Who owns the AI economy, and why does that decide who wins and loses?

He means the same technology produces opposite outcomes depending on who holds the title to it. His NickBot 2000 example gives one worker a robot that does his job, so he keeps his paycheck and gains forty free hours a week. Hand the identical robot to the boss instead and the worker is offered a dollar a day. Wolfers says that makes AI a question of law and public policy, not just engineering.

Because for roughly a century machines substituted for brawn, which pushed down working class wages while rewarding the college educated. Wolfers calls AI a cognitive revolution, so this time the people being substituted for are the ones doing thinking work. He says he now feels like a Detroit auto worker in the 1970s, watching the machines arrive on his side of the line.

Wolfers says nobody honestly knows yet, and he counsels humility. His example is the VCR, which everyone assumed would kill moviegoing and instead made people more interested in movies. A complement helps you do your job better, a substitute does it instead of you, and the same technology can look like one and turn out to be the other.

Competition, at least for now. Wolfers points to OpenAI, Anthropic, Meta, DeepSeek and a long list of others genuinely competing, which pushes prices down toward marginal cost. He says the marginal cost of a thousand queries might be about a cent, prices are extraordinarily low, and they have been halving every few months. His worry is what happens if that competition stops.

Because competition has to exist at every essential input, not just at the top. Wolfers describes a world with fierce competition among AI labs and competition in the labor market, but a single supplier sitting in between them. If only one company sells the chips, it can name an unbelievably high price and capture the value the whole chain creates.

Wolfers refuses to call it. He says anyone confidently spotting bubbles who is not already on a private island is overconfident. He draws the dot com parallel both ways: the bet that the internet would be transformative was right, while pets.com was absurd on its face, and Amazon went from bookseller to dominating global commerce.

He did it out loud as napkin math, and invited listeners to swap in their own numbers. He assumed AI could do a quarter of all work in a 30 trillion dollar US economy, which is 7.5 trillion a year, minus maybe 500 billion in costs. Discounting that forever at 5 percent multiplies it by 20 to reach 140 trillion for the US, and treating the US as a fifth of world output gets to 700 trillion. During the episode Nick looked up the Magnificent Seven at 20.3 trillion.

He says the United States currently has no AI policy, and that the absence of a policy is itself a policy, namely laissez faire. Wolfers says that default makes complete sense for something like the market for t-shirts. He does not think it makes sense when the stakes include monopolization and enormous shifts in the distribution of income and work.

Wolfers raises this as an open worry. He says he currently believes what an LLM tells him, then asks whether he should, and whether small nudges on a few topics could be nearly undetectable while still moving opinion at scale. His concrete complaint is institutional: statistical tendencies could be measured, but nobody has set up the detectors or the institutions to check.

Wolfers says it depends on market structure more than geography. Speaking from Australia, he says that if he can buy the robot at a reasonable price it transforms his economy, and if he can't, his country gets poor while America gets rich. He also notes the number of people working in AI is tiny, and that even a Chinese winner might still need Nvidia.

Wolfers points to history. The Industrial Revolution took decades, electrification likewise, and the PC revolution took decades to appear in the economic data because work itself had to be reorganized. Using that as a guide, he says AI could be big but on a two to three decade horizon, which makes speeding up the adaptation the real economic prize.

Wolfers says the question is already settled in practice and schools are failing to rise to the moment. He notes that a student applying to Harvard writes four admissions essays, and asks whether the process now rewards the cheats over the imperfect honest applicants. He is testing an alternative in his own classes, an AI run Oxford style tutorial that asks follow up questions instead of setting a copy and paste prompt.