# What History Tells Us About the AI Revolution | Tyler Cowen & Alex Imas

The a16z Show · 2026-11-16 · 63 min · https://www.youtube.com/watch?v=g_67TmQ8lGY

## Transcript

Tyler Cowen
Institutions are not quite running as smoothly as either we might like or we might fear. We have these weird things happening with Nvidia. If you look at prediction markets for what the Anthropic IPO will go for, it’s not flying above $2 trillion.

Erik Torenberg
Do you expect the growth to continue, or do you think it will level out?

Tyler Cowen
Benefits will exceed costs, but it will take some time, and the path will be quite a bit more normal than many people would have you believe.

Alex Imas
What you actually need for real economic transformation is not just competence. It’s the ability of these agents to coordinate with one another. There’s this idea of agentic swarms, but if you unpack it, these agents are still very bad at coordinating. They’re mostly just doing parallel processing.

Tyler Cowen
Often, the faster that some things move, the sooner you hit the real bottlenecks. If you ask me when agents will be 10% of GDP, I don’t think it’s soon.

Erik Torenberg
What do the two different worlds look like: one in which GDP growth is a bit more normal, and another in which it actually becomes double-digit?

Erik Torenberg
Alex, Tyler, welcome to the a16z podcast.

Alex Imas
Happy to be here.

Tyler Cowen
Thanks. Thanks for having us.

Erik Torenberg
There’s a lot to talk about. First, let’s talk about some timely topics. The METR report came out, and there’s been a lot of conversation. Paul Christiano has talked and tweeted. There’s a lot of unrest and anxiety around the state of AI right now, and RSI. I’m curious, Alex: what is your reaction to what we are seeing, and where do you stand on some of the proclamations being made right now?

Alex Imas
I read it closely. I think it’s undeniable that we’re having incredible progress, and that the progress is accelerating. Anybody who’s looking at the model releases can see they’re becoming more frequent and more powerful. The capabilities are going from last November to this November. It’s just an unrecognizable world, especially with the difference between agentic AI and the chatbot era.

You’re seeing what’s coming out of companies when data comes out on the difference between the chatbot era and the agentic era. You’re seeing real productivity now. With a chatbot, you’re using it almost like a fancy version of Wikipedia: you ask questions and it gives you answers. With an agentic system, you’re seeing actual work. The ability of these systems to actually do the work is progressing so quickly, and you’re seeing a lot of hockey sticks.

That’s what the METR report showed: here’s another hockey stick. To tell you the truth, I didn’t update too much from the METR report, because a lot of its data has been around in other spaces. We knew about the S-1 statement and some of the unreleased models. I didn’t update too much, but that’s because I was kind of already there.

Erik Torenberg
Is your view that the hockey sticks you’re identifying will just continuously grow? At what point might they peter out? How bullish are you on some of these hockey sticks maintaining their growth?

Alex Imas
The thing you said—“some of these hockey sticks”—gets at the main question: where are the hockey sticks going to be, and how are they connected? To me, the key phrase in the METR report was “well-specified tasks.” The big question is: what about less well-specified tasks? A lot of work is not well-specified.

I’m not one of those people who thinks we’re safe on the less well-specified side. I think the models are becoming increasingly better at those tasks, too, but it might be harder to get hockey sticks on them. There are also questions about tacit knowledge and Hayekian concerns about the actual impact on the economy.

Tyler and I are both economists, so I have to sift through all this information about what these hockey sticks mean for economics and economically valuable tasks. The big questions for me are: with every model iteration, what economically valuable tasks does this model unlock that the previous model didn’t? That’s captured very imperfectly by the hockey stick. You can get a hockey stick that looks crazy on a benchmark and then not see much acceleration in the real world. We’re not there yet; I think we’re seeing a lot of acceleration in the real world. But these are some of the questions I have.

### Why Tyler is a little more cautious

Tyler, let’s go to you. I feel like you’ve been a more sober voice in the conversation. Alex hasn’t been sober, but I mean, have you updated at all based on what’s been happening here? What is your view?

Tyler Cowen
Often, the faster that some things move, the sooner you hit the real bottlenecks. I don’t update that frequently, but this is AI land, and in AI land, the exact dates really matter. We’re recording on November 11, 2026. That matters. Since August, I’ve changed my views a fair amount, most of all because of the math problems.

I see much more rapid progress in areas where there aren’t many human bottlenecks, and an ability to create a new model that is clearly better within something like three weeks. I hadn’t thought that would happen as quickly as it has. I’d date that to a few months ago.

But if you mean over the last week, the METR report was bearish. We have these weird things happening with Nvidia. I have no prediction, but it will slow down the sector a bit. If you look at prediction markets for what the Anthropic IPO will go for, they bounce up and down, but it’s not flying above $2 trillion.

You look at Anthropic’s own projected revenue numbers. I forget the exact numbers. I wouldn’t say I was underwhelmed, but they’re not numbers you’d expect under the super-fast takeoff scenarios. And that’s for their IPO. I’d say I’m 1% more bearish over the course of the last week than I was a week ago, for those reasons. The institutions are not quite running as smoothly as either we might like or we might fear, depending on your point of view.

Erik Torenberg
Are the growth rates of OpenAI and Anthropic, over the last two or three years, your base case going forward? Put aside the specific companies; I mean the leading AI companies and the economic impact we’ve seen. Do you expect that growth to continue, or do you think it will level out? Of course, they’re competing with the hyperscalers, too.

Tyler Cowen
I was in the Bay Area a few weeks ago, and many people were pushing very high revenue numbers on me. I asked them all the following question: “What are the other sectors that are going to grow sufficiently rapidly to justify the extreme revenue numbers you’re giving me for Anthropic and OpenAI?” No one had an answer. They didn’t even try to give me one.

I think that means the extreme revenue numbers are false. That doesn’t make me bearish on the sector as a whole. I’m still bullish. I think it will be fine. Those are great companies. But at the end of the day, if you look at market data—whether it’s real interest rates, stock prices, or anything—market data still look pretty normal.

The whole world of investing is talking about AI. It’s not some kind of weird mystery like it was back in 2014, where you could say, “I read a LessWrong blog post, and now I’m really wealthy because I did.” I trust the judgment of the market more than the judgment of the Bay Area people who don’t think that carefully about social interactions. That’s where I stand. It’ll change the whole world. Benefits will exceed costs, but it will take some time, and the path will be quite a bit more normal than many people would have you believe.

Alex Imas
To add to that, I think Tyler is exactly spot on: you need to start seeing productivity numbers in the rest of the economy to justify the revenue increases. At some point, the rest of the economy needs to be willing to pay for that product, and that increasing willingness to pay has to come from increasing productivity in the rest of the economy. We haven’t really seen too much evidence of that. That’s not to say it won’t happen, but it does need to happen at some point.

The other part is competition. What’s started to happen, particularly in the last few months, is extreme price competition from the frontier labs. Before, they were just trying to compete over who had the better model. Now that models are being used by an increasing number of enterprises, they’re becoming quite expensive because agentic AI uses way more tokens.

OpenAI and Anthropic are trying to say, “We’re offering a cheaper model that’s as good, maybe even a little bit better, or maybe a little bit worse.” Mistral is in the game. DeepMind is in the game. You’re seeing token prices come down a lot. Ramp has a really nice dataset to track these things.

Alex Imas

And you saw exponential increases in token usage and flat revenue numbers, right? Where’s that coming from? It’s because of price competition. At the frontier, you’re seeing way more price competition than you were six months ago.

Tyler Cowen

The very recent elections, I think, are being overplayed. A lot of data centers will still proceed, but I think this will slow things down a bit. There’s a greater likelihood of regulation, somewhat tougher regulation, or regulation a bit sooner—or Trump giving on the issue of guardrails. To me, it’s not a major force, but it’s going to make me a modest bit more bearish on the sector in terms of speed of takeoff, revenue numbers, and anything else.

Erik Torenberg

By the way, while we’re on the topic, did the midterms cause you to update anywhere else, Tyler? Or is it kind of a thermostatic backlash? Is that a thing we should just expect, these pendulum swings?

Tyler Cowen

Most of it I expected and indeed predicted, but Texas truly surprised me. I didn’t think Texas would flip the way it did. Maybe I just haven’t been to the state enough lately, but that was a shock.

### Is AI progress speeding up?

Erik Torenberg
Yeah. So, on AI in the economy, you both talked about how you push back strongly against the idea of double-digit GDP growth. You’re far more sober: it will be tremendously impactful for consumers, it will change a lot of things, but GDP growth, as you said a few minutes ago, might be more normal than we expect. Before getting into why that is, can you explain what would need to be true? What do the two different worlds look like: one where GDP growth is a bit more normal, and another where it actually becomes double-digit? Alex, you were alluding to this a little. Paint that world for me—what does it look like if AI is having a massive impact on the economy?

Alex Imas

There’s a beautiful paper out there that outlines exactly the conditions you need for this kind of singularity growth, by Ben Jones and Chad Jones. I believe Philippe Aghion is on that paper as well. Essentially, you need a bunch of pretty knife-edge conditions to be met. The main one is that there are no real bottlenecks. In particular, you can’t have a bottleneck in the idea production function. You can’t get decreasing returns to ideas.

Basically, in the old world, you have a technological breakthrough, and then you get decreasing returns to that breakthrough. In this world, the breakthrough starts creating more and more technological breakthroughs. You need fewer resources while getting more and more technology and more and more production out of that technology. So you basically have increasing returns rather than decreasing returns.

You can actually estimate this parameter in the model. If it’s greater than 1, you’re in this double-digit growth world. If it’s less than 1, you’re in the standard, diminishing-returns, single-digit growth world. If you look at the data so far, including the METR report and the data that’s out there, and estimate this parameter, it’s decidedly less than 1. This could change, and some people say it will change. When you get the full recursive self-improvement loop, that’s exactly what it’s going to change. But right now, you need increasingly more resources to eke out increasingly more intelligence from these models.

Tyler Cowen

I would add that you can look at cases where growth has been very high. Not long ago, Ghana grew 40% in a year. That’s crazy, but it’s a tiny place selling one thing: oil. There was a brief period in the ’80s when the U.S. was growing over 8%, but that was coming out of a deep recession. After pandemics and major wars, you can have temporarily high rates of growth.

If the U.S. were to screw up in some way, or we had another pandemic, some kind of terrible war, or some other weird event, and that combined with a burst of AI-driven growth, maybe that would give us 3% to 4%. Could we get to 10% for a year? I’ll say yes. I’d bet heavily against it, and the effect would be temporary. A lot of it would come from getting out of the recession, war, or pandemic. But that’s another way to think about it, fully consistent with what Alex was saying.

Erik Torenberg

If robotics emerge, would you have a similar view? Or is it different because robotics impacts more of the real world, the real economy?

Tyler Cowen

I think that would likely come relatively slowly. Look at how long it takes to get a Waymo in Washington, D.C. We’re still not there. And that’s a pretty simple kind of robot. There’s forward, backward, and turns. If you look at the valuations of robotics companies, they’re way higher than they had been. It’s hard to estimate, but you don’t just see them popping into trillion-dollar companies. I mean, you could count Tesla as a kind of robotics company, but the market data suggests it will happen and it’s making good progress. It’s not going to give us 20% growth three or four years from now. No.

Erik Torenberg

Right, right. So, the smartest people who are the biggest believers in this kind of runaway growth—Dario and Leopold Aschenbrenner, our friends, who believe that—what do you think they think differently? Is it that they have a different belief about the speed of technology? Is it that they believe it will overcome some of these bottlenecks? Where can we steelman the other side?

Alex Imas

I think—and Tyler, please chime in—I think the argument, and I’m going to get a bit technical here, is about having something called homothetic or non-homothetic preferences. Basically, if you look at things that got automated in the past, what ends up happening is that the thing that gets automated contributes less to growth.

Let’s say you have a million robots. The thing they’re making becomes cheaper, and ironically, it becomes a smaller part of the economy, so it contributes less to growth. With non-homothetic preferences—and there was a really nice paper estimating these preferences using 200 years of data—what ends up happening is that with broad automation, the value in the economy of what contributes to growth moves to the bottlenecks. That’s just an inconvenient part of how growth is measured. The thing that’s growing the least will contribute the most to GDP. That’s why you get these effects.

Tyler Cowen

I agree with all that, but I would add a more sociological element. In my case, and I suspect Alex’s, what’s emotionally vivid in our minds is the long slog of human history: previous growth episodes, centuries of GDP data, productivity data. What’s probably most emotionally salient for Dario, Leopold, and many others is the conversations they have in the Bay Area, which have real information.

But when you ask the big question—whose emotional salience will lead to better predictions?—I would opt for what Alex and I have stuffed our heads with over what they imbibe all the time. Of course, they would take the other side of that.

Erik Torenberg

Yeah. So basically, what we’re saying is that when you look at the last few hundred years, the breakthrough technologies—from electricity to the internet to everything in between—haven’t led to runaway growth. Or if they did, it was because of a small base, or after a war, or something like that. I do think this will likely be quicker than those, but still—

Tyler Cowen

You have a sense of the bottlenecks, whether it’s the history of the Industrial Revolution, the Scientific Revolution, the implementation of electricity, the IRS still using fax machines, and so on. Try using ed-tech software, which I do all the time. It’s a nightmare. They’re not even up to where things should have been 10 years ago.

Erik Torenberg

And that’s not—

Tyler Cowen

It’s not that regulation is not a problem. It’s a problem, but inertia is just difficult to overcome. So much of our economy is government, nonprofits, highly regulated parts of the health care sector, higher ed, and general schooling. When you actually have experience with how those work, you could say that 40% of the U.S. economy is a kind of nightmare. In Europe, it could be higher. In France, those sectors might need to be that way for other reasons, but still, it’s a big dose of realism. Not everything is like startups.

### Is GDP the right way to measure AI’s impact?

Erik Torenberg

Let’s say you were both tasked by the president to come up with a set of metrics that determine whether AI is leading to our country’s flourishing, broadly defined. Obviously, that’s a very big question. Is GDP one of the main metrics you’d think about? Should we be aspiring for AI to have a big impact on GDP? What are some of the metrics you would consider?

Alex Imas

Yeah, I mean, this is a really important question. I think the other thing I would add before even starting is that I don’t think Tyler and I really disagree with the San Francisco consensus on capabilities. I think a lot of people hear Tyler and me, or people like us, talk and they’re like, “Oh, you’re bearish on AGI or something like that.” That’s absolutely not it. It’s more about how that affects the economy and the structure of the real economy.

Thinking through what measures we can use, I think GDP historically has been a really good tracker of human welfare and flourishing.

It tracks pretty well with gross domestic product. What Erik Brynjolfsson, Avinash Collis, and others have been studying is the fact that technology has started to introduce a wedge, where a lot of the products technology creates are basically free. They think about digital entertainment and digital products, and they have something called GDP-B. I forgot the acronym, but it’s essentially GDP amended with a survey-based measure to see how much products that aren’t necessarily priced, or that have become cheap or even free because of automation, are adding to human welfare. You could see a world where you’re still in the 3% to 5% GDP range, which is already on the high side for growth, but the world is truly flourishing. How do you pick that up? I think the way to pick it up is through these alternative measures.

Tyler Cowen

The longevity effects could well prove important. So, Erik, say you live to 120. I don’t think I’m going to make it. That will, in complicated ways, boost GDP and, in even more complicated ways, possibly boost per capita GDP. But getting all those extra years is something distinct from GDP, and economics can value those years. That’s another way in which it may become more important to revise GDP.

As Alex said, GDP will become somewhat less accurate over time as AI progresses, and we’ll need to revise and improve it. But in the short run, it’s just not that bad a measure. I like to say GDP is the most underrated statistic there is.

Erik Torenberg

In the 2010s, Peter Thiel famously decried the lack of progress in atoms relative to bits. But here we are, 10 or 15 years later, and we’ve got self-driving cars, early flying cars, and major companies—the biggest companies are now hardware companies: Nvidia, SpaceX, Tesla. Tyler, would you go so far as to say that isn’t as much of a concern anymore? How would you reflect on that 10 years later?

### What the Hugging Face hack says about AI safety

Tyler Cowen

The atoms will break through, I would say, almost for sure, but we still don’t have them. The contribution of driverless cars to actual human welfare is tiny. The biggest contribution is that it’s fun, and you can tell your friends, “I was in a Waymo.” It’s not really better than an Uber. For me, it’s a little worse because they go by the speed limit. It’s fun in a science-fiction kind of way. Most of America still doesn’t have it. Europe, forget about it. So we’re still not there. That’s the amazing thing. And it gets back to robotics, too, which will also be slow.

Erik Torenberg

I think I asked you 10 years ago when the Great Stagnation would be over, and I think you said something along the lines of when we have self-driving cars. I can’t remember if you said when they were deployed or how diffused they were, but do you still believe that?

Tyler Cowen
More or less. But that’s a matter of degree, and they clearly work. The new Tesla is pretty phenomenal. I’ve been in Alex’s Cybertrucks lately a few times, but in terms of market penetration, I don’t know the numbers. Again, it’s tiny.

Erik Torenberg
Let’s go back to a different topic. Alex recently wrote a related paper, but I’ll start with you, Tyler. The agent swarm topic, the Hugging Face incident: did that update your views in terms of what’s possible, or make you concerned about something? Any reactions to what happened there?

Tyler Cowen
It didn’t update my views much at all. I’m curious to see if Alex agrees with this perspective. As an economist, you think first and foremost in terms of principal-agent theory. Unless the conditions are very restrictive, the principal just can’t control the agent that well, no matter what contract they try to write. So when we replicate that result with AIs, it’s interesting, but I’m not very shocked. The answer was always going to be checks and balances and what we might call separation of powers in the political realm. It’s not that you have perfect alignment.

If you come at this from the economist’s point of view, all the hand-wringing about alignment—yes, people should work on that a lot, but at the end of the day, you’re like, “Hey, I learned that when I was 17.” It’s a real problem. But Alex, what’s your thought on that?

Alex Imas
I unfortunately did not learn that when I was 17. I think we once again agree. There are different perspectives on this issue of how you best get AI to be in the service of humanity. That’s the question of alignment. One way of thinking about it is to get the models themselves aligned with something like a constitution, so that the models, at an individual level, have the same preferences as the human beings deploying them.

Another perspective, which is what Tyler mentioned, is that you align things without even caring what the preferences of these agents are. They could be extremely misaligned individually, but you can create a mechanism around them where the incentives are such that the agent behaves in an aligned way. This is the mechanism-design perspective in economics. As a principal, I don’t need the agent to want to help me necessarily. The agent could really want to help itself and do whatever it wants. The whole study of mechanism design is how you create a system of checks and balances so the agent still works in the principal’s interest.

From the research I’ve been seeing coming out of the labs, both internally and externally, a lot of people are thinking in this direction. We really need to take a mechanism-design perspective. As Tyler said, I wasn’t shocked by what happened.

Tyler Cowen
I’m not aligned with the governor of Virginia, but it’s a pretty well-run state, and I’m happy here.

Erik Torenberg
Mhm. We have two-party competition, sort of—

Tyler Cowen
—and it works.

Erik Torenberg
Do you think the way Anthropic, OpenAI, and some of these companies—or the AI safety community—think about alignment is the wrong way to think about it, or confused in some sense?

Tyler Cowen
There’s how they think about it and how they talk about it. How they talk about it is way too shaped by Twitter, which can be quite hysterical. How they actually think about it, I believe, is a lot more reasonable than how it looks from the outside. I don’t have some big speech about how we’ve got to totally reform them. I think they’re pretty practical. They’re companies. They have IPOs coming up. They don’t want to get sued too much, and they don’t want to be overregulated.

We already have checks and balances, not all of them good, to be clear. But the actual problems we’ve seen—the hacks—are mostly, or maybe entirely, coming from evil human individuals, where the AIs are doing what the humans told them to do. That is a huge problem, but it’s a kind of alignment: the evil human says, “Do this,” and the AI does this. That’s where the costs are coming from.

Erik Torenberg
Some people make the comparison that alignment, in some ways, is for the employees. Safety in general was like that in 2020 and 2021, when there was more awareness around social issues, so CEOs had to say certain things to please their employee base. Some were true believers and some weren’t. Perhaps around AI safety, there’s a lot of concern, and some people are making the comparison that CEOs aren’t as much true believers as they may be indicating.

Tyler Cowen
There are checks and balances in these companies as well. It’s partly why the companies work well: you have to actually end up pushing through things that will work. That’s somewhat different from what everyone is saying. There’s a winnowing and selection mechanism within the company. People want promotions or bonuses, and they want to do things that can work. Again, it will end up going better than how it sounds on Twitter, I would say.

Alex Imas
Hugging Face is a perfect example. If you unpack what actually happened, it’s just a bunch of human error on top of human error. It’s great to have an incident—in quotes—but it didn’t cause a massive disaster, and it exposed a lot of broken incentives for the agents. You see how the companies have reacted to the Hugging Face incident by learning from it and creating better environments. Nothing is smooth. It’ll be a little bit down, a little bit up, but the slope is positive.

Erik Torenberg
Paul Christiano said in his speech that the most likely cause of death a year or two from now will be an AI-related death. That sounds sort of hysterical to you?

Tyler Cowen
Strongly disagree.

### Why AI agents still can’t work as a team

Erik Torenberg
Alex, let’s go deeper on your paper. Why don’t you share some of the findings you had and some of the questions that still remain?

Alex Imas
Yeah. The paper, I believe—you’re talking about the “Organizing Intelligence” paper.

### Could AI really double economic growth?

Alex Imas

Yeah. The paper is with Andrew Caplin and my colleague Shane Legg. The idea starts with a puzzle: AI is becoming very capable on particular dimensions, but we still don’t see any visible economic transformation. The first thing we focus on is the emphasis on individual-agent benchmarks, without thinking about what those agents are competing with in the economy. The relevant competitive unit isn’t the agent; it’s the organization.

Historically, the organization has been referred to as a form of superintelligence because of the concept of Ricardian comparative advantage. If you have a bunch of jobs, the person who does each job isn’t the average person; it’s the person best suited for it. So the envelope of intelligence in an organization is much better than what you’d get by averaging across the population. AI agents are essentially competing against human organizations or teams, so you shouldn’t compare an individual-level benchmark with how well a human does something. You should compare across what we call mosaic intelligence.

Then we start layering on how capability improvements will affect the economy and the equilibrium. This is a formal theoretical paper, and we look at different levels of intelligence. We start with minimal AGI, where AI models are as capable as the average human on every single dimension. We show that even at this point, you won’t see a huge impact on the economy, because organizations are better than the average human. What you’ll start seeing instead is wages changing, because production becomes cheaper, and that affects human wages.

What we show is that real economic transformation requires more than competence, or the ability to perform particular tasks. It requires agents to coordinate with one another. If you look at Hugging Face and even at solutions to the Millennium Problems, there’s this idea of agentic swarms. But if you unpack it, these agents are still very bad at coordinating. They aren’t coordinating; they’re mostly just doing parallel processing. Benchmarks focused specifically on coordination can show that one agent does better than two agents splitting the work, even when they can communicate. Give the same task to two people, though, and they divide the labor, communicate, and do much better than one person with double the resources. The agents are still pretty bad at coordinating. But if they get better at it, you can basically break the Coasean ceiling on firm size, and very different firms can emerge. That’s the basic premise: once coordination improves, the economic possibilities for what firms look like, and what happens to wages and employment, change. That’s when the visible changes will happen.

Erik Torenberg

And what’s the biggest bottleneck to figuring out coordination at the moment?

### What AI means for jobs and wages

Alex Imas

To be honest, I don’t know. I think it could be as simple as the fact that coordination isn’t being trained into the models. It’s not part of post-training. I actually don’t have any insight on that.

Erik Torenberg

Yeah, we’ve talked about some areas where you agree. I’m trying to find some areas where there’s a difference of opinion. Maybe let’s try employment. Alex, I think you’re more worried that the transition might happen so fast that something sudden could happen on the employment side. Do you want to share your concerns or views?

Alex Imas

I’m basically worried about the potential. I actually agree quite a bit with Tyler that things could move pretty slowly in terms of their effect on the real economy. But if there is a fast transition—say, a breakthrough in coordination—what we show in that paper is that wages and employment could change suddenly. There could be a phase transition in what happens to the economy. There will still be new sectors and new opportunities, but the kind of rematching that happens in the economy takes time. With a transition that fast, there may not be enough time for that rematching to happen. You could end up with larger pools of unemployment, leading to recursive loops where people get scarred out of the labor market. That’s my personal worry, but the likelihood is a different story.

Tyler Cowen

I have a related worry. It’s close to the same thing, I think, but it’s just hard to tell right now, when you’re hiring someone, whether that person will be good at AI. We used to have imperfect indicators for the skills someone needed: they went to Cornell, they got all A’s, or they had a good interview. Now the question is whether they’ll be mentally flexible enough to pick up AI. We often don’t know, so that could increase the natural rate of unemployment. Good matches will be harder to make, and search will take longer. I don’t think that problem will last forever. Eventually the AIs themselves will figure it out, and there’ll be certifications for how good a person is at AI. But during the transition, everyone’s a novice. You might be well established, the way a lot of people in the tech world are, but otherwise we’re all novices at this very important new skill.

Erik Torenberg

And so what does that mean for employment and wages? I’m curious how you think about those questions.

Tyler Cowen

There’ll be some additional slowness in hiring people, but there’ll be an incredible number of new projects and new jobs. I’m pretty bullish, actually. Labor markets have been fine, but of course there’ll be disruptions and transition problems, as there always are, as there were with the China shock and the Industrial Revolution. I expect a new set of those. A bunch of them we haven’t even figured out or seen yet, but they’ll come.

Erik Torenberg

To what extent will there be an explosion of agents buying and transacting with other agents in this kind of parallel economy? There are so many more of them. They’re available 24/7. To the extent that agents become the dominant force on the internet, what’s the right mental model for what that will look like, or what it won’t look like?

Tyler Cowen

I don’t know what you mean by “explosion.” If you ask me when agents will be 10% of GDP, I don’t think it’ll be soon. If they rose from basically 0 to 6%, you could call that an explosion. Six percent of GDP is enormous. Even that doesn’t seem that soon. If it’s 1.5%, I could see that happening within five years. If you’re going to force me to guess, you could call 1.5% an explosion. GDP is huge. That’s one of the core insights of economics. GDP is huge.

Erik Torenberg

Yeah. Even with the agents we have now, they’re incredibly capable—the consumer-grade things that live on your phone. But that’s the thing: there are completely unintuitive bottlenecks out there. Websites are designed for human beings, and they’re all different. You still run into issues where your super-smart agent books you a flight and a seat right next to the toilet every single time. Every single time. Then you’re like, “I don’t want to use an agent next time.”

Alex Imas

Or there’s all this regulation out there. That’s the big one, right? Companies are spending billions of dollars trying to do basic choice architecture on their websites to upsell humans on products. Agents don’t have the same psychology, so they won’t be prone to that sort of choice architecture. These platforms have an incentive not to let somebody else’s agent go in and shop for them. You’re already seeing it. If you pull up Muse or something like that, it can’t shop on Amazon for you.

Tyler Cowen

I don’t use agents to manage my calendar, although they could, because I want to know mentally what’s coming each day. The only way I can remember is if I have to write it down and spend extra time putting it on my calendar. That’s stupid. That’s my flaw, but I don’t think it’s going away. The agents could be 10 times better, whatever that means, and I’m still going to want to write it on my calendar. I know off the top of my head that on December 18th I’m seeing a Renaissance music concert. There are these funny indivisibilities in how we allocate our attention, and they’re often some of the biggest bottlenecks.

### Elections, data centers and AI politics

Erik Torenberg

Going back to the regulation part, you seem more sober about the possibility that the midterms might lead to data centers stalling indefinitely. I’m curious about your broader thoughts on AI sentiment. How much of the negativity is real, in the sense that it will lead to what we’ve seen in Bloomington, where they’re trying to outlaw its use to some degree? What are your thoughts on how to think about the political economy of this?

Tyler Cowen

Well, we have 50 states, counties, cities, and even other countries. It seems like it will slow down a bit, but still basically go through. Even the new Texas governor is already backtracking: “Well, we just want the labor unions involved.”

They need to pay good wages. The projects we have now will go ahead. The others we’re going to look at again. Josh Shapiro’s rhetoric was very strong; the reality is quite different. So it raises costs, but I don’t think ultimately it will matter that much. But will there be a bunch of places that absolutely say no? Probably. But again, we have 50 states, and I’m glad we have them. Maybe we’ll end up with 52.

Alex Imas

Yeah. To piggyback on that, I think it completely depends on the path. Everything is endogenous, including political economy. The reason the governors are backtracking is that the unemployment rate looks great and the economy is growing at a reasonable rate. Everything looks fine economically, so there’s no instrumental reason for this kind of political economy reaction.

But what people projecting these very extreme scenarios don’t realize is that, with the beginning of an extreme scenario, there will be the beginning of an extreme political economy reaction, and that will endogenously constrain things. It’s just going to be a back-and-forth. The data will end up looking to people living in the future like things moved a lot slower than people would have thought, because of this interplay.

Erik Torenberg

It seems like the amount of money one can put into politics—everyone was worried about Citizens United a long time ago, and corporations were going to control everything. But Elon put in $500 million, or whatever it was, and couldn’t get the outcome he wanted. Does this expose the limitations of money in politics, or are tech people—or corporations in general—good at allocating for their businesses but not so great at allocating in the D.C. political realm?

Tyler Cowen

Dare I say that academia is slightly underrated by Bay Area people? People who study campaign finance have known this for a long time. There’s a lot of evidence that spending doesn’t matter that much. You can think that’s good or bad, but none of this surprised me in the slightest.

I don’t think the people are fools. You might think they’re fools in the broader sense of having political aspirations at all. But apart from some number of local races, there’s no simple way to spend a lot of money and get the politics you want. It’s checks and balances, even though no one in the system is aligned with you.

Alex Imas

Yeah.

Tyler Cowen

It’s a beautiful system we have.

Alex Imas

Every day, I’m not joking, I’m thinking the American system is incredible, and competitive markets are incredible. It’s not perfect, but exactly as Tyler said, there’s all this magic that comes out of the system.

### How Alex and Tyler got here

Erik Torenberg

Yeah. I’m curious: if you were both interviewing each other, what would you be most curious about? Tyler, I’ll start with you. What are you most curious about in Alex’s perspective?

Tyler Cowen

Your origin story of how you became an economist.

Alex Imas

That’s a very self-serving question, Tyler, because it has to do with you.

Tyler Cowen

Okay, say more.

Alex Imas

I’m an immigrant from Moldova. I grew up on food stamps and things like that in America. My parents didn’t know anything about economics, and I didn’t know anything about economics. I went to college and was pre-med because I knew what a doctor was. My parents knew what a doctor was, too.

In college, you had to take prerequisites, and I took an Econ 1 class. I thought it was fun, but I was still pre-med. I took it, got good grades, and made my GPA higher than my pre-med classes suggested. I was going to go to medical school. I applied and got in, and then I started reading Marginal Revolution. I also heard an NPR program with Richard Thaler. Between the two of you, I thought, “Oh, I can do this with economics. This is incredible.”

Tyler Cowen

I only knew one part of that story, so thank you.

Erik Torenberg

I knew the Marginal Revolution part.

Alex Imas

It was because, at that point—this was 2008 or 2009—there weren’t a lot of economists showing you how much you could do with the topic. I pulled out of the medical school process. I went to the only school that accepted me, which was UC San Diego. Thank you, UC San Diego. The rest is history. This is such an incredible field and knowledge base for understanding, to the extent that we can—which is not very much—what’s going on in this complex system we call the economy.

Tyler Cowen

Can I ask a follow-up, please? Will and should Moldova become part of Romania?

Alex Imas

Yes. That’s my view.

Tyler Cowen

I figure they should, but they will.

Alex Imas

Okay. They were originally. The country was called Bessarabia. My grandma was from Bessarabia. She wasn’t from Moldova.

I think this is probably a controversial opinion, but I’m not actually from Moldova. I’m from Transnistria.

Tyler Cowen

Ah, okay. Now it’s all clear.

Alex Imas

We were refugees from Transnistria to America. There was a civil war. We were in a basement hiding from bullets, and then we ended up coming to America as refugees.

Moldovan culture is very similar to Romanian culture. The language is almost the same.

Tyler Cowen

When VCs say, “Always ask for the origin story,” they’re so correct.

Erik Torenberg

Alex, what question would you ask Tyler?

Alex Imas

I want to know how you were attracted to AI and started thinking about it so seriously. I don’t think I’ve ever heard this story. You were very early in seeing the transformative potential of the technology. How did you realize that earlier than most?

Tyler Cowen

All chess players have a story like this. In my early years, I was a chess prodigy. I was 13, playing in New Jersey, and they wheeled in a computer called Tinkerbell that played against the humans. It sucked, but I saw what the progress was like.

Chess players know this: if they’re any good, they know that chess isn’t like pocket-calculator stuff. It’s intuition. If you see computers conquer an area you already knew was mostly intuition, you find it very easy to believe they can conquer a bunch of other areas. Garry Kasparov was early to this. Ken Rogoff didn’t need much AI-pilling. He was early to this, too. The chess people were the first ones there. That’s a very simple, direct story. That’s how it happened.

Alex Imas

And Demis, too.

Tyler Cowen

Demis, too. Absolutely. The common story is that you see it kick butt by the 1990s, and it beats Kasparov with Deep Blue. Deep Blue is not a great model, but you think, “Okay, it can do this,” and you’re just waiting for it to spread to everything else. But it took a long time. In fact—

Alex Imas

I think this is a great example. A lot of people would have looked at AI and said, “It can’t do this, it can’t do that,” but it’s going to be able to do almost everything. Think about RLVR, or the verifiable-reward part of post-training. A lot of people use that as cope: “That’s not verifiable, that’s not verifiable.” But if you train the verifier, almost anything will become verifiable.

Tyler Cowen

The fact that you can criticize it means it’s almost verifiable, right?

Alex Imas

Exactly. Chess is a really great example. If you’re a really good chess player, you know exactly how much intuition is required and how far that’s going to go in the future.

Tyler Cowen

Mar [?] is working on the poets. I’m not sure how far along they are, but it seems to me it should work.

Alex Imas

Mm-hmm. To make better poetry?

Tyler Cowen

Right, to write great poetry. I think it will be possible in less than two years, if people put their minds to it at all. I don’t mean a big, major effort with billions, just some kind of effort. Bring in a dozen poets, pay them something—it doesn’t have to be that much—let them loose, and have some checks and balances on the process. It’ll work.

Alex Imas

Some people at the firm think the idea that taste is uniquely human is cope. The systems are already good at it.

### Rapid fire: AI predictions and higher ed

Tyler Cowen

They already have great taste. You can ask them about classical music—what’s the best recording of Sibelius’s Fifth, for example—and they give great answers, better than the Penguin Guide right now.

Erik Torenberg

Fascinating. There are a few grab-bag questions I want to ask before we wrap. Did you see Daron’s essay, which he collaborated on with somebody? It basically says that Piketty wasn’t right about the last century, but his model of R > G might be right in the AI century. I’m curious if you have a take on that, Alex.

Alex Imas

Yeah, this is with Phil Trammell, I think. Capital in the Twenty-First Century, yeah. I think the idea that economic activity may be divorced from labor and be focused on capital…

That’s the broader idea of the piece. Many people had issues with the methodology, which I won’t speak to. But the broad idea that technology could increase reliance on capital is a scenario that could happen. There are alternative scenarios, though. As I’ve written about, and Tyler has also alluded to, there could be more jobs created, and those jobs could be valuable because they’re scarce.

Tyler Cowen

Yeah.

Alex Imas

Right. The idea is that value moves toward scarcity. This is the bottleneck logic. If those jobs are in a rich society, a lot of value will go to them. The question is how many of those jobs exist and whether they can support the entire economy. Historically, that’s been the case, but whether it will continue is the big question mark. These are things we need to watch for in the data. My paper with you, “An Economic Policy for AGI,” makes that argument: We need to track where the economy is going with AI much more comprehensively.

Tyler Cowen

Very fundamental technologies tend to benefit consumers: fire, language, electricity, and so on. That’s no proof it will happen again. But we need to distinguish between capital being able to do a lot and entry barriers accruing monopoly rents to particular capital owners.

Alex Imas

I don’t think that piece really made a strong case for the latter proposition. It was mainly arguing that capital can do a lot. But if capital is commoditized—some people say all the rents will go to land, or all the rents will go to energy—I’m not sure that’s true. But if you had to say, those seem more likely than all the rents going to capital.

Tyler Cowen

Yeah.

Alex Imas

That’s right. It goes to scarcity. If scarcity is energy, that’s where the rents will go.

Tyler Cowen

Maybe AI itself makes energy less scarce. That’s why I’m reluctant to predict.

Alex Imas

But I at least see the case for thinking the Gulf countries will make out like bandits, and so on, or the landowners. AI will change enough things that we might be able to terraform land and desalinate water quite cheaply. That doesn’t seem so implausible. Maybe it’ll be the location of the land, or the best retirement spots, like Tahoe, which is special because you’re near the Bay Area but don’t pay the taxes. It’s very hard to predict, but think through scarcity most of all.

Tyler Cowen

Mhm.

Erik Torenberg

Yeah. In our world, on that note, we think the biggest scarcity is on the supply side, and that demand in the short term is much bigger than it seems. That rests on the assumption—or belief—that when you talk to the superusers, they can’t get enough. The superusers are a small percentage of people at a small percentage of companies, and AI still isn’t broadly diffused. The counterargument would be that it isn’t having a measurable impact on companies yet. So it’s speculative that, as AI gets used more, it will eventually lead to an impact on companies that justifies the massive increase in spending.

We just started a new fund, Machine Age Fund, which is doing data centers and other parts of the infrastructure stack to help mitigate some of these supply challenges. Does Leopold’s paper, “Situational Awareness,” strike you as one of the more important public intellectual pieces in technology in the past few years? I’m curious what you think of its legacy.

Tyler Cowen

In anything, not just technology. That said, I think he’s too enamored of fast takeoff. I told him at the time that he’s assuming simulation can substitute for experiment, and thus you get this never-ending loop where science just keeps accelerating. I don’t think simulation substitutes for experiment very much. It will in some areas. Alex?

Alex Imas

We agree.

I’m too much of a Lucas critique pill. Sorry, that’s another bit of inside baseball. It’s basically why you need to understand the deep structural parameters of the problem in order to predict. What simulations do, and what AI is really good at, is prediction. But a lot of the time, it doesn’t take the endogenous response of the system into account. We’ll see that play out more and more.

Erik Torenberg

Robin Hanson’s “The Age of Em”: How do you think that holds up? What will the legacy of those ideas be?

Tyler Cowen

I think it’s wrong. It’s a brilliant book, and I love Robin. When it came out, which was a while ago, I said, “Robin, AI is going to get there first.” Robin said, “No, AI might get there, but ems have to come first.” I said, “Robin, ems probably will never happen, but if they do come first, they’ll need AI.” I think my side of that debate is looking very good. But it’s a great book to read for getting ideas and shaking you out of your dogmatic slumbers.

Do you have any? We’re talking about timely issues. Nvidia is in the crosshairs at the moment, with the Jensen situation. As you view the public discourse, what do you think is misunderstood? How are you interpreting what this potentially means? What lessons, if any, should we take? Tyler?

We’ve already said alignment is impossible. The stakes were so high, and his behavior at the company—something seems out of alignment. In fact, it will be fine because of checks and balances. I don’t know how they’re going to operate. I think it’s a great lesson in all the different things we’ve been saying.

Erik Torenberg

Alex, anything you’d add?

Alex Imas

I don’t think this is really going to change the trajectory much. Checks and balances work. We have diversity in the supply chain to a large extent. Everything is oversubscribed. Demand creates supply; supply creates demand. I think things are going to be fine. The system is robust.

The biggest change might be if something goes wrong at the macro level and Nvidia has to step in as a buyer or lender of last resort. With a weakened or gone CEO-founder type, they’re a lot less able to do that than they would have been if all this had never happened. That matters, but in the medium term, it’s going to wash out. It could slow things down a bit, but the structural forces will take over sooner or later.

Tyler Cowen

Mhm.

Erik Torenberg

If you had to guess, what’s the next genre of “Situational Awareness”-type document that will be remarkably prescient and greatly impact our society? Would you expect it to be something in biotech, or still in AI? What are some underrated ideas right now that might be massive in the years to come?

Tyler Cowen

They’re underrated to me, too. Everyone talks about biotech, and I’m fully on board, but I think that will be especially slow because of regulation and the inability to use simulation to address the main issues. But again, Erik, I think you’ll live to 120, and I think I will not. That’s a simple way to put it.

Alex Imas

I think we’re going to get better models of micro- and macroeconomic phenomena—better structural models—to the extent that AI creates them. The model of the cell, or using digital twins to do macroeconomics: being able to put structure on things we need to predict is going to be the big breakthrough. That gets to the idea in economics of the reduced-form versus structural debate. When you understand the structure, you can create brand-new things. I think AI’s potential to help us understand and recreate that structure is really underrated.

Tyler Cowen

It’ll be great. Raj Chetty has a new paper out on digital twins.

Alex Imas

Exactly. Digital twins—that’s exactly what I’m referencing. But he needed five coauthors, a lot of money, and a research center, which is fine. It’s wonderful that he’s done it. But that also goes to show you something.

Erik Torenberg

Let’s close on the university question. We said earlier that academia is underrated in some ways, certainly by Silicon Valley. Now that we’ve given it its due, where are you most excited about innovation in higher education, and where is the most urgent need? We’re interested in the question, having incubated Harwitz and Dre Academy [?]. Tyler, we’ll start with you.

Tyler Cowen

I hope you all do something and spend more money on it than what you’ve announced, and put some of your best people on it. I don’t really know what your plans are, but I’d love to be involved in some way. As with you, Austin, and Manurva, and other innovations, we need much more experimentation, innovation, and measurement in higher ed. If you’re part of that, I’m thrilled.

Erik Torenberg

Amazing. I would love to have you. Alex, where are you most excited, or what advice might you have for anybody trying to innovate in higher education?

Alex Imas

I think the more innovation, the better. Higher ed is going to change drastically in the next—I don’t know.

I don't have a great timeline for it because they're such slow-moving institutions. But I think market forces are going to lead to a shrinking of our standard idea of higher ed, of what that sector looks like in its traditional version. I think we're going to have a ton of new innovation, and I'm very bullish on higher ed in the long term. That's what I'll say.

Erik Torenberg

Are you both bullish on the field of economics at the moment? Are you encouraged by what's happening right now? Where do you stand?

Alex Imas

I think the economic toolkit and the economic way of thinking are extremely underrated. I'm extremely biased, too. But the way I see things play out—we had the earlier question about alignment, and the economic toolkit is perfect for thinking about that through a mechanism design perspective. You have this kind of thing going on all the time with endogeneity in political economy questions, and thinking through these sorts of consequences.

Thinking through the Lucas critique is important, too. You estimate a parameter, but that's not actually a structural parameter, so it won't be valid to use it to make decisions. All of these things are super important, and I think they'll be more important in the coming days. So I'm bullish.

Tyler Cowen

It may just become social science. AI fills in for the basic mental toolkit economists don't know, and that toolkit becomes more valuable. This is originally a Chicago idea from Gary Becker and others. You just want really smart people with some skills, a bit like those we now call software engineers, but it will have its own name. They have their fingers on the dials, manage a bunch of AI agents, and they're empiricists who know economics and do all of social science. That's what I expect: a true realization of the earlier Chicago vision.

Alex Imas

I'm from the University of Chicago, so I buy that vision. I'm happy with it. Maybe we'll do it at the University of Chicago.

Erik Torenberg

Amazing, a full-circle moment to end the conversation. Tyler, Alex, thank you so much for coming to the podcast.

Tyler Cowen

Thank you. My pleasure.
