Man Bot

The first things I was paid to design were CD-ROMs. Business titles, training courses, the sort of thing that shipped on a disc in a jewel case. Websites alongside them, back when a website was a stack of tables and a few images you apologised for the file size of.

What made it exciting was not the discs. It was that the PC and the early internet had just put publishing within reach of ordinary people. You no longer needed a press, a broadcast licence, or a specialist to set your type for you. Anyone could get to the thing, anyone could put something on it, and neither side of that needed deep technical knowledge any more. If that sounds familiar, it should. I am aware I am describing the argument people are having about AI right now, thirty years early, and I am aware that the people making it then were both right and much too relaxed about who was going to pay for it.

Because then the web ate the CD-ROM, and it did not take long. Inside about three years the interactive multimedia industry was simply gone — the studios, the specialists, the people who had spent a decade getting genuinely good at authoring for a disc. Most of them were not offered a seat at whatever came next. I moved, largely because I was young and cheap enough to be allowed to. Several people I worked with were neither, and that was the whole difference.

That is the part that usually gets left out, and leaving it out is why people stop listening. A shift can democratise something, produce more work in total, and still destroy the livelihood of the person standing closest to it. All three at once. The rest of this essay assumes you can hold them together.

The example everyone reaches for is the bank teller. Cash machines arrived, and teller numbers per branch fell — and then banks opened more branches, because a branch got cheaper to run, and total teller employment went up for about two decades. It is a lovely story. It is also not the end of the story: what finally took the job was not the machine in the wall but the one in your pocket, arriving thirty years later. Today it has the lowest score on this site, 13.

This is the single idea that makes everything else on this site readable, so it is worth saying slowly. Almost nobody's job disappears. What happens is that the tasks inside it get redistributed — some to machines, some to other people, some to customers — and what is left is still called by the same name even though it is a different job.

Here is my own week, three times over. In 1999, well over half of it was production — making the artefact, by hand, slowly. Today it is about a fifth. I lead a team now, so most of the making belongs to them — but a fifth of my week is still my hands on the work, and I defend that on purpose. The rest did not come back as free time. It went into deciding and into talking: working out which problem is worth solving, and then persuading a lot of people that it is.

Notice what that does to the value of the job. The part that shrank was the part I could be taught in a year. The part that grew was the part that took me a decade to be any good at, and it is almost entirely made of judgement and other people. That is not automatically good news, and chapter three is about why.

The third bar is the one that stopped me. My daughter's first year looks very like my 1999 — heavy on production, light on everything else. That is not a coincidence and it is not a criticism. It is what the beginning of this job has always looked like: you earn the judgement by doing the production. Which raises an obvious question about what happens when the production is the first thing to go.

A software developer is the clearest version of the same pattern: the most assisted occupation on this site, and one of the more exposed, at the same time. Those two numbers are kept separate everywhere here, because collapsing them into a single "exposure" figure is how a helpful tool gets reported as a redundancy.

I should say what this chapter is not. I am telling this from my side of it, because it is the only side I have. She has read this and told me what to cut. Where I have guessed at what she was thinking, I have taken the guess out. A parent watching a twenty-three-year-old is not a reliable narrator about a generation, and I am not going to pretend the vantage point is better than it is.

What I saw was this. She started in product design, and within a year she had moved towards fashion — a small independent label, the kind where the person who designs the garment is also the person deciding how it is cut, what it is made of, and whether the sample that came back is right. I had opinions about that at the time. I kept most of them. The reason I am writing any of this down is that when I finally went and looked at the numbers, the move she made was the one the data would have recommended, and I had spent a year being quietly worried about it.

I want to be careful here, because this is the point where a parent starts explaining a decision that was never theirs to explain. I do not know why she moved. I know what she moved to, and I know what the data says about it, and those are different things from a reason.

The thing I do have standing to talk about is the apprenticeship problem, because I have been on the hiring side of it. Juniors have always learned by doing the work nobody senior wanted: the document review, the third variation, the tidy-up. That work is exactly what is being automated first. If you remove the bottom rung, you do not get a shorter ladder. You get a ladder nobody can get onto.

I do not have a solution. I have noticed that the firms worrying about it out loud are mostly the ones who cannot currently afford to fix it, and that the ones who could are not the ones worrying.

Here is every occupation on the site, placed twice over: left to right by how much of it a machine can already do outright, and bottom to top by how much of its physical work robotics still cannot reach.

That second axis is doing more work than it looks. Being physical is not, on its own, protection — it is only protection for as long as no machine can do it. So the height of a point is physical work minus whatever robotics has already got to.

The top-left corner is the interesting one. That is work which is hard to automate and still out of a robot's reach — and it is where the trades sit, almost alone. A plumber scores 82 on this site, higher than every professional occupation on it. Not because plumbing is simple, but because a crawlspace under a terraced house is a harder engineering problem than a contract review, and always has been. Truck driving is just as physical and sits nowhere near it.

This is Moravec's paradox, and it is about fifty years old: the things we find hard, machines find easy, and the things any four-year-old can do — pick up an unfamiliar object in a badly lit room — remain genuinely difficult. We built the reasoning machine long before we built the hands.

And there, near the middle, are the two roles this essay is about. Product design at 55. Fashion design at 65. She moved up the map, and the reason is cloth: a thing you have to touch to judge, on a body that is never the same twice.

Nobody knows how fast this goes. Anyone who tells you they do is selling something, and the honest response is not to refuse the question but to show your working. So here are three stated assumptions applied to the same map. Press one and watch what moves.

Under slow, very little rearranges. Under fast, two things happen at once: the desk jobs slide right, and the physical jobs fall. Watch truck driving drop off its shelf while plumbing barely moves — same amount of physical work, completely different exposure, because doubling robotics reach gets you a motorway but it still does not get you into that crawlspace.

Uneven is the one I find most plausible, and it is the least comfortable. It assumes adoption concentrates where it has already started — that the jobs already heavily assisted get transformed quickly, and everywhere else barely moves. That is roughly how the last four shifts went. It also means the disruption lands hardest on people who thought of themselves as safe because their work was cognitive.

Truck driving is the clearest test. It is the one occupation here where the robotics number, not the language-model number, decides the outcome — and it is the one that moves furthest between slow and fast.