Everything else on this site describes what I’m building and why. This page is different — it’s my own opinion, based on 50 years of hands-on, cross-disciplinary experience, not something I’m asking you to take as settled fact. I could be wrong. But I think it’s important to say plainly, because it’s the belief underneath everything else here.
I don’t think this round of automation is like the others. Historically, when a new technology automated a category of work, people moved to a different category — farm labor gave way to industrial jobs, industrial automation gave way to service and knowledge work. That pattern gets cited constantly as a reason not to worry this time either: automation has always eventually created more jobs than it destroyed, so why would now be different?
Here’s why I think it’s different: every previous wave of automation was narrow. A mechanical loom could weave. A robotic arm could weld a specific joint on a specific assembly line. Each new machine was good at one category of task, which is exactly why humans could always move to the next category the machine couldn’t yet touch. What’s converging right now — neural-network-based AI, married to increasingly capable sensing (machine and computer vision) and increasingly capable physical actuation (permanent- magnet servo motors, modern motor drives, the whole stack of VLSI/CPU/GPU/TPU/FPGA compute driving it) — isn’t narrow. It’s general. It can be trained to do a very wide range of tasks, intellectual and physical, that previous automation could never touch. I don’t think there’s a comparably large “next category” for people to move to this time, because the new technology isn’t leaving one uncovered.
I also think the “Dirty, Dangerous, and Dull” framing is a deflection, not an explanation. It’s a common talking point from people building physical automation: we’re just taking over the unpleasant work, freeing people up for better things. I don’t doubt that’s part of it. But I don’t think any serious person believes companies will stop there. If a task is clean, safe, and genuinely interesting — but a machine can do it reliably and more cheaply — nothing about business incentives suggests it gets left for a human out of consideration. “Dirty, Dangerous, and Dull” describes where automation started being politically easy to justify, not where it stops being economically attractive.
So is there work that’s genuinely safe from this? Almost certainly some, for now. Building trades — plumbing, electrical work — are a good example people often raise, and they’re right that these are currently very hard to automate, largely because existing buildings were never designed with automated maintenance in mind. Fitting a robot into a century of accumulated, inconsistent, human-scale infrastructure is a much harder problem than building a robot that could do the underlying task in principle. My guess is that maintenance of existing structures stays a human trade for a long time, right up until it doesn’t matter anymore — because future buildings will increasingly be designed for automated construction and automated maintenance from the start, the same way modern factories are designed around the robots that will run them, not around a robot squeezed into a hundred-year-old floor plan.
I’m not just theorizing about this. A few weeks ago, I used an AI agent to design a printed circuit board — generating the netlist with Skidl and driving the layout through a KiCad tool integration — for a small soldering-practice project. I’ve spent decades doing this kind of work by hand, and I still didn’t need to be the one operating the tools directly; I prompted, and usable design work came out the other end. That’s the kind of task that used to require real specialized skill and practice to do well. I intend to use the same workflow for EduBot’s own control board, and likely for most of my PCB design going forward. This is a small example, but it’s exactly the kind of “clean, safe, interesting, skilled” work that the Dirty/Dangerous/Dull framing implies should be safe from automation — and it isn’t.
But even granting that these trades stay protected for a good while, the more important question is: what share of total employment do they actually represent? A relatively small number of “protected for now, for structural rather than technical reasons” trades doesn’t change the picture for the vast majority of jobs, which don’t have that kind of physical/legacy- infrastructure shield. That’s the actual, expanding concern — not that literally every job disappears, but that the number of jobs genuinely insulated from this wave is a shrinking minority, and shrinking is enough to be the problem this project exists to address.
I’m not asking you to simply take my word for it. If there’s a large, durable category of human work that general-purpose AI and physical automation genuinely can’t reach — not “hasn’t reached yet,” but structurally can’t, and one that accounts for a meaningful share of total employment — I’d like to know what it is. I haven’t been able to identify one at that scale, and that’s exactly why I think building something like this, now, matters.