productivity," I figured there had to be a better way. Turns out, that better way involves teaching silicon-based life forms to do the heavy lifting while I sip coffee and pretend to look busy.
PLCs: The Final Frontier (Or So I Thought)
This week marked a milestone in my quest for ultimate laziness—I mean, efficiency. I successfully got AI to program PLCs. For those blessed souls who've never encountered ladder logic, imagine if electrical schematics and programming had a baby, and that baby was raised by electricians from the 1960s. It's visual, it's quirky, and it seemed about as AI-friendly as teaching a cat to do your taxes. The language was literally designed so electricians could program without learning to code, which in retrospect, makes it hilariously ironic that I'm now teaching code to program without electricians.
Here's what I fed into the AI blender: original task instructions, requirements, a sample program (because even AI needs to learn our special brand of dysfunction—I mean, conventions), and specs for our various machine types. The mundane stuff that makes engineers question their life choices—matching inputs and outputs to schematics, spotting tiny differences between nearly identical machines—suddenly became a five-minute prompt engineering exercise instead of a three-hour slog through documentation.
The Plot Twist Nobody Saw Coming
But wait, it gets better. Not only did the AI figure out the differences between the sample machine and the target machine (including those special snowflake features that someone inevitably adds because "the customer requested it"), but I also threw it a curveball. The original program was in Rockwell Studio v33, and I casually asked it to upgrade to v36. Different commands, different structure—the kind of version migration that usually involves swearing, Stack Overflow, and at least one existential crisis. The AI just... did it. Created a v36 program for a different machine type based on a v33 sample from another machine entirely. I'm pretty sure that's the industrial automation equivalent of translating Shakespeare into Klingon while juggling.
There was one hiccup—a truncated routine that made me briefly question whether I'd need to actually do some work. But in a cosmic joke that would make Douglas Adams proud, Claude announced an increased context window literally the next day. The timing was so perfect I briefly wondered if the AI overlords were watching and decided to throw me a bone. This whole breakthrough happened months earlier than I expected, which has become the running theme with AI development. At this rate, by next year AI will probably be writing articles about how it taught humans to be more efficient.
The Increasingly Vertical Hockey Stick
The technology improvement curve isn't just exponential anymore—it's practically vertical. Every time I think "well, surely AI won't be able to do X for at least another year," I'm proven wrong within weeks. It's simultaneously thrilling and mildly terrifying, like riding a roller coaster designed by someone who thinks physics is more of a suggestion than a law. But hey, if it means I can finally retire my keyboard before it retires me, I'm all for our new AI overlords. I, for one, welcome them—especially if they keep handling the ladder logic.