No AI Jobs Apocalypse (Yet) - and a Debt Problem (Now) | Martha Gimbel (Yale Budget Lab)

This week we’re joined by Martha Gimbel, executive director and co-founder of The Budget Lab at Yale. Martha has worked just about everywhere economic policy gets made — the Joint Economic Committee, the Obama and Biden Councils of Economic Advisers, and Indeed’s Hiring Lab — and she’s now one of the clearest voices on what the data does (and doesn’t yet) say about AI and the labor market.We start in Washington: what politicians are actually asking about AI, the case for “no regrets” economic policy, and why the unemployment insurance system “is not prepared for someone to sneeze within 50 feet of it” — the tech, the financing, and Mississippi’s $200-a-week maximum benefit. Along the way: why almost no policy “pays for itself” (except funding the IRS), whether one hacker per state plus Claude Code can fix government IT, and the Anthropic finding that agentic coding rewards domain expertise, not coding skill.Then the big empirical question: is AI already taking our jobs? Martha walks through the Budget Lab’s labor-market tracking — no sign of broad macro disruption yet — and why the narrative says otherwise: 1.7 million layoffs in a normal month, CEOs with an incentive to blame AI, and Challenger data attributing seven times more layoffs to AI than to tariffs (”this is implausible”). From there we turn to her Senate testimony and Atlantic essay on the national debt: deficits since 2015 are already costing new mortgage holders about $2,500 a year, and “AI will grow us out of it” is a bet — one the Budget Lab has actually modeled. We close with token taxes, sovereign wealth funds (”the great thing about the government — we can tax it”), what CEA is really like from the inside, and a lightning round that ends in a Red Rising roast.Links & ReferencesMartha’s work* Martha Gimbel — The Budget Lab at Yale · budgetlab.yale.edu* Martha’s Atlantic essay on how deficits are raising costs for households.* Testimony before the Senate Finance Subcommittee on Fiscal Responsibility and Economic Growth — “The Fiscal Outlook: 2027–2036” hearing (March 11, 2026): debt held by the public ~99% of GDP in 2025 → 120% by 2036 → 175% by 2056* Evaluating the Impact of AI on the Labor Market: Current State of Affairs — the Budget Lab’s occupational-mix tracking; no sign of broad AI disruption in the macro data yet* What Might AI Adoption Mean for the Fiscal and Economic Outlook? — the AI-and-the-debt scenarios built on the Karger et al. expert forecasts; the fiscal gains are not a free lunch once you add support for displaced workers* Abhi Gupta, The Impact of Deficits on Costs for Households* The Budget Lab Small Macro Model (BLSMM) — the open, interactive macro model discussed in the R-vs-G section* Long-term Impacts of the One Big Beautiful Bill Act — ~zero growth impact at 10 years, negative at 30 (crowding out)* Coming soon from Martha: a token-tax piece in Tax Notes, and Budget Lab work on AI, capital taxation, and sovereign-wealth-fund economics — stay tunedConcepts, papers & people discussed* Anthropic, “Agentic coding and persistent returns to expertise” — the Claude Code study Martha cites: domain expertise, not coding background, predicts success with AI agents* Challenger, Gray & Christmas layoff announcements — the data attributing ~7× more layoffs to AI than to tariffs in 2025* Brynjolfsson, Chandar & Chen, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI” — covered previously on the podcast.* Olivier Blanchard, “Public Debt and Low Interest Rates” — the “if r is low, debt has no fiscal cost.” * Danny Yagan and Neil Mehrotra — the Budget Lab’s designated R-vs-G thinkers* The Windfall Trust — the AI economic scenario-planning (”war gaming”) exercises Andrey asks about* The ROAD to Housing Act — “Congress is doing something obviously good. Fingers crossed.”* Reinhart–Rogoff and the Excel error — Seth’s aside on debt-threshold doom predictions* The Council of Economic Advisers and the Joint Economic Committee — partisan vs. unbiased, and the puppy-distribution test* Trade-offs of the pandemic UI plus-up — the $600 flat add-on existed because state systems literally couldn’t compute 90% wage replacementSci-fi corner* Becky Chambers, The Long Way to a Small, Angry Planet — Martha’s pick for the future we actually want (and a direct appeal: Becky, the people want to know what’s next)* Adrian Tchaikovsky, Children of Time — “so good”; the one with the spiders* Pierce Brown, Red Rising — “angry Harry Potter”; sorry, BasilOur sponsor* This episode is brought to you by Revelio Labs, providing data products useful for many questions about the economy.Chapters* (00:00) Intro & sponsor* (00:59) What politicians are actually asking about AI — and the case for “no regrets” policy* (02:30) D.C.’s misconceptions: politicians are still learning the tools* (03:22) Scenario planning, war gaming, and automatic stabilizers* (04:43) “The UI system is not prepared for someone to sneeze within 50 feet of it”* (05:54) Three reasons UI is broken: the tech, the financing, and Mississippi’s $200/week max* (08:53) Should we federalize it? The IRS, and why almost nothing “pays for itself” (the 7× rule)* (11:28) Why politicians can’t do the obviously good thing: concentrated pain, diffuse benefits* (12:46) Can one hacker per state + Claude Code fix government IT? DOGE, caves full of paper records, and returns to domain expertise* (16:58) Is AI already affecting the labor market? What the macro data shows (and doesn’t)* (19:05) Why the narrative says otherwise: youth outcomes, wrong numbers, and 1.7 million layoffs in a normal month* (22:22) Challenger data: 7× more layoffs attributed to AI than tariffs — “this is implausible”* (23:16) What is behind the entry-level slowdown? The low-hire, low-fire puzzle* (25:16) Will we know it when we see it? Weavers, export controls, and the Napoleonic Wars* (27:54) “I can’t travel to Earth 2”: the counterfactual problem, even at the Budget Lab* (28:43) The fiscal outlook: debt from ~100% to 170%+ of GDP — and why thresholds aren’t the point* (30:44) Fiscal crisis risk, sweet sweet T-bills, and the Blanchard low-rates argument* (32:05) Why didn’t we issue 100-year bonds at 2%?* (32:44) Crowding out: deficits since 2015 ≈ an extra mortgage payment every year* (33:55) Housing affordability, interest rates, and mortgage lock-in* (37:14) What deficit spending crowds out — and the diapers-bill standard for government spending* (40:25) Fiscal gap accounting vs. talking so D.C. actually understands* (44:26) “AI will grow us out of the debt”: R vs. G and the Budget Lab’s AI fiscal scenarios* (47:05) Who pays when the robots work? Payroll taxes, capital taxation, and AI finding every loophole* (48:33) Is capital more or less elastic in the AI age? Token taxes and the IRS* (52:16) A sovereign wealth fund for AI? “The government doesn’t have to own things to get money from things”* (55:41) Fairness and the nerd’s-nerd case for simplifying the tax code* (56:47) Inside the JEC and CEA: partisan vs. unbiased, and the puppy test* (1:00:34) What working at CEA feels like: second best, third best, fourth best* (1:02:06) The CEA junior staff and their “extremely benevolent and well-reasoned rule”* (1:03:34) Budget Lab vs. Penn Wharton vs. CBO: 30-year horizons and what you’re buying with paid family leave* (1:06:38) Private vs. public data: Indeed, benchmarking, and the shutdown’s three contradictory hiring estimates* (1:08:31) What data do we actually want from the AI labs?* (1:09:41) Lightning round: the biggest bottleneck to AI productivity gains (Hollywood vs. healthcare)* (1:10:53) Gun to your head: pre-distribution or redistribute-after?* (1:12:59) Favorite sci-fi: Becky Chambers and the future we want* (1:14:38) Worst sci-fi takes: Red Rising, Children of Time, and a message for Basil* (1:15:35) Sign-offJustified Posteriors is the podcast that updates its beliefs about the economics of AI and technology, hosted by Andrey Fradkin and Seth Benzell. If we changed your priors, subscribe, share it with a friend, and keep your posteriors justified.TranscriptWhat Politicians Are Asking About AI [00:00 – 04:43][00:00:04] Seth: Welcome to Justified Posteriors, the podcast that updates beliefs about the economics of AI and technology. I’m Seth Benzell, building, with your support, a podcast community which is hopefully more fiscally sustainable than the federal government. Coming to you from the Pocono Mountains of eastern Pennsylvania.Andrey: And I’m Andrey Fradkin, coming to you from San Francisco, California. We are sponsored by the fine folks at Revelio Labs, providing data products useful for many questions about the economy. And we’re very excited to have Martha Gimbel with us today. Martha is the executive director and co-founder of the Yale Budget Lab, and has worked in an enormous variety of impactful roles related to economic policy. Martha, welcome.Martha: Thank you so much for having me.Andrey: To get started: we know that you talk with a lot of politicians and staffers. What questions are they asking you about AI?[00:01:10] Martha: Some of the questions that politicians, policymakers, and staffers are asking are the same ones everyone is asking, right? What is going to happen? How should we think about the economic impacts of this? What is plausible? What seems unlikely? Politicians — they’re just like us. They have the same questions everyone else does.I think the other thing is that people are really trying to figure out how much of the potential future requires something that is different than what we’ve done before. And I don’t just mean “we know that workforce training hasn’t worked particularly well in the past, so we should update how we do it” — the things that you would do no matter what — but a true paradigm shift in how we do policy. And I think that politicians are still sort of confused about that. They don’t know how much they should be thinking about a totally different way of doing things. I should say, I’m one of the people who’s been on record saying that this is a time to do “no regrets” economic policy. We don’t know what’s going to happen, and so now is a great time to do things that are a good idea no matter what. But you shouldn’t wager policy on a specific version of the world.Andrey: What are the most common misconceptions that politicians have about AI, in your experience?Martha: I think part of it is that a lot of people in DC are still learning how to use these tools — as, to be clear, are all of us. They’re still kind of at the baby stages there. And also, think about a typical politician’s life — and I’m not trying to say this is everyone, but think about what they’re doing. They’re talking to constituents, they’re getting briefings from staff, they’re attending a committee hearing. They actually aren’t spending that much time sitting in front of their computer trying to generate memos or do analysis — they have staff who are doing that. So some of it is just people trying to make sure that they have the time to sit down and play with these things and figure out how they work.[00:03:22] Andrey: Have you seen efforts by, let’s say, the Windfall Trust and other organizations to do scenario planning — essentially war gaming, but for AI and economic policy? Do you have an opinion on those sorts of exercises?Martha: In general, the scenario planning is helpful and is a good way of approaching this. It’s also one that DC is very familiar with. The great thing about scenario planning, if you’re trying to think about influencing economic policy in DC, is you’re speaking policymakers’ language. It’s also something people think about a lot in economic policy already: we talk a lot about automatic stabilizers — things like unemployment insurance that kick in when the economy is doing worse, and then trigger off in various ways, or you don’t file for them when the economy is doing better. Thinking about these types of policies that can stretch to fit different scenarios is a place that people in DC really understand, and it has been a subject of a lot of debate — how do we structure these, et cetera — over at least the last 10 years.The Safety Net Isn’t Ready for a Sneeze [04:43 – 08:53][00:04:43] Andrey: How well do you think the current set of automatic stabilizers will do if we have a shock — let’s say a white-collar unemployment shock?Martha: One of the things that is frustrating to people like me is that you will sometimes see... I’m gonna call out someone’s tweet, but fortunately for them, I don’t remember their name, so they totally get to get away with it. Someone tweeted, “People don’t understand — the unemployment insurance system isn’t prepared for the tsunami of white-collar job loss that’s about to come at it.” And all of us who worked on unemployment insurance were like, “You don’t understand. The UI system is not prepared for someone to sneeze within 50 feet of it. What are you talking about? Wave of white-collar — it can’t handle anything.” So, you know, we can have a longer conversation about how well Silicon Valley and DC understand each other. The unemployment insurance system is not in good shape.Andrey: What are the top three reasons why it’s not in good shape?[00:05:59] Martha: Let’s do it. First of all is the tech. Just legitimately filing for benefits, getting them processed, all of those things, is really, really cumbersome. I should say, for some states, that is a feature, not a bug — it is a way to keep from paying out benefits. This is a thing I think a lot of people experienced during the pandemic. You might assume it has been fixed. It has not. And this flows through to UI policy: there are fixes you might want to make to the UI system, but the systems are so buggy that you can’t touch them. I had friends who were parachuting in to try to be helpful to the state UI systems during the pandemic, and if you’ll forgive a metaphor that doesn’t really work, they basically opened up the door to the closet where the code is kept, and then closed it slowly and said, “Okay — if no one breathes, it might not break down. No one look at the code, no one think about the code, and it will be fine.”There was a lot of debate during the pandemic about the plus-up to UI benefits that we did, where we just added a certain amount onto people’s benefits. That meant some people were making more on UI than they had been when they were working, and people were asking, “Is this a good policy or a bad policy? How do we think about incentives to work?” Guys — the reason we did that was that the systems literally couldn’t handle saying we’re going to do a return of 90% of your previous wages. It seems very simple, but they couldn’t handle it. Which is a very long-winded way of saying the first issue is the tech.[00:07:47] The second issue is the financing — how people are paying in for benefits. Trust funds and things like that are not always in the best shape. And third, I think people don’t realize how ungenerous a lot of these benefits are. It varies a lot by state, but in Mississippi, the maximum — not minimum, maximum — benefit that you can get is a little over $200 a week. That’s not a huge amount of money.Seth: It’s not paying the mortgage.Martha: It’s not gonna pay the mortgage. And particularly if you’re worried about higher-paid workers losing their jobs, it’s a huge problem. Which is all to say: no, I don’t think the social safety net is particularly well set up for broad technological disruption. I don’t think the social safety net is particularly well set up for a standard-issue recession. I don’t think the social safety net is well set up for a 4.3% unemployment rate, which we have right now. We have let these systems languish.Why Nothing Pays for Itself (Except Funding the IRS) [08:53 – 12:46][00:08:53] Seth: Is the answer to move in the direction of federalization? Is this incompetent states and the government letting them run amok?Martha: We have federalized programs that also don’t work particularly well. Part of this is just: you have to spend the money for things to work. And who knows if we’ve allocated that in the right way, et cetera. As another example, there was a lot of money that was supposed to be going to the IRS for modernization and increased enforcement, and that’s just been eaten away and eaten away and eaten away. Money for the IRS is particularly frustrating to budget nerds. To get a little bit nerdy about the budget for a second—Andrey: Do it.Martha: Everyone always thinks that their preferred policy will, quote-unquote, “pay for itself.” You see these things: if we spend a dollar on... I don’t know, name a policy you like.Seth: Corporate tax reform.Martha: Well, corporate tax reform might pay for itself, but I mean on the spending side.Seth: Schools.[00:10:00] Martha: Yeah — “additional dollars for schools will pay for themselves.” And from a societal perspective, that is often true: you will have a more highly educated populace, it affects crime, all of these different things. But in order for a policy to pay for itself from the budget perspective, it has to induce economic activity that is basically seven times the size of what it costs — because we don’t tax the whole amount.Seth: Right, that’s just the share of GDP that’s taxed.Martha: Exactly. And so it is almost impossible to have a policy that pays for itself. Funding the IRS does. If you think about what funding the IRS does: one, it makes things easier — everyone’s always annoyed during tax time when they have to call the IRS and they’re on hold for hours and hours. It also allows them to go after more complicated cases that require more resources, but also where you get larger amounts on the back end. And just to state explicitly: when they go after these cases and get this money, this is money that we as taxpayers were owed. People owed us this money. It’s our money. Pay it to us. It’s not a new tax being levied. But even so, the IRS has been starved of resources.[00:11:28] Seth: Why is it impossible to get politicians to do the obviously good thing?Martha: That is the eternal question in DC. First of all, just to state explicitly: the obviously good thing can be very bad for a small group of people, and they’re going to be much louder about it than everyone else. The benefits of improved tax collection are diffuse — we all benefit from that, but I don’t see it. No one sends me a thing that says “the IRS went after this many tax scofflaws this year, and that means...”Seth: JD Vance is on it.Martha: But for the people they go after, that’s a real concentrated pain. They have to pay more in taxes, they didn’t wanna pay that, and so they’re gonna be much louder about it — which is always a problem. And yeah, we’re gonna have to think about state governance and capacity issues. They’re also just not very sexy. You go to your constituents and you’re like, “I’m really making sure that the code that operates our UI system has been updated — please return me to office...” No one cares. But it’s important.Can AI Fix Government? DOGE and Domain Expertise [12:46 – 16:58][00:12:46] Seth: Martha, some people have told me that AI makes coding way easier. Is there some sense in which AI will just solve this problem? Because we’ll just get one hacker in each state, plus, you know, Claude Code Opus, and we’ll knock one of these out in an afternoon?Martha: To be clear, I am hopeful that AI will make some of these issues easier. I also think, as we all know, any kind of AI implementation is never that easy. You also have to think — with government, so much of this stuff has been neglected for so long. There were all those stories last year about a bunch of government retirement records which are still on paper in a cave, I think in West Virginia. This is a broader issue that comes up: people say, “Oh, we can just use AI to solve this,” or some other thing that seems easy, and it’s like — well, wait a minute. Let’s get in there. You have to actually understand the environment in which you’re operating. You have to understand the existing issues. For instance, if you’re the IRS, there are huge privacy concerns — you need to make sure that you’ve really locked up all the data. I can feel people rolling their eyes at me, but it’s a real concern.[00:14:04] And frankly, you saw this with DOGE. There were people in DC who, frankly, were kind of excited about DOGE — “a bunch of people who understand technology and have a mandate to operate, and we might actually fix these things!” ...and we’ve closed down USAID. Okay. It’s not that people in DC are hostile to doing this. It’s just that it’s complicated, and you have to cut through a lot of red tape and barriers and a lot of years of doing things like putting records in a cave in... I apologize if the cave is not actually in West Virginia. I cannot remember where it is. But it’s somewhere.Seth: I was at a conference where I actually talked to some DOGE-ers over this past weekend — I don’t know what they call themselves. Something like this is never values-free, because you can palpably see how excited they are to fire people, right? And so, yes, obviously if we can have AI do a lot of the work of the government, we’re gonna need fewer federal employees. But it doesn’t come across as credible.[00:15:14] Martha: Or alternatively, you free up a bunch of people at the IRS to actually help people with their tax returns. This gets at the whole debate about AI and labor markets and what’s going to happen: if you’re freeing up certain capacity, do people lose their jobs, or do they switch into doing other types of things that are higher value? I also think — Anthropic actually just put out a piece on this last night, or this morning, I’m not sure — about what they’ve seen with the use of Claude Code: there’s this increasing return to domain expertise. If you just send people in to work on something who understand the tech but don’t understand the broader context in which they’re operating, it’s not nearly as effective. So I do think you really need to make sure that you’re having these conversations with people.Not to keep picking on DOGE, but just as an example — I remember there was this story that they found all this real estate that wasn’t being used, and so they’re like, “We’re gonna sell it and put it up for auction, and isn’t this great?”Seth: It’s free real estate.Martha: It’s free real estate! We should do something with it! And look — there are all of these random sites in Northern Virginia that no one says what they’re for, and everyone’s like, “Take the list off the internet. Take it off the internet. If you don’t know what something is in Northern Virginia, go into a SCIF — a classified area — and ask someone. Don’t put it on the internet.” The list came down in five minutes. But it had already been put up.Is AI Already Affecting the Labor Market? [16:58 – 23:16][00:16:58] Andrey: This is a natural transition point. Your lab has been one of the key groups documenting what’s going on in the labor market in real time. So — the big question that everyone’s asking: is AI already affecting the labor market, and if so, how? Is it gonna take our jobs? Or has it already taken our jobs and we don’t know yet?Martha: I would say: I don’t know if it’s going to take your job. There is currently no sign that AI is causing broad disruption in the labor market at the macro level. This does not mean that no one has lost their job to AI yet. This does not mean that no one’s job has changed due to AI — I wanna give all these caveats here. What it means is: if you’re looking at the macro data about the economy, it doesn’t seem to be the case that AI is causing broad disruption at this time. I do want to say, I think this take sometimes gets me labeled as an AI Luddite or an AI skeptic—Seth: God forbid.[00:18:11] Martha: —which I think says something about where this conversation has been. This technology is still very, very new. The ChatGPT that was released in November 2022 is not the same as the tools that we have today. And it is interesting to me that people assume that after only a couple of years, we would have this kind of broad disruption that’s easily measurable. These things just take time. Have you ever met a compliance department? They have questions. So I think it’s really important to be clear about what we’re seeing and not seeing now, while also not dismissing the potential for disruption in the future.Andrey: I totally agree, by the way. I think in the media, the way this gets reported is oftentimes frustrating. They’ll take companies at their word, at face value, when they say they’re firing due to AI. There are all sorts of prominent people already saying that the youth can no longer find jobs, when, as far as I can tell, that’s not at all what’s going on. Why do you think this is happening, and is there something to be done about it?[00:19:40] Martha: First of all, I think there are just things that people get in their heads that feel right to them. In 2022, my friends kept asking me when the recession was going to be over, and I kept saying, “We’re not in a recession.” And they were like, “What are you talking about? We’re definitely in a recession.” I don’t know what to tell you — GDP growth is positive. So there’s the question of how people feel versus what is happening.On the youth, and on the headlines: on the youth side, it is legitimately the case that you have seen some deterioration in outcomes for younger workers. This seems to be coming because overall we’ve had a slowdown in hiring, which, again, does not seem to be related to AI — but it’s not something that labor economists understand particularly well. And that’s just frustrating to people, understandably. I graduated in 2009. My friends and I had a really hard time finding jobs, but we knew why. It was very clear. Whereas now it’s like, “Oh man, things aren’t looking so great. No idea. Good luck out there — have a good time!” So people are searching for an explanation. I also think people have wrong numbers in their heads. I’ve heard senators talking about 20, 30% youth unemployment — which, to be clear, we do not have.On the CEO announcements: part of this is just that people don’t have context in their heads. You’ll see these things like “X company lays off 10,000 people due to AI.” Well, first of all, we have a relatively low-layoff environment right now, and we still have 1.7 million layoffs every month. So 10,000 is just not that many. To be clear — if you are one of those people who’s gotten laid off, it is terrible. It is really hard for you, it is really hard for your family. I’m not dismissing that. But from an economy-wide perspective, it’s just not that big.[00:21:40] There are also huge selection effects in what gets covered, and in who makes announcements. The hardware store down the street, if it starts hiring a bunch of people, doesn’t make an announcement about it. And if it fires a bunch of people, it doesn’t make an announcement either. But these big companies do, and so that gets a lot of coverage — but they’re not the totality of the US labor market. And again, there’s also just an incentive to ascribe any layoffs to AI. It gets covered weirdly kind of positively for your company — you’re forward-looking, you’re doing all the things.Seth: I mean, not weirdly. You’re lowering costs, right? It should be positive.Martha: Well, but also people say, “Oh, they’re a forward-looking company making investments in tech for the future.” There’s this data from this company called Challenger — which has some pros and cons — but what they do is gather layoff announcements, and they attribute the layoff announcements to a source. And between April and November of 2025, about seven times as many layoffs were due to AI, according to Challenger, than tariffs. This is implausible. It just is. But if you think about it — if you’re a CEO, which one is more fun to put in your announcement? There are huge incentive effects here, and I think people aren’t thinking those through, and they aren’t thinking through the scale, and they aren’t thinking through the selection effects in both who makes layoff announcements and whose layoff announcements get reported.The Entry-Level Puzzle and the Limits of Attribution [23:16 – 28:43][00:23:16] Seth: But thinking about that attribution challenge — if you had to come up with a story for why there seems to be a differential job market for entry-level workers versus more advanced workers, would you point to work from home? Would you point to increased interest rates, and some sense in which the worker has to pay off sooner? What else would you point at?Martha: The honest answer is I don’t know. There have been people pointing to work from home — there was a new paper out on that. Interest rates in general have probably slowed down hiring, which is going to affect young people more than older people, because older people tend to have jobs and young people tend to need their first job — so hiring is more important for them. But this kind of gets at the point: a bunch of economists have been debating why we’ve been in this low-hire, low-fire labor market for a couple of years now, and we’re just not sure. There’s been this question of — the unemployment rate’s been about 4.3% for a while. Are we at equilibrium? Is this a labor market that’s at that magical soft-landing, just-chugging-along point? In that case, you’re just not gonna have that much hiring — and that’s going to be bad for young people. That’s one theory. We can’t prove anything.I think part of the reason why the “AI is holding back hiring for young people” theory has really taken off is because we don’t have an alternative. We don’t have an “oh, it’s actually this thing.” We don’t know, and so of course people are going with an explanation. I should also say: it is not insane to think that AI will affect young people first. That’s a very plausible outcome — it’s what we’ve seen in past technological change. It’s not some totally cuckoo theory that people are working with here. It’s just that it doesn’t really seem to be showing up clearly in the data.[00:25:16] Andrey: There’s a paper that we’ve covered on the podcast, “Canaries in the Coal Mine,” and I don’t think we need to rehash that. But my sense has been that if it’s really starting to affect the labor market, we’ll know — we’ll see large shifts in overall employment rates. Is that kind of your sense of the AI impact on the labor market?Martha: I am just constantly the person who’s like, “It’s hard, and who knows what we’ll know, and who knows where this is going?” This is why no one likes hanging out with economists. But if you think about the Industrial Revolution, and you think about the weavers who lost their jobs — that was really clearly due to technology, right? They didn’t need a Bureau of Labor Statistics to tell them they were losing their jobs to technological change. However, what I just said is actually not quite correct. One of the things going on at the time was the French Revolution and the Napoleonic Wars, and England actually put export controls on the textile industry. They couldn’t export.Seth: The Continental System.Martha: It’s a problem! So they sent the industry into recession, and then you had technological change on top of that. How much of the job loss that happened very quickly was about the technology? How much was about the export controls? How much was about some third factor that I don’t know about, because this isn’t my area of expertise? I don’t know. All of this gets very, very complicated — all of these factors going in and out, compounding and building on each other. The economy doesn’t happen in a vacuum. I think part of the problem here is that everyone wants to know right now where this is going, who’s going to be affected, and how and when. And so much of this is probably only going to be clear in retrospect. Which is not a fun answer.[00:27:18] Andrey: And it might not even be clear then, right? There was this recent Wall Street Journal article that I think you were a part of, where people were asked what the net effect of AI on jobs is gonna be. I know I just said we’ll know — I think we’ll know in very specific cases, if certain occupations truly are automated. But I think we won’t know the general counterfactual. It’s not like we’re gonna be there five years from now and know the exact trajectory of the economy without AI. I don’t know how we’d ever get at that.Martha: People ask me all the time, “Has hiring at Budget Lab been impacted because of AI? Are you hiring fewer research assistants?” And I say no — how people do their jobs has changed, the tasks that people do have changed, but the number of people that we’re hiring has not changed, as far as I can tell. But you shouldn’t ask me that question! I don’t know! I can only see the version of Budget Lab where AI happened. I can’t travel to Earth 2, which has no AI, and see how many research assistants we hired. I have no idea. We want this kind of certainty from people that is just really implausible.The National Debt: Why the Trajectory Matters [28:43 – 33:55][00:28:43] Andrey: Okay, this is a good time to move on to our next topic. Seth, you wanna take it away?Seth: I’m super excited to dig into this next topic, ‘cause you’ve been speaking a lot about it recently — you had a recent Atlantic article, and in March you spoke to the US Senate about the fiscal outlook over the 2027-to-2036 time horizon. You seem to be concerned about the national debt. Can you tell us a little bit about our current trajectory there?Martha: Everyone always worries about debt and deficits when people are talking about spending — by which I include both literal spending and also tax cuts — on things they like. “Deficits are fine if it’s on the thing I like.” I often have this experience where I’ll talk to people about deficit spending and how it can increase interest rates, and I’ll give the example of a specific policy, and someone will go, “Well, wait a minute — but that’s a good policy. We should do that.” Sure. But if we think we should do it, we should pay for it.[00:29:46] Seth: So the number that you put in your recent Atlantic article — and perhaps this was for your Senate testimony — was that the debt-to-GDP ratio in 2025 was almost 100% in terms of debt held by the public. By 2056 it is projected to reach over 170%. Why is that bad, Martha? Debt’s good.Martha: So I actually think this is a really important thing. One, I actually don’t think that specific percentages are necessarily bad.Seth: Ooh, you get snaps for that.Martha: There has been a problem in the debt-hawk conversations in DC where people would go, “If we get to this level, it’s all gonna fall apart.” And then we hit that, and it’s still okay.Seth: It turns out that was an Excel spreadsheet error. Reinhart and Rogoff.[00:30:44] Martha: Yes. But you still have this thing where people said, “It would be really bad if we got to 100% debt to GDP” — and the sky has not yet fallen. I think there are two reasons to worry about this. One is that you do risk a fiscal crisis, where markets all of a sudden go, “Oh no, no, no — we don’t like it.” Those aren’t predictable. They could happen at any time, and they’re very, very expensive if they happen. You’re not gonna like what we have to do if we run into a fiscal crisis. The US is weird for all sorts of reasons — we are not even like the United Kingdom — and I think we’ve been really relying on that for the fact that we’re not gonna have a fiscal crisis. Markets don’t have a good place to go that’s not the United States.Seth: There’s an infinite demand for T-bills. I mean, if they want the T-bills so much...Martha: Yeah, those sweet, sweet T-bills.Seth: Actually, the sophisticated version of that argument is the Olivier Blanchard argument from a bunch of years ago — the idea that if safe interest rates are super low, there’s some sense in which there is no fiscal burden to the debt. People just really, really are looking for safe investments. Give it to the market — you’re not actually hurting anybody.Martha: Sure. And then now we have higher interest rates, and so everything is much more costly. Right? You make a bet.[00:32:05] Seth: Why weren’t we issuing 100-year bonds 20 years ago — or whatever, 15 years ago, when we were at zero? Why did that never happen?Martha: I am not a treasury person, so I should really not be talking about this. I’m going to anyway, and probably gonna get this wrong. My understanding was, one, it ended up being much more complicated than people thought it was going to be — you have to figure out market demand and all of these things. This is another example of: you have to talk to people with actual domain expertise. We can all sit here and say, “We should’ve been issuing these really long-term bonds,” and it’s actually much more complicated.Seth: We were borrowing at negative real rates. It seems like a good idea. Okay. All right. Killjoy.Martha: Sorry! Sorry. You should talk to the treasury markets room about how complicated these things get.[00:32:44] I think one of the things that we’ve been trying to really emphasize at Budget Lab is there is this evidence that as debt-to-GDP goes up, it puts upward pressure on interest rates, because of what economists call crowding out. There are so many people who wanna buy so much debt; the Treasury Department issues more debt; all of us poor little people who wanna take out mortgages now have to compete against the great big US government. Some research by Abhi Gupta here at Budget Lab says that if you’re taking out a new mortgage today, Congress’s fiscal actions since 2015 have raised your average mortgage payment by about $2,500 a year. On an average mortgage, that means you’re basically making an extra payment each year. So congratulations to everyone taking out a new mortgage. And I think this gets to the point: there are costs to this. You don’t have to wait for a fiscal crisis for things to be a problem. There are still costs.Mortgages, Housing, and Crowding Out [33:55 – 39:56][00:33:55] Seth: I wanted to pick you up on this emphasis, because in both your Atlantic piece and your Senate testimony, you really emphasized the cost of borrowing for homeowners. One takeaway I had was: this means that instead of talking about the fact that our national debt could fill all 32 NFL stadiums with two tiers of construction pallets filled with hundred-dollar bills — I love those — or that we’re getting hit with a meteorite the size of Texas made of hundred-dollar bills, we should be talking about how deficit spending is making it harder to pay our own bills. So help me understand something: is this actually the substantive argument, or is this a rhetorical move? Because when I think about it, on net, Americans are creditors. And if on net we’re creditors and have assets, don’t we on net want high interest rates — or at least the old people who actually vote?Martha: The old people may be fine with higher interest rates. But we are still talking about a situation where there’s a huge emphasis, for instance, on housing affordability. One of the things that is hard about housing affordability is that one of the best tools we have for it is to increase housing supply.Seth: We’re passing the Road to Housing Act as we speak, right?Martha: I believe we are, yes. Has it passed?Seth: Congress is doing something obviously good. Fingers crossed.[00:35:18] Martha: But increasing housing supply takes time. This is not to say that we shouldn’t do it. But it’s not the case that we pass this bill, or a locality improves zoning, and housing affordability immediately improves. It takes time to build a house. They are building a house on my street right now. They have been building it forever. But interest rates are a lever we can pull to improve housing affordability.Seth: How strong a lever is it? Doesn’t it just bid up the housing prices?Andrey: Yeah, I don’t think it matters at all — I think it gets passed through.Martha: You don’t think it matters at all? I mean, yeah, that’s fair — there’s supply and demand. But I do think it makes a difference. People don’t like high interest rates.Seth: That’s not obvious to me. I think the old voters might like high interest rates.[00:36:23] Martha: No, no, no. I think this is very, very clear: if you look at studies of consumer sentiment and things like that, people do not like higher interest rates. Fair point on “you bring down the interest rates and then housing prices will go up,” et cetera. But—Andrey: Redistribution-wise, I think it really matters. Depending on whether people have flexible mortgages or not, and when they locked them in, it affects different people in very different ways.Seth: And specifically generationally, as we’re pointing out.Martha: Yes. And also, it can unlock housing supply, because you do have this issue right now—Seth: There’s a lock-in issue. If you have a 2% mortgage rate...Martha: ...why in God’s name would you move?[00:37:14] Seth: If you were to ask me why high debt is bad in general equilibrium, I would have definitely talked about risking a fiscal crisis. I would have definitely talked about transferring from young debtors to old savers. But isn’t the number one crowding-out concern that we’re not getting the physical capital investment that we want? Isn’t that kind of the general-equilibrium story — wages are lower because interest rates are high?Martha: Yes. And in general — if you’re thinking about the types of investment that will drive economic growth moving forward, I don’t know that issuing more treasuries so that we can pay more Social Security benefits to older people would be high on most economists’ lists.Seth: It’s the boomers. It’s the boomers again.Martha: It’s always the boomers. To be clear — we pay Social Security benefits for a reason. We don’t want old people to live in poverty. I should say elderly people. I’m gonna get in trouble with my mother; I just called her old.[00:38:13] But again, there are reasons why we do government spending, which I think is really important. Sometimes when we have these conversations about government spending, there’s too much of a focus on “is this going to increase GDP growth by 20 basis points” or whatever. Sometimes government just does things because we live in a society and we should do them. There was a bill someone introduced in Congress, as an example — and I’m not saying I support the bill or don’t support the bill; I actually know very little about the bill — that would give money to new parents. And there was a criticism of the bill that said this amount of money is not going to increase fertility. And my frustration with that was: no one who introduced the bill thought it was going to increase fertility. They just think diapers are expensive and the government should help people pay for diapers. Now, you can agree with that or not — is that a good use of government funds? That’s a question. But the standard for government spending cannot always be “is this a magical thing that will solve all of our economic problems?”At the same time, you can’t just dismiss the cost of the program. You have to think: this program pays for diapers, this program pays for roads, this program pays for missiles — how do we value those different things? And the solution is not to say “we’re just gonna deficit-spend ad nauseam,” because the deficit spending has costs.Talking Fiscal Policy So D.C. Understands [39:56 – 44:26][00:39:56] Seth: Here’s another question for you, Martha. When we started this conversation, you said, “I’m not particularly concerned about the debt-to-GDP number — I’m interested in this continuum of risk of more crowding out, of important borrowing.”Martha: I shouldn’t say that I’m not concerned about debt-to-GDP. I just mean it’s not that I think we’re gonna hit 115% of debt-to-GDP and it’s all over for all of us.Seth: Can I ask a follow-up question then? There’s this popular other way of thinking about fiscal sustainability, which is fiscal gap accounting — taking the present discounted value of all revenues and the present discounted value of all expenditures. Would you favor us thinking about debt through that lens rather than the accounting number? There’s a correct answer.[00:40:51] Martha: I think we can argue about different ways of thinking about this, but this actually gets at an overall point about economists in DC that I wanna make. We can get ourselves into these more complicated concepts — other people would say we should be thinking about the risk that R is greater than G, and how should we be plotting that, all the things—Seth: That’s the next question.Martha: There we go. I think sometimes we overcomplicate things, and I think that that is a problem. We are trying to bring these very complicated concepts to people, and I think it is useful to pull back and think about illustrative numbers, or back-of-the-envelope calculations, that help people understand what’s going on — without, and I’m not trying to pick on you here, retreating into our ivory towers and having these very interesting debates about how we think about R versus G. I’m not saying those debates aren’t important or that we shouldn’t be having them. But if we’re thinking about trying to drive these conversations in DC, it is really important to be talking about things in a way that people understand and that really speaks to them. Because otherwise we’re not going to have an impact.Seth: So the correct answer was that fiscal gap accounting is the correct way to think about fiscal sustainability — but unfortunately, maybe the politicos aren’t caught up to that yet.[00:42:45] Martha: If you wanna write a memo and send it to the Hill, you can do that. But I actually do think this is important. I used to be at the Council of Economic Advisers, and one of the things that we worked with senior economists on really closely was: you can overcomplicate this, and you need to be able to talk to people about this in a way that helps them really understand what is going on — what is the thing they have to worry about and what they don’t have to worry about. Someone was asking about our interest rate work — a totally reasonable question — what about the impacts of government spending on real disposable income? You’re talking about the impacts on mortgage costs, but there’s the other half of this: how does it affect people’s income? And there’s two answers there. One answer is, I would need to know what the actual policy is and run it through a full macro model, and that’s very complicated. The other answer is that we are trying to give people an illustrative calculation that helps them get their heads around something. I just think that that is a really important service to provide to policymakers.Seth: To continue retreating into the ivory tower — and moving into R and G...Martha: I just wanna state: if you wanna go full R-and-G, you actually need Danny Yagan from the Budget Lab, or Neil Mehrotra, who have done a lot of thinking about this. But yes, let’s do R and G.Seth: We’ll ease into it.Martha: Should we define R and G, or is that not necessary for this group?Seth: Well, I’m gonna build up to it.Can AI Grow Us Out of the Debt? [44:26 – 48:33][00:44:26] Seth: So I’ll tell you how we get there. The way I wanna get there is: Director Gimbel — or Martha, whatever you prefer — some people have suggested that actually it is foolish to worry about the debt-to-GDP ratio, or even the fiscal gap, because AI will boost G — will boost the growth rate — so much that we will simply grow out of the debt. So what do you say to those who call you fiscal hot-bed-wetters?Martha: One: that’s a bet, and that’s a big bet. And if you wanna make that bet, you can make that bet — and to be clear, there are a lot of people who are making that bet right now. I think there are a couple of other things that go into this. One is, you’d likely also have some impact on interest rates, and we can debate how big that is—Seth: There we go. This is where we go R and G. So, counter: sure, the growth rate will go up. But also, if we think that robots will bid up interest rates — people are gonna wanna invest in these robots and not invest in T-bills — that’ll make borrowing more expensive. So actually now we’re in a race between how much it boosts interest rates versus boosting growth rates. Do you have a stance on that? Does the Yale Budget Lab have a stance on that?[00:45:50] Martha: We don’t. Or rather, I should say: we have a new macro model that people can play with online that has some feedback — not one-for-one — from growth into interest rates. But I think there’s another aspect of this, which is something people are starting to talk about more and more: yes, there’s G, yes, there’s R, but there’s also how much spending you have to do, and how much revenue you’re getting. We put something out about the fiscal impacts of AI using the Karger et al. forecasts — which is not a statement that those are correct; we were using those as inputs. And one of the things we point out is: yes, under their forecasts, debt-to-GDP improves — in some cases improves dramatically. But then you can start wearing away at that, depending on what you assume around government spending and government taxation.Seth: Are you thinking about social support, or are you thinking about arms race?Martha: I’m thinking about social support. They have forecasts of what’s gonna happen to the labor force. You can assume that we spend on displaced workers the way we spend on usual unemployed people — which is not very much money, by the way, per our previous conversation. Or you can assume that we spend on them more similar to how we spend on Social Security, which then has further deterioration. There are then questions around capital share and inequality, which impacts what taxation looks like. I will say, spoiler alert, that is something my team is thinking about right now. And it’s not just these questions around capital versus labor. It’s also questions like: how much of the payroll tax are you still paying? We have these gaps in our taxation system. I should also say: we tax capital less than we tax labor in this country, and we also have many loopholes for capital taxation. And if you think that AI can’t find every loophole in capital taxation possible, I have a bridge I’d like to sell you. I will sell you the IRS cafeteria where they have been keeping all of the paper records.Taxing Capital in the AI Age: Token Taxes and Loopholes [48:33 – 52:16][00:48:33] Seth: Maybe let’s talk about tax policy for a minute there. I’m very curious to hear what your answer is. We know the traditional reason why we tax capital at a lower rate than labor is the idea that labor is inelastically supplied, but capital is very elastically supplied. Is AI gonna change that? If anything, it seems like capital would be even more elastic in that age.Martha: I’m so glad you asked this question. I am so interested in this question. We obviously have no idea, but I think it’s fascinating.Seth: It’s the country of geniuses on a cloud. You can put it anywhere you want.Martha: Well, but also, it’s plausible that it becomes less elastic — if the returns are so much higher there. I don’t know; I can argue this either way. And that’s kind of what I feel like with these questions around elasticity and capital taxation: you can come up with whatever theory you want, and we’ll see what happens. I don’t think it’s clear. I do think it’s clear that we should deal with the loophole situation.Seth: So, Martha, is consumption taxation the answer? Is a Georgist land value tax — as it’s always the answer — the answer? Can we get Rand Paul to finally support a VAT?[00:49:55] Martha: If you could get Rand Paul to support a VAT, I think everyone in DC would be lining up to hire you as a lobbyist, because that is a level of persuasion I don’t think people realized was possible.Andrey: We need the AIs to do that persuasion.Seth: The super-persuasive AIs.Martha: That’s right. You know, this kind of goes back to where we started, actually. One of the things that I worry about is that some of the conversations around taxation and AI are — people are trying to come up with a whole new different way of doing things, and they’re thinking about it theoretically, as opposed to working with the real-world constraints that we have. So there’s been a lot of discussion about a token tax, and I won’t get into the pros and cons of a token tax from an economic perspective—Andrey: I will. A token is a—Seth: You know this is a podcast for that, Martha.Martha: I have a piece coming out on this in Tax Notes in like two weeks! Don’t spoil me.Seth: We’re gonna scoop you.Andrey: Okay, okay — I’ll let you have it. No, I just think that, I mean, obviously the details matter, but tokens are a meaningless unit. So I think it’s just crazy to think you’re gonna have a per-token tax.[00:51:25] Martha: Yeah. There are so, so many things to say here. One of the things I will say is: you get the IRS to figure out how to implement a token tax. What are you talking about? And again, that’s not a shot at the IRS — they have a lot on their plate. Let’s really not try to overcomplicate this. I worry a lot that people are trying to come up with a new way of addressing problems. If the issue is that people are concerned about increasing returns to capital and us not taxing those properly — if that’s the thing you’re concerned about, we should figure out how to properly tax returns to capital.Seth: Rather than invent some micro thing to target the tax at.Martha: Yeah.A Sovereign Wealth Fund? “We Can Tax It” [52:16 – 56:47][00:52:16] Seth: Let me ask you about one last pie-in-the-sky economic reform policy before we move on to our last topic.Martha: I swear to God, if this is gonna be UBI...Seth: It’s UBI-adjacent. This one, I’m gonna say, is actually in the Overton window, because perhaps you know that our president recently talked about having a stake in the big frontier labs. And when you talk to people about transformative AI, a lot of the discussion is around: well, we should have a national sovereign wealth fund, so the people can own the robots. The people will own the AI, and then we can share the wealth. Is that realistic at all, given the national debt? Or is there some universe where actually we should sell T-bills and then buy Anthropic stock as a government?[00:53:08] Martha: I realize that I keep saying “oh man, the Budget Lab is working on a paper around this” — but this is one of the things that we’ve been trying to think through: what are the economics around this? So stay tuned. I will say, in general, I get itchy about government owning parts of corporations.Seth: I mean, it’s free money, Martha. You just borrow at 2% and invest at 5%. What could go wrong?Martha: I do love free money. Whomst among us doesn’t love free money? I think this actually gets at a broader thing, which is: this conversation, by definition, basically assumes that the benefits of this will be fully captured by — name your preferred AI lab here. What that policy doesn’t do: if some other organization is able to use AI to massively increase their profits, and fire all of their workers, and they don’t hire any other workers and it’s just pure capital returns — we don’t get any revenue from that. They’re still operating under the old system. So this goes back to this thing of: you’re assuming that the profits from this technology are going to end up in a very specific place. But we don’t know that.Andrey: The steel man of this could be just that this is a hedge — not that, for sure, the big labs are gonna take all the rents.[00:55:04] Martha: Sure. Or we could think about improving capital taxation, which is a good thing to do. We should think about improving how we tax capital in this country no matter what.Seth: You know, the government doesn’t have to own things in order to get money from things.Martha: Truly. This is the great thing about the government: we can tax it. ...Someone’s gonna pull that out — “this is the great thing about the government, we can tax it” — and put my name and a really scary picture of me on it.Seth: That’s the TikTok clip. Exactly.Martha: “This woman is coming for your...” But, you know.By the way, one thing I will also say about this — and this is not really econ, but I’m gonna go with it anyway. One question that’s come up repeatedly in all of this, for a lot of people, is this question around fairness. I actually do think that one of the issues we have right now is that people feel like the system is not fair. And this comes up on the taxation side partly because there are so many loopholes and ways to avoid paying taxes on capital gains. One of the things that you get by simplifying the tax code, and making it much clearer who pays what for what reason, is that people go, “Oh, okay. That is this person’s contribution. This is my contribution. We all live in a society. We go from there.” In a situation where people feel like things are unfair, decreasing the complexity in the tax code actually can make a difference. I realize this is a nerd’s nerd’s take, but I actually do think it’s important.Seth: This is the midwit meme. It’s like: just do proportional taxation... “no, we need crazy developed policy”... just do proportional taxation.Inside the CEA and JEC: Partisan vs. Unbiased [56:47 – 63:28][00:56:47] Seth: All right. The last topic I wanted to ask you about, Martha: you’ve worked in so many fascinating positions inside and outside the government, adjacent to policymaking — from the Joint Economic Committee, to the Obama CEA, the Biden CEA, and now in this think-tank role running the Yale Budget Lab. Can you talk to us a little bit about, when we as outsiders read a report from something like the JEC versus the CEA, how much is that a real pure-economics document versus a political document? How should we read it as outsiders?Martha: One thing I will say is, JEC in particular often has leadership changes — it can switch with every Congress, depending on who’s in control of the Senate or the House. CEA obviously switches over every four or eight years. So part of this is just gonna depend on who’s in charge.I’m gonna get up on my high horse a little bit on this one. There was something that you saw people saying for a couple of years about the Council of Economic Advisers — that it was a nonpartisan place to work. CEA is not nonpartisan. You work for the President of the United States. You are within the Executive Office of the President. You are not nonpartisan. What CEA’s job is, is to be unbiased. You are the person whose job it is to sit in the room when they say, “We wanna give every American a puppy,” and you say, “That’s a very interesting policy. How will we distribute the puppies? How do we think about different preferences for puppies? What about allergies? How do we deal with the refuse problem?” Your job is to provide the unbiased analysis.And one of the things I think you see when CEA is working really well is that people in DC can kind of tell. You’ll notice that they’ll send the CEA chair out to talk about certain topics, and then on other topics they just never get sent out to talk to the news media. That’s because you wanna send the economists out to talk about things where the economists think it’s a good idea — their credibility is really important. Different administrations are gonna approach this differently, but I think historically that’s the way CEA has worked, and that’s really important.[00:59:40] Andrey: It also matters whether you have a charismatic person — I mean, Goolsbee was out there all the time in a way that I think others weren’t necessarily.Martha: That’s also a personality thing. There are people who enjoy talking to the press and people who don’t. There are people who enjoy being public figures and people who don’t. There are people who are really good at it and people who are less good at it. But I do think it is important that you are only sending the CEA economists out on things that they legitimately feel they can speak to from the economic literature and the underlying economic reasoning. Because otherwise, what are we even doing here?Andrey: I’m curious, on a personal level, working in these roles — what does it feel like? Do you feel like you’re making a difference? How does it compare to other types of jobs?[01:00:46] Martha: I am biased here. I worked at CEA twice — under Jason Furman and under Cecilia Rouse. It is hard to get better bosses, so I had a great experience. Most of my closest friends are from when I was at CEA, each time, so that kind of speaks to what my experience was like.I also think that working there can be hard, particularly for economists — and this gets at some of the conversations I was having with Seth earlier. A lot of the time, economists come in and say, “We have the one right way to do things.” And then comms freaks out: “You can never say that publicly.” And then the White House Counsel’s Office freaks out: “That is so illegal — please never tell anyone you even thought about that.” And then legislative affairs freaks out: “That will never get passed. Please don’t do that.” So you have to think really carefully: okay, this may be the optimal economic option, but what’s second best? Maybe what’s third best? If we get to the fourth-best thing, is that actually worse than doing nothing? You do have to operate under these real-world constraints. Some people find that incredibly frustrating, and some people find it incredibly interesting. I find it incredibly interesting.Andrey: Very cool. Is there something you can talk about that you’re particularly proud of from your work at the CEA?[01:02:14] Martha: Oh, man. I try not to do that. Mostly because your job is to be an advisor behind the scenes, and not to be someone who’s out there saying “I did this thing.” You’re working for a president. You’re working for a policymaker — they are the person.The actual answer to your question, not to get overly soppy: particularly when I was at CEA the second time, I was in charge of the junior staff. And the CEA junior staff are kick-ass. I will say to all of the economists out there thinking about whether or not AI is going to take your job: AI may or may not take your job, but the former CEA junior staff are definitely taking your job. And they’re gonna be so much better at it than you are, and it will be fine. We just had such amazing people there, and it’s been really amazing to watch them do all the things they’re doing after leaving CEA. I look forward to their extremely benevolent and well-reasoned rule. I think it’s gonna be great.Budget Lab vs. the Other Modeling Shops [63:28 – 66:38][01:03:34] Seth: My last question is about these different forecasting and policymaking groups. Do you see differences in high-level philosophy between how you at the Yale Budget Lab, or maybe the Penn Wharton Budget Model, or the Joint Committee on Taxation approach things? You’re all academic economists — are you all coming at it from the same direction, or are there big philosophical differences?Martha: I don’t know that I would necessarily say philosophical differences. I should say, basically all of the modeling shops think of JCT and CBO as the one ring to rule them all. If we get an estimate that’s markedly different than theirs, generally we’re like, “Hey, staff, what on earth is going on here?” And sometimes there’s a reason — a difference in a parameter, et cetera.One thing Budget Lab is interested in that I think is different is this very long-term approach. The costs and benefits of policies can look very different at 10 years versus 30 years. If you take the One Big Beautiful Bill Act: at 10 years, its average impact on growth is basically zero. If you go out 30 years, it’s negative — because of what we were talking about earlier: the rise in interest rates, the crowding out of private investment that slows down economic growth.We are also interested in non-monetary economic benefits. As an example, you can look at paid family medical leave. Paid family medical leave is not gonna double GDP growth — that’s just not what it does. It enables people to spend time with a newborn, for instance, and one of the things we know is it has some impact on neonatal mortality. That’s a thing a lot of people care about, and it’s not an estimate that’s part of the traditional budget modeling process. I think allowing people to think about what is it that we’re buying with this policy — we are spending money for a reason; what are we getting; do we think what we’re getting for those dollars makes sense — is important. That’s been relatively easy in the past on tax policy, because what you’re trying to do with tax policy is redistribute, raise more revenue, et cetera. So that’s easy — or I should say easier; the poor people who do the tax modeling for me are screaming right now: “Martha, our jobs are not easy!” But particularly on the spending side: what is it that you’re buying, and how efficiently are you buying that thing?Private Data, Public Data, and the AI Labs [66:38 – 69:41][01:06:38] Andrey: All right. You also have spent some time at Indeed—Martha: Oh, no. To be clear, I loved Indeed, but now I’m terrified about where this is going.Andrey: No, no — we were curious what you think the role of private versus government data will be.Martha: I think private data is incredibly important. It allows you to get much more disaggregated. You can sometimes see things faster. You can see things that don’t show up in public-sector data, because they may not be measuring it. But I think sometimes people talk about private-sector data as a panacea — “oh, we won’t have to have public-sector data anymore.” I do not think that is correct. One, private-sector data shows you what one company sees. We are often interested in what is happening with the overall economy, so someone needs to do the aggregating to help us see that. People should also keep in mind that a lot of the private-sector data indices you see are benchmarked against official government data — “hey, do we have the right number of job postings for leisure and hospitality, given what we know about hiring in that industry from official government data?” So there’s really a symbiotic relationship there that is really important, and it’s important not to say, “Oh, we’ll just use private data and it’ll be fine.” You saw this, by the way, when the government was shut down last fall, and we had three different private-sector data estimates come out: one said “hiring is fine,” one said “we’re having no hiring,” and the other said “hiring’s fallen off a cliff.” And everyone was like, “Well... who knows?”[01:08:31] Andrey: There’s been also a call for data sharing between the AI labs and the government. Obviously, all else equal, it would be great to have more data from the labs. Is there anything particularly interesting in terms of data that the labs have that we would want?Martha: To be clear, per your point: I love data. Data is great. I personally should have access to all the data. I will say — if I think about the things that I’m wondering about in the labor market right now, what I kind of want to know is: if a company adopts AI, how quickly and how do they adjust their hiring? And that’s not something, as far as I’m aware, that the labs can answer. So I think there’s been a little bit too much focus on the data that the labs have, and not on starting with: what are the questions that we need to answer right now, and where does that data live?Lightning Round: Bottlenecks and Redistribution [69:41 – 72:58][01:09:41] Andrey: Related to this point — this is also in our speed round — economists talk about bottlenecks to AI adoption leading to productivity growth. What do you think is the biggest bottleneck?Martha: Oh, man. The biggest?Andrey: You can do a big one. Your favorite.Martha: Actually, I’m gonna cheat here, which is to say that this really depends on different industries and different occupations — and that that is a really important part of this overall conversation. It’s going to look different in different workplaces, and I think sometimes we talk about this as if it’s overly homogeneous. For instance, Hollywood is not particularly regulated. The biggest barrier I see there is questions around consumer sentiment: are consumers gonna revolt if they think a movie was made with AI? On the other hand, in healthcare, there are real regulatory and liability concerns, so that’s going to make a difference there. People need to do much more thinking about the dynamics in specific industries and occupations, rather than some of the broader economy-wide thinking.[01:10:53] Seth: Fair. All right — so imagine some of those bottlenecks get unlocked. We start getting these productivity gains and some disruption to labor demand. Gun to your head—Martha: No.Seth: —an imaginary gun — are you on team pre-distribution, to save good jobs? Or are you on team let-’er-rip — second fundamental welfare theorem, get efficiency, and then redistribute after?Martha: I think this gets at one of the things about working in policy versus being in academia. The economist answer is: let ‘er rip and then we redistribute, and we get the most economically efficient whatever. People don’t seem to like that very much.Seth: Why don’t people love us? Martha, why don’t people love us?Martha: It’s a real question! I actually legitimately do think this is one of those hard questions for economists working in policy. We can say, “Look, you get the most economically efficient outcome if you allow this thing to happen, and then obviously we can just redistribute on the back end.” The citizenry just doesn’t seem to like that. And so I think we need to do more thinking about how to balance those issues.Seth: Okay, but gun to your head.[01:12:18] Martha: I’m an economist, right? I have this instinct. But I’m just saying — again, this is part of the problem. We are the problem. People’s preferences are real, and we have to take into account people’s preferences, even if we are sitting there saying, “No, no, you don’t understand — GDP growth will be this much higher, and then we just redistribute it, and it’ll be fine, and your welfare will be higher.” People don’t like that.Seth: All right, we should keep it halfway.Martha: But also, people should let us redistribute as necessary, and not get mad at us for it. But that’s not how things work.Sci-Fi Corner and Sign-Off [72:58 – 76:14][01:12:59] Andrey: Let’s wrap it up. What is your favorite sci-fi book or author?Martha: Oh, that’s mean. That’s really mean. I should say, I was a huge sci-fi nerd as a kid, and obviously still am. Because we have been here talking about AI, I’m gonna say Becky Chambers — just because I think Becky Chambers is the version of the future that we all kind of want for ourselves. It’s these amazing space-going communities with these friendly AIs. It just seems very nice. So I’m gonna choose the optimistic one. And maybe we’ll get that.Seth: Great recommendation. Do you have one book by her in particular?Martha: I think you kind of have to start with — I’m gonna get the title wrong — A Long Way to a Small, Angry Planet, which is the first one. It’s just so, so great. There’s a follow-up that takes place on a space station, with a very cute child that they all start looking after together — I forget the name of that one, which is really bad, but I loved it. She’s working on a new thing right now, and I keep refreshing her website to see if she’s announced what it is yet. So if by some miracle this makes its way to Becky Chambers: the people wanna know what’s coming next. Please let us know.Seth: Amazing. And definitely a better sci-fi take than some of the takes we’ve had on this podcast. I won’t name names.Martha: I wanna know what the worst sci-fi take was. Okay, don’t name names — but I wanna know.[01:14:38] Seth: I’ll name names. Basil Halperin said he loved Red Rising — which I just finished. Which is so mediocre.Martha: No! Oh my God. I’m sorry — I love Basil, but that is... I’m legitimately very upset about this.Seth: Next time you see him, you can tell him Seth really did not enjoy Red Rising.Martha: Let me rephrase: Red Rising is a fun read you can get through in two hours. It’s terrible, but, like, sure.Seth: Angry Harry Potter.Martha: So angry. The angriest of Harry Potters. By the way, if I can do a second one, it would be Adrian Tchaikovsky’s Children of Time, which is so good.Andrey: Great, great book.Seth: Oh, that one’s great. With the spiders?Martha: Yeah, with the spiders. It’s so good. Children of Time is great. People should read Children of Time.[01:15:35] Seth: All right, Andrey, shall we wrap it on that? Everyone read Children of Time and...Martha: A Long Way to a Small, Angry Planet.Seth: And everyone out there, when you’re done reading, you should...Andrey: The Yale Budget Lab?Seth: Say the line. Keep your posteriors justified. We have a line, Andrey.Andrey: I know our line. You gotta constantly be keeping your posteriors justified.Seth: Never stops. Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe