Luis Garicano on AI, Messy Jobs, and Spain

Spain has the sun of Texas, the coastline of Florida, and a life expectancy that ranks fifth highest in the world. It also has the fifth lowest fertility rate on the planet and a science Nobel Prize drought stretching back to 1905 (12:58). Ask Luis Garicano, a Spanish economist who taught at the University of Chicago, served in the European Parliament, and has spent his career tracing how institutions bend incentives, and he explains the paradox less through talent or luck than through a simple question: who gets to vote, and who does the system actually serve.

Start with housing. Before the 2008 financial crisis, Spain built up to 800,000 houses a year (01:22), an extraordinary pace for a country its size. The boom turned into a bubble, and the bubble turned into a scandal that tied local politicians, developers, and savings banks together in corruption. The public drew a lesson from that, but arguably the wrong one. Instead of blaming weak oversight, many Spaniards came to associate building itself with graft.

"People identified easy building with corruption and people identify easy building with high prices." Luis Garicano, [02:09]

So the country passed rules stripping municipalities of much of their power over land use, on the theory that less discretion means less corruption (02:47). The result is a housing supply that barely moves. Spain now builds under 100,000 houses a year while absorbing hundreds of thousands of immigrants annually (02:47). Drive through the country and much of it sits empty, ringed by biodiversity rules and a green belt around Madrid that recalls London's. Barcelona residents insist their city is full even though the land around its airport sits fallow. Garicano's shorthand is that an anti-corruption reform aimed at politicians ended up disempowering the very people who could unblock supply.

The demographic math compounds the problem. Spain has gone three decades without meaningful productivity growth (04:37), while its population skews old, with the fifth lowest fertility rate and the fifth highest life expectancy in the world (05:04). As Garicano puts it, essentially all of the country's GDP growth since 2008 has gone to pensioners (05:04). High-speed rail lines and highways built in better decades now go without the maintenance they need (05:28), not because Spain lacks money but because the money has a prior claim. A country can have cheap solar and wind power, empty land for nuclear plants, and excellent weather, and still stall, because the voters who show up are old, comfortable, and unthreatened by the status quo. Cowen calls it a kind of resource curse, and Garicano agrees: a place with almost nothing wrong day to day has almost no appetite for the disruption that fixing anything requires. Real reform in Spain, he notes, has historically arrived only through crisis, in 1959, in the early 1980s, and again in 2012 (07:05).

A domestic version of the same dynamic plays out between Madrid and its regions. Because the Basque Country once had a violent independence movement, it won the right to collect its own taxes and negotiate its contribution to the national budget directly with Madrid, a deal Catalonia now wants too, though extending it to a region as large as Catalonia would break the national budget (17:44). Garicano's point is not about grievance. It is that the deal was cut under threat, and threats, unlike need, tend to get rewarded. The same logic scales up to Brussels. The European Union requires unanimous agreement on anything touching the core functions of a state, taxation, defense, and foreign policy, and majority rule almost everywhere else. In practice, one country's veto can block a shared defense procurement system, so European militaries end up building slightly different tanks and incompatible artillery shells rather than sharing one design. Garicano, who served as a member of the European Parliament, also noticed something about the kind of politician the system rewards: parties promote the most loyal candidate, not the sharpest one, because a candidate whose entire career depends on the party has no incentive to break ranks (20:57). That is one reason, he argues, the presidency of the European Commission has gone repeatedly to people chosen precisely because national leaders did not want someone in Brussels capable of overshadowing them (21:20).

What survives the machine

Garicano's new book, Messy Jobs, co-written with Jim Li and Yanhui Wu, turns the same institutional eye on a different question: what happens to work when artificial intelligence gets cheap. His answer cuts against both the doom and the reassurance usually on offer. Look, he says, at what happened during the 2000s offshoring boom, when companies shipped call-center and back-office work to India. The jobs that moved were the ones a manager could specify completely: a single task, a clear procedure, an output that is easy to check. Those are exactly the jobs most exposed to AI today, because a task that is easy to verify is also one a reinforcement-learning system, a method that improves an AI by scoring its output against a target, can be trained to grade and refine (36:35). What survives is messier work: jobs that bundle several tasks together, that involve reading a room or making a judgment call nobody wrote down in a manual. A relationship manager at a bank can use AI to prepare for a client meeting, pulling balance sheets and flagging products to pitch, but still has to sit across from the client and manage the relationship (40:51). The AI enters the bundle. It does not replace it.

That protection carries a cost. Retraining, in Garicano's account, mostly does not work, and not for purely technical reasons. A doctor whose diagnostic work gets automated cannot simply become a plumber, not because plumbing is unlearnable but because a job is bound up with identity, with who a person understands themselves to be (37:57).

"The way we retrain ourselves, as Kuhn said, talking about scientists, is by dying, and the new generations do new things." Luis Garicano, [38:21]

It is a grim line, but consistent with his skepticism about formal retraining programs. What he finds more promising is narrower: teaching people not new careers but new use cases, like giving a bank's relationship managers a short list of questions an AI can help answer before a client meeting (39:10). His own 84 year old mother, he notes, has picked up ChatGPT for long conversations with no formal training at all (38:43). Increasingly, the barrier is not technical skill but knowing what the tools can actually do.

Europe's second-mover bet

Given how far behind Europe sits on computing hardware, on frontier language models, and on data centers, with the Dutch chip-equipment maker ASML as close to an exception, Garicano does not think AI sovereignty, meaning control of the entire technology stack from chips to models, is a realistic goal for Spain or for the continent (45:16). His prescription is a second-mover strategy: adopt AI aggressively, become a large enough customer that vendors lobby their own governments to keep selling to you the way China did with the chipmaker Nvidia, and insist that models stay interoperable so switching providers is as simple as changing an API call (46:39). The advantage of open-weight models, AI systems whose underlying parameters are published rather than kept secret, is that a company or country can run them locally and keep its own data and know-how in-house, even while trailing the technical frontier by a few months (48:05).

The through-line across housing, pensions, party lists, and AI is not that Garicano is a pessimist. It is that he keeps finding the same mechanism: whoever already holds power, whether pensioners, party leaders, or legacy carmakers, tends to capture the rules meant to fix the system, so the fix ends up protecting insiders instead of solving the problem. What makes his messy jobs argument notable is that it locates a kind of protection that does not depend on anyone lobbying for it. Messiness, the tangled, relational, hard-to-specify part of work, shields ordinary people the way incumbency shields politicians, except nobody had to arrange it that way.