A Brief History of Trying to Replace the Human
June 24, 2026
Robert Shiller published an essay this week arguing that the fear around AI is a bigger threat than the technology itself. He won a Nobel Prize for showing how stories move economies, and Narrative Economics is the long version of the case he makes here. Narratives shape what people do with their money. When millions of people expect the worst, they stop spending and they stop hiring, and the expectation starts to make itself true. He traces the pattern back through the Luddites, through Brave New World, through the Automation Recession of the late fifties, all the way to the singularity talk that runs today. His read on the Great Depression is the sharp part. The crash itself reached almost nobody, since only about two percent of households owned stock at the time. What deepened it was a collapse in consumer spending driven by uncertainty about the future, in a public soaking in stories that told them to be afraid.
He is right, and he names the source correctly. A great deal of the AI doom is coming from the people who build AI. Shiller closes by appealing to Silicon Valley to stop, because the attention that comes from warning how dangerous your own model is helps you sell it, right up until the recession that the fear helped create.
There is a second turn he does not make, and it is the one that matters most for anyone actually putting this technology to work. The doom narrative is not only depressing how people spend. It is shaping how the technology gets built. If you believe AI replaces people, you build it to replace people. You point it at the human in the workflow and you try to remove them. Then it fails, because the value was never in the removal.
Look at where the money went and what it bought. MIT's NANDA study last year found that ninety-five percent of organizations deploying generative AI got no measurable return on it, against thirty to forty billion dollars spent. The report puts the cause on the tools, which do not learn the work and do not connect to how a business actually runs. Separate consumer data shows roughly eighty percent of people getting nothing real from the AI they touch. This is the productivity paradox running again, the same one Robert Solow named in 1987 when he said you could see the computer age everywhere except in the productivity figures.
This is what fear builds. Aim the technology at replacing judgment and you get a tool that guesses in the spot where a person used to decide. The result comes out worse than the person and more expensive than the old way of working. Hammer and Champy ran a version of this in the nineties with reengineering. Imposed from the top down, it failed by their own count somewhere between fifty and seventy percent of the time. The instinct to remove the human is old, and it has a track record, and the track record is bad.
The alternative is not a slogan. I have been living it. My career trajectory was decades owning and operating small businesses, not writing code; AI changed that trajectory. Over the past several months I designed a full software architecture, worked through the engineering tradeoffs with a CTO who builds these systems for a living, and produced a financial model that holds up to scrutiny. I had the idea. What I did not have was the technical training to carry it the distance from a concept to a built thing. AI closed that distance.
I drove the thinking; I corrected it constantly. I threw out every answer that was merely plausible and held my own frame against its defaults more times than I can count. The output is good because I refused to let it settle for average. That is the part no diagram shows and no model supplies on its own. The judgment stayed mine. The reach got extended.
That is augmentation, and it is the whole game. The technology earns its keep when it sits under a person and gives them more range than they had alone. It breaks when it is sent in to stand where the person was standing. One direction compounds, because the person keeps getting sharper and the tool keeps extending them. The other direction walks straight into the ninety-five percent wall, because a tool with no stake and no purpose of its own cannot carry a decision that needs both.
Shiller wants the fear to stop because of what it does to spending. I want it to stop for a reason that sits one layer underneath that. The replacement story is more than bad economics. It is a design brief, and right now it is only writing one half of the software. Every dollar aimed at removing people is buying the failure the studies already measured. LLMs are only one side of the same coin. Put the person in charge with the machine underneath, extending their judgment rather than standing in for it, and you get the thing the fear keeps insisting is impossible, which is a person with more reach than they ever had on their own.
