AI Can't Replace the Plumber, but It Will Take Out Everyone Sitting at a Computer
Fukuyama says AI's bottleneck isn't intelligence, it's the physical world: nobody in Dhaka lacks the knowledge to build a water system, the water mafia won't let them. What actually gets taken out is the middle layer of people sitting at computers.
The video won't play here. Listen to the audio instead:
The argument · timestamps estimated from transcript position
Making furniture is the antidote to big tech dependence
Fukuyama elevates his hobby into a question of ‘individual autonomy’: we live in a world dominated by Meta, X, Apple, Amazon and Google, and being so-called tech-savvy just means knowing how to use their software — it doesn't penetrate any deeper. What he hates about Apple is that it deliberately stops you from opening your own machine: even opening the computer requires Apple's proprietary fastener, it isn't modular; a PC you can take apart. When the OS on his big home PC recently broke, he took the whole thing apart, reinstalled the system, put it back together, and it worked again. He worries the AI era will raise a whole generation that can't do anything without AI, the way most young people today can't read a map — if GNSS went down, they wouldn't know how to cross the city to visit a friend.
— Francis FukuyamaHis survivalist template is Ukraine
Fukuyama says the world state in a lot of survivalist scenarios — Cormac McCarthy's The Road, Octavia Butler's Parable of the Sower — amounts to a return to the cave age, and nobody can prepare for that. The more realistic case is Ukraine's situation after Russia invaded in 2022: suddenly facing a far stronger enemy, able to rely only on their own brains, and it turned out they were very good at robots, drones, and standing up command-and-control systems — and they built it all themselves. He says that's roughly his own technical level: he can wire sensor inputs to an Arduino and a Raspberry Pi, drive servos, and has built a pile of robots (not the lethal Ukrainian kind). Drop him into a situation where he has to work with that tier of technology himself, and he can do it.
— Francis FukuyamaThe path to learning any craft is YouTube
Fukuyama never took a woodworking class or a robotics class; it was all self-taught. His furniture making began in the 1980s, before YouTube existed — it was PBS's This Old House, and the carpenter on that show had his own furniture-making program; watching him build simple tables and desks is how he got started. He says YouTube is simply incredible: there is no hobby so obscure that it doesn't have a whole set of instructional videos. The visual part is what matters, because with any hobby that involves handling physical things — fixing cars, fixing computers, making furniture — if you don't actually watch someone do it, you won't understand how it works.
— Francis FukuyamaThe price of using Claude Code is skill atrophy
Fukuyama used Claude to migrate several large databases: years ago he wrote a database program himself in Python, without SQL, which he now sees was a dumb move; migrating it to a more robust open-source product was only possible with ChatGPT and later Claude. He says he has written thousands of lines of code, and he used to always get stuck at the point of ‘I want to do something but I can't figure it out even digging through the manual’; now with Claude Code he just says add a button that does such-and-such, and it does it, he runs it to see if it works, and if it doesn't he says ‘can you fix it’, and that's that. But he sees the danger: he has already forgotten a lot of the programming he once knew, because he no longer needs those abilities. He also admits software skills come back — the way he once drew furniture plans in AutoCAD, built a Georgian house in software with 3ds Max, tuned the lighting to watch morning light come through the windows and fall on the furniture, and could even make a video of walking through the house; he can't do that now, because you have to be immersed in that highly complex software.
— Francis FukuyamaWhat outsourcing threw away was tacit knowledge you can't say out loud
Fukuyama thinks the loss of American manufacturing capability didn't have to happen, and the blame lies with economists: the outsourcing frenzy was driven purely by efficiency — if someone else can make a mouse more efficiently, why not buy it? They didn't understand the tight link between high-level skills like designing a computer or writing complex software and the low-level skills that turn it into a physical product, involving complex supply chains, social relationships with suppliers, and a great deal of basic engineering capability grounded in tacit knowledge — once it's gone, it's gone, and it's extremely hard to rebuild. He gives the example of tightening a bolt: loosen the bolt on a gasket cover and too much force strips the thread, too little and it won't come off — it's all feel, and a good mechanic knows how. Japan and Germany preserved these mechanical skills longer than the US — Japan has Living National Treasures, about a hundred people socially honoured for a particular skill, some cultural like the tea ceremony, others machinists who can make extremely complex parts to high precision.
— Francis FukuyamaGovernment is the only organisation that can do a moonshot
Fukuyama traces ‘can't build anything’ back to the political turn that began in the 1970s and 80s: in the Progressive era and the New Deal, Americans had strong confidence in their own government; an ambitious young progressive in the 1930s would want to go to the Roosevelt administration and run a federal agency, electrify the upper South through the TVA, or be an engineer building the Hoover Dam; by the 1960s there was a backlash, and Reagan fixed it in his inaugural address as ‘I'm from the government, and I'm here to help’ being the scariest words. He concedes the backlash had a point — Robert Moses was cast as a villain for tearing up communities, and many agencies really were captured by corporate interests — but the result was a general abandonment of government, and government is the only social organisation capable of undertaking a project like the moonshot. Today people who want to do social justice graduate from Yale or Stanford Law School and don't go into agencies; they go to public-interest law firms to sue the government and stop it from doing things.
— Francis FukuyamaAI's weakest link is that it can't connect to the physical world
Fukuyama says the weakest part of the AI revolution is the interface between LLMs and the real physical world. He cites Musk's interview with The Economist's Zanny Minton Beddoes: asked what money will be like in ten years, Musk said people won't need money because everyone will have everything they want — Fukuyama calls the idea absurd, because it's completely detached from the material preconditions of economic growth; close the Strait of Hormuz and the entire global economy staggers, because it depends on oil and gas as a physical resource. He thinks a lot of the AI speculative frenzy comes from very smart people who are only smart in a mathematical sense, and who overestimate intelligence's own capacity to solve every problem. He heard an OpenAI engineer talk about AI ending global poverty, every village in every developing country having clean water — and he has taught cases on water supply in South Asia: in a city like Dhaka, what good water supply looks like isn't rocket science; it doesn't get built because a water mafia controls the existing supply and makes big money from it, and the Bangladeshi government is too weak to enforce the law against them. How does a smart machine solve that? It's a purely political constraint, and no amount of intelligence gets around it.
— Francis FukuyamaAI will take out the middle layer, and that's bad for democracy
Fukuyama thinks AI will be far more transformative than the internet, because it's a general-purpose technology usable in almost every aspect of a modern economy; and most of the wealth in a modern economy like America's isn't generated by physical interaction, it's generated by people sitting at computers providing cognitive services, and these machines will replace them. He says many of the people pushing the technology downplay the impact, because it will take away a lot of people's livelihoods. He judges AI's immediate effect to be economies of scale: giants like Meta and Google exist because they're software-based, and software scales more easily than manufacturing; AI is slightly different in that it has a physical substrate — you have to build large data centres — but the returns flow to the people at the top of the corporate hierarchy building these systems, while taking out the livelihoods of the middle layer (the bottom still needs plumbers and electricians). That inequality and concentration of wealth is bad for democracy, possibly to the point where the state itself can no longer check these private companies. Cartels and monopolies like the Standard Oil Trust at the end of the nineteenth century triggered a political backlash and antitrust, but he isn't sure future governments will still have the power to control these big companies.
— Francis FukuyamaIn their own words · checked verbatim
I think that dependence on big tech is a problem for society as a whole. I do have this worry about the coming AI age, because we're already getting very dependent on these AI systems, and I think as time goes on, we could raise a whole generation of people who basically don't know how to do anything if they don't have access to an AI.
Francis Fukuyama4:30
It also has to do with basic engineering skills, because a lot of that stuff is based on a kind of tacit knowledge, and once you lose that tacit knowledge, it's gone, and it's really, really hard to recreate.
Francis Fukuyama21:00
I really think a lot of the speculative fervor behind AI is driven by people who are very smart, but in one rather narrow way—mathematically smart—and they overvalue intelligence, by itself, as having the ability to solve all problems.
Francis Fukuyama33:00
It's simply a political constraint that no amount of intelligence can get around—the intelligent machine will say, boy, this is a really stupid political system , but it's not going to have a way of getting around it.
Francis Fukuyama33:00
I think people are understating, for example, the job loss—the threat to service-sector jobs from these machines is going to be huge. I think a lot of the people pushing the technology are trying to soft-pedal the impacts it's going to have, because it is going to take away the livelihoods of a whole lot of people in society.
Francis Fukuyama39:00
That kind of inequality and concentration of wealth will get you to a point where the state itself may not be able to counterbalance the power of these private companies.
Francis Fukuyama39:00
Figures
| Number of people honoured as Japanese Living National Treasures | about one hundred | 21:00 |
| Global population | 8 billion | 33:00 |
Glossary
- tacit knowledge
- Feel and experience that can't be taught in words, only accumulated by doing the work.
- Living National Treasures
- A Japanese designation honouring holders of a particular craft, from tea ceremony to high-precision machinists.
- vetocracy
- A system in which any party can sue or stall, making it hard to get projects done.
- GNSS
- The umbrella term for satellite positioning systems such as GPS.
How to listen
Founders, investors and policy researchers who care about where AI meets the real economy, manufacturing reshoring, and the prospects for the regulatory state and democracy.
The first ten minutes or so of furniture and hobby talk — fast-forward to after 21:00.