Most of the jobs you and everyone you know do right now didn't exist in 1940. David Autor, the MIT economist, ran the numbers in a 2024 paper in the Quarterly Journal of Economics and found the majority of current US employment sits in job categories that were invented after 1940. Not a few. The majority. Every time a machine retires a bit of old work, the same machine coughs up new work nobody had a name for yet. We keep doing this dance and we keep forgetting we've done it.
The looms are the cleanest example. The Luddites smashed the textile frames in the 1810s because they were dead certain the machines would end weaving work forever. By 1830 British textiles employed more people than before, just doing different things. Mokyr and his crew laid that out in a 2015 review, and the pattern holds. The work didn't vanish. It moved.
So here's the position i'll actually defend. AI reshapes where the effort goes. The grunt shifts onto the machine and you're left holding the part it can't touch, which was always the point of the job.
Run a cafe long enough and you learn this in your hands. The machine pulls the shot and holds the temp. Dials the grind tighter than a tired human ever could at 6am. Gets you most of the way there. But the reason someone drives past four other cafes to sit at yours is the bloke behind the counter who remembers their kid's name and their usual and clocks when they've had a rough week. No machine buys that. Thats the last stretch, and its all human.
That's the whole shape of it. AI gets you about 80% of the way for about 20% of the effort. The draft, the first pass, the numbers, the grunt. Then you pour your full effort into the last bit, the polish and the feel, the finish someone actually notices. And the finished thing comes out better than either of you could've done alone.
The data backs the shape, with caveats. The World Economic Forum's Future of Jobs 2025 report, which is employers telling you what they expect, projects 170 million new roles created and 92 million displaced by 2030. Net gain of about 78 million. Same report reckons around half of the affected tasks get augmented rather than automated outright. Means a machine sitting next to the worker making them quicker, keeping the job on the desk. Erik Brynjolfsson's 2025 work on AI in real workplaces found the same thing on the ground. The tools raised what people could produce. They lifted the floor instead of clearing the room.
Now i'll give the other side its due, because pretending there's no risk is how you sound like a salesman. Every wave of automation before this one came for narrow work, mostly physical, mostly one task at a time. This one comes for broad thinking across a thousand tasks at once. So "this time is different" isn't a crazy thing to say. It might be. I don't know for certain and neither does anyone claiming they do.
But the bet i'm making with my own business, embedding these tools into small operations every week, is that the human part gets more valuable, not less. The touch, the read of a room. The judgment call you can only make because you've stood in that spot and felt it go wrong before.
The machine does the 80. You do the 20 people actually remember. Same job it always was.