Is AI Replacing Programmers or Boosting Them?

This week we’re joined by Robin Hanson, economist at George Mason, author of the Overcoming Bias blog and The Age of Em, and the person behind a long list of chewy, influential ideas, from prediction markets and futarchy to grabby aliens.📜 Read the full transcriptWe start with AI and growth. Robin’s median forecast for the next decade is a burst about the size of the internet boom, concentrated in software. He expects the software industry to roughly triple, with 20% of the new revenue going to AI and most of the rest going to the top engineers who can manage it.Then we turn to prediction markets. Why are long-horizon markets expensive, and why does it matter how long a piece of information lasts rather than when the question resolves? How should you subsidize a market so you pay for information near your actual decision threshold and decision date? Can a whale just buy the outcome? From there we move to AI alignment. In an old post Robin argued we should “prefer law to values” when it comes to AI. If you’re picking a country to retire to, you check the crime rate, not opinion polls about people’s values. He also argues that the push for draconian AI regulation is out of proportion to the harms we’ve seen so far. Even this summer’s Hugging Face incident, in which an autonomous AI agent broke into the company’s internal systems, is in his telling “teenagers going on a joyride, doing some graffiti.”The second half covers Robin’s newest project: cultural drift. Culture is humanity’s superpower, Robin says, “and we broke it.” In three centuries we went from hundreds of thousands of peasant cultures under strong selection pressure to a single rich, peaceful, fast-changing global monoculture that celebrates activism. That moves every parameter of a working selection process in the wrong direction, and falling fertility is the clearest symptom. Robin proposes three fixes, each a big ask:* Unleash capitalism everywhere, including giving parents a share of their kids’ future tax revenue* Govern groups by “number go up,” using futures markets on their long-run adaptive success* Rally a large polity around a sacred goal that civilization can’t collapse on the way to, such as when do a million people live in space?We close with a lightning round on science-fiction authors (Vinge, Asimov, Stephenson, and Le Guin), which includes the moment Robin reveals which sci-fi character LLMs say he most resembles.Links & ReferencesRobin’s work* Overcoming Bias, Robin’s blog* Robin Hanson, The Age of Em: Work, Love, and Life when Robots Rule the Earth, Oxford University Press, 2016. In Robin’s words, it’s “like science fiction, except there’s no plot, and there’s no characters, and it all makes sense”* Robin Hanson, “Shall We Vote on Values, But Bet on Beliefs?”, the original futarchy / decision-markets proposal* Robin Hanson, “Logarithmic Market Scoring Rules for Modular Combinatorial Information Aggregation”, the automated market maker that most subsidized prediction markets use* Robin Hanson, Ryan Oprea & David Porter, “Information Aggregation and Manipulation in an Experimental Market,” Journal of Economic Behavior & Organization (2006), the lab evidence behind “manipulators are sheep”* Robin Hanson, “Prefer Law To Values”, Overcoming Bias (2009)* Robin’s recent posts on cultural drift (book in progress):* “Skinner, Cattell, Stewart, Wilson on Deliberate Culture”, which summarizes what’s distinctive about his cultural drift analysis* “Five Slogans Re Cultural Drift”* “Directed Culture Trials”* “Without A Sacred Quest, Our Civilization Falls”, the case for proposal 3* Robin Hanson, “Explain The Sacred” (2023), the hub post for his series on the sacred, which Seth mentions* Grabby AliensEconomics of AI* Paul David, “The Dynamo and the Computer,” American Economic Review (1990), on why general-purpose technologies take decades to show up in productivity statistics* Timothy Bresnahan & Manuel Trajtenberg, “General Purpose Technologies: ‘Engines of Growth’?” Journal of Econometrics (1995)AI safety* Hugging Face, “Security incident disclosure — July 2026”, the autonomous-agent intrusion Andrey brings upPrediction markets* MetaDAO, which runs crypto governance by decision market* Kalshi and Polymarket* Robert Shiller, Macro Markets: Creating Institutions for Managing Society’s Largest Economic Risks (1993), on long-run markets for hedgingChapters* (00:00) Cold open* (00:35) Welcome, Robin Hanson* (01:32) GDP over the next decade: an internet-scale boom, mostly in software* (03:10) Bottlenecks, priors on automation bursts, and why intelligence isn’t one-dimensional* (07:11) General-purpose technologies: reorganization takes decades* (08:33) Invention vs. diffusion* (09:40) Long-run prediction markets and the cost of capital* (12:32) It’s how long the information lasts, not when the question resolves* (14:58) Subsidizing markets: paying near your thresholds in price and time* (17:39) Hedging, Shiller, and what Kalshi and Polymarket are proving* (20:15) Decision markets in the wild: the MetaDAO* (22:18) Can a whale buy the decision? Why manipulators are sheep* (26:11) Seth’s apes: short squeezes without information* (27:20) AI forecasters: substitute or complement for prediction markets?* (31:24) AI property rights: prefer law to values* (33:38) Psychopaths, spite, and where you’d retire* (36:43) Humans as amalgams of heuristics, and training AIs on reputation* (40:06) Adaptive evolution of AI products, and regulation out of proportion to harms* (42:59) Cultural drift: “we broke it”* (44:03) Evolution of vs. in: coral reefs vs. the open ocean* (46:37) Natural selection as a control process: the four parameters* (48:09) From peasant cultures to a global monoculture* (50:20) The one-time dividend of cultural merger* (51:36) Ought from is, and the one-suitcase choice* (54:11) Why does adaptation have to mean fertility?* (55:35) Is Robin a nihilist? Three proposals* (57:17) Proposal 1: unleash capitalism, and give parents a share of kids’ future taxes* (01:01:19) Proposal 2: “number go up” on adaptive success* (01:04:46) Proposal 3: a sacred goal: when do a million people live in space?* (01:08:08) Lightning round: sci-fi authors, overrated or underratedJustified 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 and keep your posteriors justified. Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe

This week we’re joined by Sarah Bana — a friend of the show and colleague from the Stanford Digital Economy Lab, now a professor at the London School of Economics. She is a leading researcher on questions around how AI is changing jobs, with a unique emphasis on how employers are using new digital tools as part of the hiring process. Our conversation focuses on one example: pymetrics, a company that helps hiring companies filter applicants by having them play “fun games” like a balloon popping video game. Andrey is quite skeptical that these screens are more efficient than coin flips, but Sarah’s main point is, whether or not they are 100% arbitrary, this somewhat arbitrary screen may lead to negative social consequences. Specifically, “algorithmic monoculture,” where arbitrary distinctions randomly prevent some deserving people from getting jobs. Indeed, Sarah’s research shows that position-level racial disparities are present despite the vendor’s audits.From there, the conversation moves to measuring workers and work generally. Andrey offers the hot take that the most serious labor economists in the world now work at Mercor and Surge AI; Seth nominates sabermetricians instead, and we ask whether being above or below the corporate panopticon is the big question moving forward, or merely another compensating differential. We then discuss use of AI in academic reviewing. Sarah’s Management Science paper proposes an AI first-line referee whose report the authors answer before any human reads the paper. Andrey declares it reviewer malpractice not to run a submission through AI, and Seth predicts the future of academic publishing is basically Goodreads — every work coexisting with publicly posted reviews, which your agent helps you navigate and rank for your own purposes. We close with a few questions about faith and AI: Pope Leo XIV’s encyclical, what the Ismaili Imam has said to graduates about AI, and Sarah’s useful metaphor for categorizing positive and not so positive uses of AI. LinksOur guest* Sarah Bana at LSE and at the Stanford Digital Economy LabAI, signaling, and algorithmic monocultures in hiring* Bommasani, Bana, Creel, Jurafsky & Liang, “Algorithmic Monocultures in Hiring”, FAccT ‘26 (ACM version) — ~3 million applicants, ~4 million applications, all screened by one vendor’s algorithms. About 4% of people who applied to 10 positions were recommended for rejection by all 10, more than independent decisions would predict.* “Q&A: Algorithmic Monoculture in Hiring” — Stanford Digital Economy Lab interview with Sarah and her coauthors* “Harver Acquires pymetrics” — Sarah’s “bought up by Harver” at 9:24 * Lejuez et al., “Evaluation of a Behavioral Measure of Risk Taking: The Balloon Analogue Risk Task (BART)”, Journal of Experimental Psychology: Applied (2002) — the psychology instrument behind the balloon game. * Galdin & Silbert, “Making Talk Cheap: Generative AI and Labor Market Signaling” — the Jesse Silbert paper Sarah mentions: on Freelancer.com, the wage premium for tailored applications disappears after LLMs arrive* NYC’s Local Law 144 on Automated Employment Decision Tools requires published bias audits; the four-fifths threshold Sarah also mentions comes from the EEOC’s Uniform Guidelines, 29 CFR § 1607.4(D)* JP: Does AI Cheapen Talk? (Bo Cowgill Pt. 1) and Epistemic Apocalypse and Prediction Markets (Bo Cowgill Pt. 2) — Sarah’s favorite episodes, and Bo’s paper on AI helping the bottom of the applicant pool more than the top* JP: Can an AI Interview You Better Than a Human? — our episode on Jabarian & Henkel’s AI voice-interview field experiment* Coasean Singularity Links: Shahidi, Rusak, Manning, Fradkin & Horton, “The Coasean Singularity? Demand, Supply, and Market Design with AI Agents” — see also JP: Seb Krier on AGI, the Coasean Singularity, and EDMMeasuring work: Surge, Mercor, GDPval, and the corporate panopticon* Surge AI and Mercor — the expert data-labeling firms Andrey nominates as the world’s most serious labor economists* Patwardhan et al., “GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks”, and our episode on it, JP: Evaluating GDPVal, OpenAI’s Eval for Economic Value — Sarah’s other favorite* Abowd, Kramarz & Margolis, “High Wage Workers and High Wage Firms”, Econometrica (1999) — the “AKM” firm-effects framework behind Sarah’s point that high-wage firms also tend to have the free coconut waterAI and peer review* Bhargava, Bana, Zhang, Brandimarte, Choudhary, Li, Loupos & Zantedeschi, “Fighting Fire with Fire: Infusing Artificial Intelligence into Peer Review to Sustain Quality Scholarship”, Management Science (2026) — the AI-first-referee workflow, where authors respond to the AI report before human review* Refine — the AI technical-review tool Andrey advises, and our conversation with its cofounder, JP: Ben Golub: AI Referees, Social Learning, and Virtual CurrenciesFaith and AI* Pope Leo XIV, Magnifica Humanitas (2026) — the encyclical on protecting the human person in the age of AI (Wikipedia overview) JP: Litigating the Pope’s AI Encyclical with the Lawyers of Scaling Laws Pod — our own reading of it* Aga Khan V, 50th hereditary Imam of the Nizari Ismailis, and “AKU celebrates Class of 2025” — the Aga Khan University convocation where the Chancellor’s message to graduates touched on artificial intelligence and the power of new scholarly toolsChapters(02:06) AI, signaling, and algorithmic monocultures in hiring (25:57) Measuring work: Surge, Mercor, GDPval, and the corporate panopticon (37:48) AI and peer review: “Fighting Fire with Fire” (54:36) Faith and AI; AI as Steroids vs. sneakers vs. coachTranscriptAndrey: One version of the future is a corporate panopticon, where everything we do at work is observed by AIs and measured.Sarah: One of the other things that’s going to differentiate certain jobs from others is how much freedom you have.Seth: Are you above or below the panopticon?Sarah: What’s happening in the hiring space is a little bit different because it’s increasingly mediated by third-party vendors. They claimed that this was unbiased. We found that wasn’t the case.Seth: Would your long-run bet be on match quality and jobs improving in, you know, five, 10 years because of AI?Sarah: I’m optimistic that we will come up with different solutions than the ones that we have now to be able to connect workers to jobs.Andrey: In the Coasean singularity, the AIs will match all of us together efficiently. [laughs]Seth: Welcome to “Justified Posteriors.” Today, we’re pleased to be hosting our colleague, Sarah Bana, of the London School of Economics and Stanford Digital Economy Lab. We’ll be talking about AI and reviewing, AI and Islam, but first, AI and cultural homogeneity. What is cultural homogeneity, and do I have too much of it? I’m Seth Benzell, coming to you from Chapman University in sunny Southern California.Andrey: And I’m Andrey Fradkin, coming to you from San Francisco, California. I’m very curious to learn about the monoculture of podcast guests that exist [all laugh] in the world. Welcome to the show, Sarah. I’m very excited to have you.Sarah: Thanks so much. Longtime listener, first-time caller.Seth: Yeah, it’s so great to have an ascended fan on. Can we briefly ask you what’s your favorite episode before we get into it?Sarah: Oh, okay. That’s a great question. I did like the two-parter that Bo did.Seth: Okay.Sarah: I thought that was really fun. I work in a similar space, and so I’ve been thinking a lot about AI and signaling, and so then, that was definitely a very fun episode. And I think it also got a little bit into metaphysical things, as sometimes you do.(02:06) AI, signaling, and algorithmic monocultures in hiringAndrey: Before we get into specific papers, do you think that AI is already ruining signaling? So the idea here would be that it’s so easy to apply to jobs that this, you know, traditional way of applying no longer works. And I guess the flip side, the counterargument maybe is that ev… Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe

This week we’re joined by Phil Trammell, economist at Stanford and senior researcher at Epoch AI, and co-author (with Chad Jones) of the “When GDP Misleads” paper we covered earlier. Phil is one of the most original thinkers on the economics of AI, and he brought his priors to an unusual assignment: a science-fiction scenario.We read the AI 2040 “Plan A” essay: a future in which the US and China agree to pace and monitor AI development, data centers are placed where they’re easy to bomb (hello, Mongolia), and 2040 arrives with abundance, a citizen’s dividend, and lie detectors. Before reading, the three of us stated our priors on two questions: will an international AI pacing and transparency treaty exist by 2030, and would one with teeth be a good idea? Phil started at two-thirds and 75%; Seth at 30% and 60%; Andrey at 20% and 49%.Along the way: whether the essay’s growth forecast is load-bearing (Phil: it’s close to a wash, and Seth and Andrey push back), why a world where AI looks like nuclear weapons might be easier to coordinate on, Chad Jones’s risk-aversion argument, whether AI itself makes monitoring and commitment cheaper, why monitoring every data center sounds absurd to Andrey and like scrubbers on smokestacks to Phil, mind-reading and privacy, and Seth’s worry that a one-time lump-sum rebate of the universe ends in an extreme power-law distribution of wealth. Then Phil delivers a mini-lecture on renegotiation-proofness, the piece of game theory he thinks the essay is missing, and on why it may be safer to go faster if most risk is state risk rather than transition risk.We close with posteriors, a discussion of Phil’s work on recursive self-improvement (RSI), and a lightning round: why Epoch instead of Stanford or Anthropic, his giving pledge and wild animal welfare, and how a Piketty question from Dwarkesh turned into one of the most-read things Phil has written.Links & ReferencesPhil’s work* philiptrammell.com — Phil’s website · Epoch AI* Charles I. Jones & Philip Trammell, “When GDP Misleads: Inferring Living Standards from the Value of a Statistical Life” — our earlier episode on the paper;* Dwarkesh Patel & Philip Trammell, “Capital in the 22nd Century” * Phil’s Epoch piece on parallelization constraints and his in-progress follow-up on whether RSI is imminent.* Aschenbrenner & Trammell, “Existential Risk and Growth” (a Chad Jones tweak on state risk vs. transition risk) — Leopold’s Situational Awareness.* Cunningham et al., “The Economics of Recursive Self-Improvement” (Elasticity Institute; Tom Cunningham led, Phil is a co-author)The essay and the scenario* AI 2040: Plan A — the global pacing/transparency treaty, data centers in Mongolia and Canada, the “citizen’s dividend”* Dan Kokotajlo and the AI Futures Project, authors of the essay* Lydia Nottingham’s reply to the “if everyone would just” framing of coordination plans.Papers, concepts & people discussed* Chad Jones, “The AI Dilemma: Growth versus Existential Risk” — willingness to take on existential risk as a function of wealth and mortality; our earlier Chad Jones episode* Hall & Jones, on the value of life and the rise in health spending — people get more risk-averse as they get richer* Armstrong, Bostrom & Shulman, “Racing to the Precipice” — why you want a big gap between the leader and the follower* Renegotiation-proof equilibria and the grim-trigger strategy in a repeated prisoner’s dilemma.* C.S. Lewis, The Abolition of Man — technology and the “middle of history” when we’re most free to plan the future* Thomas Piketty, Capital in the Twenty-First Century and Brad DeLong’s early point about capital substituting for labor* Gell-Mann amnesia, which Phil invokes on the verification discussion* Tom Davidson’s work on takeoff speeds and automating AI R&D, e.g. Davidson & Eth); algorithmic efficiency, sample efficiency, and RLVR.Previously on Justified Posteriors* How much should we invest in AI safety? — existential risk vs. growth with Chad Jones* Our “When GDP Misleads” episode.Chapters* (00:00) Cold open* (01:22) Intro, and our priors on an international AI pacing treaty* (08:18) The evidence: a walk through Plan A and its 2040 utopia (Mongolia included)* (11:25) Is the growth forecast load-bearing? Immortality, the FDA, and a unipolar 1990s* (19:06) AI as a normal technology, and could AI make coordination easier?* (24:39) Monitoring every data center, and can the plan last until 2040?* (31:02) Lie detectors, mind-reading, and privacy* (34:38) Concentrated wealth: lump-sum rebates vs. power-law inequality* (37:59) Renegotiation-proofness and Racing to the Precipice* (42:14) Can competition deliver safety? Profits, moats, and a spirited disagreement* (44:48) State risk vs. transition risk: is it safer to go faster?* (46:29) Posteriors* (53:03) Recursive self-improvement: is it imminent?* (1:02:58) Lightning round: Epoch vs. Anthropic, the giving pledge, and the Piketty/Dwarkesh essayJustified 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 and keep your posteriors justified.Transcript:Cold Open [00:00 - 01:22][00:00:00] Andrey: Do we think that we’re gonna have an international AI pacing and transparency treaty by 2030?Phil: In the extreme case where AI was only disruptive, it would be easiest to coordinate a slowdown. What makes it hard is the prospect of abundance if we don’t.Seth: Isn’t this a universe that’s inevitably setting itself up for an extreme concentration of wealth?Phil: How you feel about that depends on whether you think there’s something intrinsically undesirable about wealth inequality. Not sure there is.Andrey: One of the requirements of this plan is to create monitoring devices and monitoring software in every single data center in the world.Seth: Will the Turing police come to my house if I have too nice a graphics card to play my video games?Phil: It seems a bit inconsistent to be worried about that, but then not feel that AI would really lower the cost that would go into a regime to coordinate its development or deployment.Andrey: What if Russia builds a data center and decides to just go for it?Phil: You really want a big gap between leader and followers, that the leader can slow down for safety purposes. But if we’re all sharing information and the coordination is fragile, then when it unravels, we’re as neck and neck as can be. Their proposal to mitigate that, of course, is for us to destroy each other’s big data centers in Mongolia and Canada.Andrey: Suppose countries actually are going to abide by it and it substantially slows down the pace of AI progress. Is that a good or a bad thing?Intro and Our Priors on an AI Treaty [01:22 - 08:18][00:01:22] Seth: Welcome to Justified Posteriors. This week, we have on the incomparable Phil Trammell of Stanford and senior researcher at Epoch AI to talk to us about AI-driven economic growth. But first, we’ll get his views on AI 2040’s Plan A, a vision of a possible future where AI is globally regulated for the good of all. I’m Seth Benzell, always ready to pause my work for a good time, coming to you from Chapman University in sunny Southern California.Andrey: And I’m Andrey Fradkin, coming to you from San Francisco, California. Very excited to have Phil on the show, who I consider one of the most original thinkers in this space, and also one of the most, morally upright. So welcome to the show, Phil.Phil: Oh, thank you. I’ll try to live up to that.Andrey: I did wanna kind of ask for our priors. And in particular, do we think that we’re gonna have an international AI pacing and transparency treaty by 2030?Phil: I think it’s far from a sure thing, but I’ll, I’ll give it two-thirds.Seth: Two-thirds probability. We get a sixty-six percent chance for Phil. I have to say, I put it closer to thirty percent. I think there’s a real probability, but I think there are some big, implementation problems, some of which are discussed in this, essay and some of… Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe

This week, we’re joined by David Deming, top labor economist, blogger at the Forked Lightning Substack, and — since 2025 — Undergraduate Dean of Harvard College, to ask how the value of social skills will change in the age of thinking and talking machines. We start with his QJE paper “The Growing Importance of Social Skills in the Labor Market,” which models social skills as a factor that reduces coordination costs, plus his new work arguing that meetings must be important because Norwegian workers spend about 12% of their hours in them. We ask how the coming Coasean singularity — the ability of AIs to communicate and bargain on our behalf — will change all this; David thinks that making routine communication cheap just frees up more time to argue about the questions that have no right answer. We then turn to how this research informs his leadership at Harvard. Andrey asks about his research with Raj Chetty and John Friedman, which shows that an Ivy-Plus admission is a lottery ticket for joining the elites, and we talk about the cause of that finding, and how AI might change it. David argues that although social skills will be increasingly important, admissions shouldn’t select on them until we can measure them. We discuss possible new approaches to integrating AI into schools, including controlled first drafts, voluntary Yondr pouches, and oral exams. We close the discussion of Harvard with the faculty’s vote to cap A’s at 20% plus four.The lightning round covers how he would allocate his social skill RPG stats, why ChatGPT use turns out to be “broad but shallow,” what economists get wrong because they lack social skills, and a tribute to David’s advisor Larry Katz. Most interesting in Seth’s opinion was David’s decision to favor composure and charisma over manipulation in RPG stats; in the bulk of the interview, he seems to suggest that raw charisma is less important than the ability to coordinate and smooth over differences in a work context, which would be closer to manipulation. Perhaps downplaying the importance of some social skills is the most important social skill of all…LinksOur guest* David Deming’s website and his Substack, Forked LightningSocial skills* Deming, “The Growing Importance of Social Skills in the Labor Market”, QJE (2017) — the paper we open with: social skills as reduced coordination costs in team production, tested in the NLSY and in occupational data, 1980–2012* Becker & Murphy, “The Division of Labor, Coordination Costs, and Knowledge”, QJE (1992) — why the economy isn’t one big team* Deming, Løken, Willén & Xu, “Meetings”, NBER WP 35706 (2026) — the Norway paper: meetings take ~12% of work hours and ~14% of firm wage bills, and meeting intensity correlates with wage growth* Shahidi, Rusak, Manning, Fradkin & Horton, “The Coasean Singularity? Demand, Supply, and Market Design with AI Agents”, NBER WP 34468 — Andrey’s paper, which (note the question mark) does not predict that it will actually happenYudkowsky, “Coherent Extrapolated Volition” (2004) — the rationalist term Andrey reaches for* Deming, “Efficient social learning is the human advantage over AI” — the “last human skill” argument: sample-efficient humans hold a comparative advantage in sparse, high-context tasks, and nothing is higher-context than social lifeAI at Harvard (plus admissions and the Ivy-Plus premium)* Chetty, Deming & Friedman, “Diversifying Society’s Leaders? The Determinants and Causal Effects of Admission to Highly Selective Private Colleges” (NBER version) — waitlist design: attending an Ivy-Plus college rather than a flagship public raises the chance of reaching the top 1% by about 50% and nearly triples the chance of working at a prestigious firm. The paper also identifies athletic recruitment as one of three drivers of the high-income admissions advantage, relevant to my editorializing at 30:22.* Derek Thompson, “Why Harvard’s Dean ‘Encourages’ Students to Use AI” — David’s longer interview on the same themes* Yondr — the phone pouches David bought as a voluntary commitment device* Panos Ipeirotis, “Fighting Fire with Fire: Scalable Personalized Oral Exams with an ElevenLabs Voice AI Agent” — the NYU professor’s AI oral examiner Andrey mentions (36 oral exams for about $15 total)Grade inflation* “70% of Faculty Vote to Overhaul Harvard Grading With A Cap” — Crimson on the 458–201 vote capping A’s at 20% plus four, starting fall 2027Lightning round* A note on my character sheet. Sharp-eyed Storytellers will notice that the social attributes I read out at 52:41 (Charisma, Manipulation, Composure) come from no published edition. Classic World of Darkness has Charisma / Manipulation / Appearance. The 2004 Storytelling System behind Chronicles of Darkness renamed Charisma to Presence and swapped Appearance for Composure. This was, of course, entirely intentional. I kept the new edition’s Power / Finesse / Resistance structure and restored the classic Charisma (Presence sounds like a meditation app). Call it a house rule. * Chatterji, Cunningham, Deming, Hitzig, Ong, Shan & Wadman, “How People Use ChatGPT”, NBER WP 34255 (2025) — the source of “broad but shallow”* Garicano, “Hierarchies and the Organization of Knowledge in Production”, JPE (2000) — the knowledge-hierarchy framework David thinks AI will reshape* Baker, Gibbons & Murphy, “Relational Contracts and the Theory of the Firm”, QJE (2002), and Gibbons & Henderson, “Relational Contracts and Organizational Capabilities”, Organization Science (2012) — “highly rated but still underrated”* Deming, “Early Childhood Intervention and Life-Cycle Skill Development: Evidence from Head Start”, AEJ: Applied (2009) — the Larry Katz labor-class paper that became David’s first publicationConnections to previous Justified Posteriors episodes:* JP: Seb Krier on AGI, the Coasean Singularity, and EDM — the optimistic case for agent-mediated bargaining at scale* JP: Noah Smith on Blogging, AI Economics, and Elite Overproduction — Coasean utopia vs. the collapse of all meaning (answer: why not both?)* JP: Does AI Cheapen Talk? (Bo Cowgill Pt. 1) — makes the point that cheap communication so far means more bad communication, not more good* JP: Alex Imas — Demand Collapse, Bargaining with Machines, and Behavioral AI Economics and Imas, “What will be scarce?” — the demand-side “relational sector” argument David contrasts with his supply-side story* Hackenburg et al., “Scaling language model size yields diminishing returns for single-message political persuasion”, PNAS (2025) — the “AI persuasion maxes out” result I ask about, and our episode on it, Scaling Laws Meet PersuasionChapters(01:25) Social skills: coordination costs, the Coasean singularity, and “the last human skill” (20:39) AI at Harvard, teaching and selecting for social skills and the Ivy-Plus lottery ticket (46:05) Fighting grade inflation (52:41) Lightning round: Social Skills as RPG stats, “How People Use ChatGPT,” whether economists need better social skills, and Larry KatzTranscriptDavid: I’m making an argument this is like the last human skill.Andrey: So should Harvard just stop teaching hard skills and go all in on social skills?David: No.Andrey: Damn it. I thought I got you. [laughs]Seth: Nobody’s going to have to communicate anything. The AI will tell each other, and then the AI will just tell me what to do, and no one will ever have to go to a meeting.David: My guess is it will decrease the quantity of communication but increase the quality of it. AI will not have the meeting that tells you what the company’s strategic direction will be over the next two years, because there’s no right answer to that.Andrey: If we all do have a lot more free time, then the most important thing that we should be doing is arguing over what the big AI system should be doing for us.David: We spend a lot more time firing takes at each other and arguing about things, in part because most work is not necessary for human survival. Most of us are entertainers. It would seem crazy to our ancestors that w… Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe

We’re honored to host Daron Acemoglu, Institute Professor at MIT and 2024 Nobel laureate. Daron has spent a decade arguing that AI should be built to make human expertise more valuable rather than to replace it, most recently in “Building Pro-Worker AI” with David Autor and Simon Johnson, and in a Nature essay on why we should stop talking about AGI. He also has a new book out, What Happened to Liberal Democracy?.We came to steelman pro-worker AI and to find the tensions in it. The big question: is technology even directable, or is AGI the golden path to everything? Daron says technologists give him three reactions — “we’re already doing it,” “it’s impossible,” and “just wait for AGI” — and admits he can’t disprove any of them. From there: we discuss comparative advantage under superintelligence (Ricardo doesn’t save you if the other side is cheap enough), what he’d do if he were Dario (build a production line for applications, and move the knob from 95/5 toward 80/20 or 60/40), and why he doesn’t trust the market to sort this out.We close with capital taxes and pensioners, whether Daron would join the AI-pause coalition (”slow down and pause are not the same thing”), what the US should learn from China on applications and classrooms, how he’d update The Simple Macroeconomics of AI for agentic AI. Lastly, we ask him what he really thinks of that Economist piece about him. “They’re the elite. I’m the working class.”Links & ReferencesDaron’s recent work* Daron Acemoglu, David Autor & Simon Johnson, “Building Pro-Worker AI” — The Hamilton Project / Brookings, February 2026 (PDF · NBER w34854).* Daron Acemoglu, “Why we must stop talking about artificial general intelligence — and instead build ‘pro-worker’ AI” — Nature, August 2026* Daron Acemoglu, What Happened to Liberal Democracy? Remaking a Politics of Shared Prosperity — Penguin Random House, August 2026 (MIT News)* Daron Acemoglu, “The Simple Macroeconomics of AI” — Economic Policy, 2025 (NBER w32487); * Conversations with Tyler: Daron Acemoglu on Liberalism, Automation, and the Educated Elite — August 2026;The task-based framework* Daron Acemoglu & Pascual Restrepo, “The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment” — American Economic Review, 2018; * Daron Acemoglu & Pascual Restrepo, “Automation and New Tasks: How Technology Displaces and Reinstates Labor” — Journal of Economic Perspectives, 2019; * Daron Acemoglu, Andrea Manera & Pascual Restrepo, “Does the US Tax Code Favor Automation?” — Brookings Papers on Economic Activity, 2020 (NBER w27052);* Daron Acemoglu & Pascual Restrepo, “Automation and Rent Dissipation: Implications for Wages, Inequality, and Productivity” — NBER, 2024; Other concepts, papers & people discussed* Rich Sutton, “The Bitter Lesson” — the technologists’ argument that general, compute-heavy methods always beat specialized ones* David Ricardo and comparative advantage — “Why did England trade with Portugal?” “Because nothing was very, very cheap.”* Dell’Acqua, McFowland, Mollick et al., “Navigating the Jagged Technological Frontier” — the source of the “jagged frontier” (Ethan Mollick’s term, not Josh Gans’s)* Bastani et al., “Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics” — PNAS, 2025; one of the classroom studies Daron has in mind* Emmanuel Saez & Gabriel Zucman, “Wealth Inequality in the United States since 1913” vs. Smith, Zidar & Zwick, “Top Wealth in America: New Estimates under Heterogeneous Returns” — the wealth-inequality measurement debateThe Economist piece* The Economist, “The world’s most influential economist is oddly unconvincing” Previously on Justified Posteriors* The Simple Macroeconomics of AI — our December 2024 episode on Daron’s paper.Chapters* (00:00) Cold open: three reactions from technologists* (00:30) Intro — “help me or destroy me”* (02:00) What is pro-worker technology? AI is not a TFP term* (04:20) The general-equilibrium objection: new tasks in industry A, lower wages in industry B* (08:56) Why care about the labor share if productivity is soaring? Three answers* (10:04) Solow vs. AK: why not aim for growth forever with a labor share of zero?* (11:37) Labor must be essential: dignity, self-worth, and “why would anybody listen to them?”* (12:59) Is technology directable? Sutton’s bitter lesson and the AGI golden path* (13:59) “We’re doing it already,” “it’s impossible,” “just wait for AGI” — and why Daron can’t disprove any of them* (14:43) Is Anthropic’s “augmentation” pro-worker? Why the taxonomy can’t be read off the technology* (15:37) AGI → ASI: nothing can be pro-worker; relational goods; Ricardo, Portugal, and the subsistence wage* (18:29) If you were Dario: build a production line for applications, and move the knob from 95/5* (20:26) Andrey pushes back: aren’t tons of companies already building applications?* (21:36) The pre-agentic age had remarkably few usable applications; where the talent and compute are going* (22:31) Will the market sort it out? “I haven’t convinced you that there is a need for proposals”* (23:28) Nobody is making money: capital allocated on “future projections, vibes, and beliefs”; sovereign wealth looking for a place to park* (25:32) If there’s an overabundance of capital, why isn’t pro-worker AI funded? Data centers and talent scarcity* (26:47) Two standard market failures: the capital-biased tax code and excessive automation under labor-market rents* (29:51) Who holds the capital? Life-cycle savings vs. the very rich; Saez–Zucman vs. the skeptics* (31:10) Is the long-run supply elasticity of capital infinite? The data says moderate; OLG vs. representative agent* (32:25) Having your cake and eating it too: equal treatment of capital and labor raises GDP in the calibration* (33:56) How close is AGI? “Not that near,” and not the productivity-maximizing strategy either* (34:23) Would Daron join the pause coalition? Safety vs. epsilon-from-AGI; hitting the brake to steer* (37:45) China, classrooms, and the missing US regulatory infrastructure* (39:11) Updating The Simple Macroeconomics of AI: agentic AI raises the numbers, diffusion is still in the money* (40:28) The three real bottlenecks: engineers, organizations, and the jagged frontier* (43:25) AI in Turkey: public use, lagging firms, and the state* (44:44) Acemoglu memes* (45:31) That Economist article: “if you’re going to have a hit piece against you, pray that it’s an incompetent one”TranscriptCold Open & Introductions [00:00 – 02:00][00:00:00] Daron: Whenever I talk about pro-worker AI, which I’ve been doing quite often for about ten years, when I talk to CEOs or managers of companies, they are on board. Whenever I talk to technologists, I get three reactions. One is, “We love it. We’re doing it already.” Second, “It’s impossible.” Third, “Yes, just wait for it. We’ll get to AGI first.”[00:00:30] Seth: Welcome to Justified Posteriors. This week, we’re honored to host Nobel laureate Daron Acemoglu of MIT. He’ll answer the question “What happened to liberal democracy?” — the title of his new book. But first, we’re excited to learn about his evolving views on AI, and especially to pin him down on what pro-worker technology actually means. I’m Seth Benzell, always racing with the machine — and I don’t mean using a treadmill — at Chapman University in sunny Southern California.Andrey: And I’m Andrey Fradkin. Very excited to have Daron on the show. He was an inspiration for me going to economics grad school, so this is truly an honor. Coming to you from San Francisco. Welcome to the show.Daron: Thank you, Andrey. Thank you, Seth. This is my true pleasure and honor.[00:01:14] Seth: Thank you, Daron. Okay, so we have limited time, so we’re going to jump right into it. In a recent series of essays, including a paper with David Autor and Simon Johnson called “Building Pro-Worker AI,” and a new piece out in Nature about why we shouldn’t be talking about AGI, you’ve been calling for this thing called pro-worker AI. And of course, we love workers here on o… Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe

This week, we’re joined by Martin Beraja of UC Berkeley Haas, one of the most prolific macroeconomists working on AI. We start with his newest paper “Growth at the Resource Frontier” — with Noam Yuchtman — which argues that NYT v. OpenAI is just the latest instance of a very old pattern — innovators appropriate resources nobody had bothered to assign property rights to (oil, the electromagnetic spectrum, ad data), and the courts legitimate it afterward. Could it be that unclear property rights are actually a catalyst for innovation?We then turn to Martin’s China work (including “AI-tocracy” — with Andrew Kao, David Yang, and Noam Yuchtman, Quarterly Journal of Economics, 2023), where facial-recognition firms winning data-rich surveillance contracts went on to produce more commercial software, and where local unrest predicts AI procurement which then suppresses further unrest — a feedback loop that cuts against the standard “autocracy taxes innovation” story. Andrey pushes back on this hard. Third: AI as a codification technology, successor to the scientific manual and the McDonald’s franchise binder — if new entrants were born with the organizational knowledge of forty-year-old firms, Martin’s new paper “The Value of Organizational Learning Technologies,” with Eduard Talamás, argues that US GDP would roughly double. Lightning round covers saltwater versus freshwater economics, Bay Area agglomeration externalities, and why AI displacement may be less inefficient than robot displacement.Relevant Links* Martin Beraja’s website and X account* “Growth at the Resource Frontier” — with Noam Yuchtman, March 2026 (the Summer Institute paper)* “Data-intensive Innovation and the State: Evidence from AI Firms in China” — with David Yang and Noam Yuchtman, Review of Economic Studies, 2023 (published version)* “AI-tocracy” — with Andrew Kao, David Yang, and Noam Yuchtman, Quarterly Journal of Economics, 2023 (NBER version)* “The Value of Organizational Learning Technologies” — with Eduard Talamás, March 2026* “Inefficient Automation” — with Nathan Zorzi, Review of Economic Studies, 2024* “Fast and Slow Technological Transitions” — with Rodrigo Adão and Nitya Pandalai-Nayar, JPE Macro, 2024* “Exporting the Surveillance State via Trade in AI” — with Kao, Yang, and Yuchtman, 2025* The New York Times Co. v. Microsoft Corp. et al. — the complaint that started the argument in London; OpenAI’s response* The Green Mountain Boys and Ethan Allen — America’s tradition of appropriative innovation* “Pacing the Frontier” — the July 2026 statement from frontier-lab employees on coordinated slowdown toolsTimestamps(01:10) “Growth at the Resource Frontier” — what the property-rights tradition misses (09:13) Why not just Coase it? DuPont, Conoco, and the plastics counterexample (15:10) Ethan Allen, the Green Mountain Boys, and conflict at the literal frontier (25:25) “Data-intensive Innovation and the State” — surveillance data and commercial AI (34:54) “AI-tocracy” (40:40) “The Value of Organizational Learning Technologies” (48:19) If entrants were born old, US GDP would roughly double (59:26) Lightning roundTranscriptMartin: The demand for surveillance that the Chinese state had was allowing firms to have access to large quantities of data that they wouldn’t have had otherwise, and hence allowed them to innovate more in this type of technology — facial recognition — that needed a lot of data to be successful.Andrey: I’m gonna push back on this.Martin: I like that Seth is the good cop and Andrey is the bad cop in this. [laughs]Seth: Welcome to Justified Posteriors. Today we’re joined by Martin Beraja, a top economist at Berkeley who has exploded onto the scene with a raft of influential macroeconomics of AI papers. He’ll be talking to us about AI-enabled dictatorship, economic growth when no one knows how AI companies are allowed to use the data they hoover up, and so much more. I’m Seth Benzell, always ready to fight to appropriate a share of the world’s knowledge, coming to you from Chapman University in sunny Southern California.Andrey: And I’m Andrey Fradkin, coming to you from San Francisco, California. Welcome to the show, Martin.Martin: Hi, Seth, Andrey. Thank you for having me. It’s a pleasure.(01:10) “Growth at the Resource Frontier” — what the property-rights tradition missesSeth: It’s lovely to have you on. We saw you present some of this work at NBER Summer Institute, and we were already talking about having you on, but it was so exciting we just had to get you scheduled.The first thing we wanted to talk to you about is the work you presented there — a paper in progress called “Growth at the Resource Frontier.” Maybe a good place to start us off: what do you think the previous literature misses about economic growth that this paper is designed to capture?Martin: I would say that something we teach undergrads, and even later on, is the idea that having secure property rights is good for innovation and economic growth. There’s been plenty of evidence and plenty of theoretical models in that tradition. In fact, when you read reports from the IMF or the World Bank or any international organization — I come from Argentina, a middle-income country — a lot of those reports say, “The problem with these countries is that they do not have good property rights institutions.” That’s a very strong tradition in economics.Then there’s another tradition, more on the innovation side, that asks what incentivizes innovators to innovate. The key idea there is that you need to give innovators some form of monopoly rents, either through patents or through trade secrets. They need to be able to exploit their ideas in the market, and those rents must not be competed away — otherwise they wouldn’t innovate. That idea, that rents in the product market drive innovation, is in basically all the models of innovation we have, in one way or another.Seth: So what’s missing? [laughs]Martin: When we started with this paper — it’s kind of the story — we began with a conversation with Noam Yuchtman, my co-author. He’s in London, and I think it was just about when the New York Times had filed its complaint against OpenAI, accusing them of appropriating all the Times articles and using them to train ChatGPT. So we’re debating this, and Noam was more like, “Obviously this is terrible that they appropriated all this.” I’m caricaturing the conversation a bit, but— [laughs]Seth: No, I love that. Make your co-author the straw man. That’s good.Martin: Yeah, there’s some drama. And obviously, it was illegal, blah, blah, blah. And I said, “Well, Noam, surely you think the innovation wasn’t just the algorithms behind the LLMs and the training, but also the idea that we could use all this public data” — in this case, public or not, that’s the question — “under some notion of fair use, to train LLMs at scale.” That was the big conceptual and technological breakthrough: that more data was enough, at the end of the day.So we started there, thinking about what that conceptual breakthrough was. It was the idea that something previously not recognized as valuable — online text, images, whatever was out there — could suddenly become a new resource, namely training data for LLMs. And once we made that connection, we started to realize: hold on, this has been going on forever in human history. A lot of the history of human progress was really about expanding the resource frontier. In fact, historians name ages after their materials — the Iron Age, the Bronze Age—Seth: Ah.Martin: —the Stone Age—Seth: Ah.Martin: —and so on, up to more modern eras where knowledge became important and, through the Enlightenment, the idea that we could master nature.Seth: If there were secure property rights over iron ore, the Hittites would have never conquered the Near East.Martin: Right. Well, we’re gonna get there—Seth: We’ll get there. [laughs]Martin: We’ll get there. And then we say: oil was another example, where we brought oil into the resource frontier. Th… Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe

In this episode, we discuss “When GDP Misleads: Inferring Living Standards from the Value of a Statistical Life”, written by Stanford/Anthropic economist Chad Jones and Stanford/Epoch economist Philip Trammell. The word “AI” never appears in it, yet every argument about whether AI will show up in the statistics runs through the question of whether the statistics were ever measuring the right thing.We start with priors. Has real GDP per capita over- or underestimated welfare growth? Andrey’s case for over: China’s real GDP per capita is up over 70 times since the early 1950s, and life is not 70 times better.The paper poses a dilemma, suppose you have two goods (food and string quartets) with different productivity growth. The counterintuitive result in the paper is that inventing the low-productivity good makes people better off while slowing measured GDP growth. The paper then proposes to use the value of a statistical life as a measure of welfare. It is the nominal price for being alive, so its growth rate, deflated by the marginal utility of consumption and pinned down by the intertemporal Euler equation, gives you the growth rate of lifetime utility. Seth likes it, while Andrey explains why dissatisfaction with the Euler equation was in the top three reasons he didn’t become a macroeconomist.The results: welfare growth of 2.3% a year since 1940 — and a decline from 1980 to 2024. We get into how VSL is actually estimated (hedonic wage regressions identified off coal miners and other people with unusual preferences about dying), the 1% discount rate that arrives uncited in a one-sentence paragraph, whether mortality risk is being double-counted, and the robustness table where moving the interest rate by a single percentage point in either direction swings the 1940–2020 welfare gain.Links & ReferencesThe paper* Charles I. Jones & Philip Trammell, “When GDP Misleads: Inferring Living Standards from the Value of a Statistical Life” — headline results: 2.3%/year welfare growth 1940–2024 (6.9×), negative 1980–2024; robustness range 16.1× to 3× on a ±1pp interest-rate change* Charles I. Jones — Stanford GSB* Philip Trammell — Global Priorities Institute, Oxford* Charles I. Jones, “The AI Dilemma: Growth versus Existential Risk” — the previous Chad Jones paper we covered; bounded utility drives an ever-widening wedge between welfare and GDPThe value of a statistical life* US DOT departmental guidance on VSL in economic analysis — the paper cites $13.7 million for 2024 (we said $14.5 million on air; see Corrections)* Dora L. Costa & Matthew E. Kahn, “Changes in the Value of Life, 1940–1980” — Journal of Risk and Uncertainty, 2004; the compensating-wage-differential estimates the paper’s VSL time series rests on. “I don’t think any value of statistical life paper is very good.”Concepts discussed* The intertemporal Euler equation — and Andrey’s objections: it fails for a large enough subset of people, individual Euler equations don’t aggregate into a linear functional form, and it implies Ricardian equivalence, which is “so provably false as to invalidate this entire approach”* New goods and variety growth — the food-and-string-quartets example; why the moment of invention (price falling from infinity to finite) is the hard part of any variety adjustment; and the composite-smartphone problem in quality adjustment* Robert Nozick’s experience machine — the eudaimonia button, and what it would mean for measured welfare* Derek Parfit on personal identity and future Tuesday indifference — is your discount rate even something a social welfare function should respect?* Charles I. Jones & Peter J. Klenow, “Beyond GDP? Welfare across Countries and Time” — AER 2016;* Trammell’s bull case for AI welfare: not more stuff per person, but vastly more beings capable of having utility — a total-utilitarian argument that this paper’s single representative agent can’t representPreviously on Justified Posteriors* How much should we invest in AI safety? — our earlier Chad Jones episode (existential risk vs. growth)Corrections* VSL figure: On the episode we said the US DOT value of a statistical life was $14.5 million for 2024. Jones & Trammell cite $13.7 million for 2024 (DOT guidance; current DOT table also lists $13.7M for 2024 / $14.2M for 2025). The slip doesn’t affect the paper’s growth-rate results.* China GDP multiple: On the episode we said China’s real GDP per capita was up over 50× and as high as 72× since 1952/1962. A cleaner figure: 2025 real GDP per capita was about 82× its 1962 level.Chapters* (00:00) Cold open: would you rather be middle-class today, or the king of China?* (00:27) Intro — the paper, and why an AI podcast is covering a paper that never says “AI”* (02:12) Priors: has GDP per capita been a good proxy for welfare?* (03:47) Everything good is correlated with GDP — until you look closely* (05:29) China, 1952 to today: 72× GDP per capita. Is life 72 times better?* (06:32) Two concerns: diminishing returns, and growth in varieties* (07:08) Over or under? Andrey’s split verdict on China and the US* (07:43) 116% since 1980 — “they already had pinball machines”* (08:51) Seth’s prior: diminishing returns dominate, and why the AI age might flip the sign* (10:09) Haven’t we already had huge variety growth? Podcasts, Prairie Home Companion, and the eudaimonia button* (11:15) Putting numbers on it: 95% and 80% that GDP still overstates* (11:36) How much better is life since 1986? Andrey says 25%* (12:12) The benchmark: life expectancy × log consumption, and 41% since 1980* (13:25) The paper’s setup: food, string quartets, and a productivity gap* (15:05) Why inventing the new good makes us better off and slows measured growth* (16:37) Quality adjustment, and the composite-smartphone problem* (17:08) The problem that exists even before invention: satiation* (17:41) Varieties vs. abundance: the king of China, at length* (19:01) Trammell’s actual bull case: more beings, more utility* (19:44) The clever idea: the value of a statistical life as a nominal price for being alive* (21:10) $14.5 million — but 14.5 million what?* (21:50) The deflator problem, Weimar Germany, and the marginal utility of consumption* (24:12) “Micro or macro?” — the Euler equation and the conditions it needs* (25:54) Discount rate vs. mortality risk — is something being double-counted?* (27:17) The magic equation, in two equivalent forms* (28:59) “The intertemporal Euler is getting some intertemporal shade”* (29:40) Ricardian equivalence, aggregation, and “I have told Chad this”* (31:04) Seth’s defense: welfare on the left, nominal on the right* (31:49) How VSL is actually measured: Costa & Kahn, and the death risk of coal mining* (33:04) Identification off the highest-risk jobs — and the people such jobs attract* (34:44) “It’s just made up”: the $14.5M highway-safety number* (35:22) The other inputs: a 1% discount rate, uncited, and T-bills plus a convenience yield* (35:53) The results: 2.3% a year since 1940* (37:19) …and negative since 1980. Life peaked in 1984* (38:05) 6.9× vs 2.4×: could you convince me life is seven times better than 1940?* (39:27) “For whom?” — the representative agent, the 10th percentile, and changing demographics* (41:19) Should a social welfare function respect your discount rate at all? Parfit and future Tuesday indifference* (42:07) Posteriors* (43:37) The robustness table: one percentage point, 16.1× or 3×* (45:10) Sign-off Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe

Justified Posteriors reads “Writing Code vs. Shipping Code” by Mert Demirer, Leon Musolff, and Liyuan YangIn this week’s episode of Justified Posteriors, we update our beliefs with evidence from an ambitious new paper estimating the impact of AI on software production productivity. Demirer (friend of the show), Musolff, and Yang combine public GitHub records for over 100,000 developers with confidential Microsoft data to trace the effect of distinct generations of AI coding tools — autocomplete, sync agents, and async agents — on the code production hierarchy: lines of code, files, commits, pull requests, projects, and releases. The main empirical finding is attenuation of the effect of AI at each step. Enormous gains of 1000% productivity increases or more at the top of the chain translate into about a 30% increase in shipped releases. The model they use to explain this result is closely connected to Kremer’s O-ring logic, which regular listeners will recognize as from a few episodes back. Seth likes the spirit of the model, but feels it is overcomplicated for this context, for reasons he explains. In addition to discussing the data analysis, Andrey and Seth have a good back-and-forth about what we can conclude from it and extrapolate to the economy more generally. The implication that grabbed Seth’s attention is a sentence in this paper’s abstract. Nested in the summary of the careful empirical exercise is an estimated elasticity of substitution of 0.25 between AI and human effort. Big if true! Seth points out the enormous long-run implications of humans being complements to AI in what seems to be the most AI-friendly of tasks: as AI gets cheaper, the human share of income goes up, wages skyrocket, and ultimately AI boosts jobs instead of taking them. That’s a huge real-world hook. Seth and Andrey discuss whether, and if so how much, we update our beliefs in this direction, with Andrey being careful to point out the difficulties of extrapolating from a partial equilibrium elasticity to long-run macro consequences. This episode is sponsored by Revelio Labs — a great source of labor economics data for academics and firms. Now available on WRDS.Priors → PosteriorsPrior 1: Does access to AI coding tools boost lines of code written by more than 100%?* Seth: 95% → 99%. Seth came in confident and left more so. Great to see giant numbers.* Andrey: 80% → 95%. A high yes, hedged because “which developers” and “which tools” do a lot of work in that sentence.Prior 2: Does AI boost economic value by 50% or less of the factor by which it boosts lines of code?* Seth: 95% → 97.5%. I have personally produced a great deal of economically worthless code lately. * Andrey: 85% → 95%.Prior 3: Are AI coding tools a gross complement to human labor?Seth’s answer depends on the level of aggregation:The average normie programmer 20%→20% (unchanged)A human engineering department 33%→40%A software company / open source project 60%→85–90%The economy as a whole 33%→33% (unchanged)Andrey: 75% complement at the sectoral level, and he’d put it as low as the programming department — because right now the code that comes out is not shippable without substantial human input. Posterior: still a complement, mildly supported. He declines, on the record and repeatedly, to extrapolate to the macroeconomy. No fun!ReferencesThe paper under review* Mert Demirer, Leon Musolff & Liyuan Yang, “Writing Code vs. Shipping Code: Productivity Effects Across Generations of AI Coding Tools,” NBER Working Paper 35275 (May 2026). * The authors’ own summary: “Writing code versus shipping code”, VoxEU, June 2026. Prior work by the same team* Kevin Zheyuan Cui, Mert Demirer, Sonia Jaffe, Leon Musolff, Sida Peng & Tobias Salz, “The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers,” Management Science (2026). 4,867 developers, roughly a 26% increase in completed tasks, larger gains for the less experienced. Related Research and Prior Episodes* METR, “Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity” — the RCT Andrey refers to, in which 16 experienced developers took 19% longer with AI while believing they were 20% faster. arXiv version.* METR’s own update, “We are Changing our Developer Productivity Experiment Design” (Feb 2026) — developers increasingly refuse to be randomized into working without AI, which biases the estimated speedup downward. * Michael Kremer, “The O-Ring Theory of Economic Development,” Quarterly Journal of Economics 108(3), 1993. Our episode on it: Weak Links, Strong Predictions: Kremer’s O-Ring at 30.* Josh Gans & Avi Goldfarb, “O-Ring Automation,” NBER Working Paper 34639 — and our conversation with Avi: Avi Goldfarb on Prediction Machines, O-Ring Tasks, and How AI is Reshaping Economics.* Aaron Chatterji, Tom Cunningham, David Deming, Zoë Hitzig, Christopher Ong, Carl Shan & Kevin Wadman, “How People Use ChatGPT,” NBER Working Paper 34255 — a reference for Andrey’s point about substitution toward home production. Non-work usage grew to over 70% of messages; computer programming is a small share.* Our episode on the same phenomenon in a different market: The Economics of Book Slop — more books, not obviously more valuable books. More apps, not obviously more downloads. Rhymes.* The universal token budget: Alex Imas — Demand Collapse, Bargaining with Machines, and Behavioral AI Economics and Seb Krier on AGI, the Coasean Singularity, and EDM.Table of Contents* Introduction and Today’s Paper — [00:00]* Priors: [02:58]* The Evidence: Ideal and Possible Experiments— [18:34]* The Evidence: Adoption Events and the Attenuating Waterfall — [24:56]* The Evidence: Model and Simulation Results — [42:56]* Aggregates, App Stores, and Posteriors — [56:00]TranscriptIntroduction and Today’s Paper [00:00]Seth: The abstract of the paper has the following sentence: blah, blah, blah, based on the results in the model, there’s an estimated elasticity of substitution of 0.25, so high complementarity between AI and human effort, which indicates strong complementarities. Wow. As the people say, big if true. Humans and AI, 0.25 complements. Everybody worried about AI taking all our jobs — wrong. All labor share to 100%.Welcome to the Justified Posteriors podcast, the podcast that updates beliefs about the economics of AI and technology. I’m Seth Benzell, with a 0% productivity impact on my code writing, as measured by my podcast release schedule, coming to you from the Pocono Mountains of eastern Pennsylvania.Andrey: And I’m Andrey Fradkin, coming to you from San Francisco, California. Justified Posteriors is sponsored by the fine folks at Revelio Labs, and please do sign up to our podcast and our Substack whenever you get the chance.Seth: Today we’re talking about a really interesting empirical study investigating the impact of AI tool use — autocomplete, synchronous agents, asynchronous agents — on people’s productivity in writing code. This is that kind of hard empirical data that maybe has the potential to move our beliefs. So I’m cautiously optimistic that I’m going to learn a lot from this one.Andrey: It’s the big question, in many ways. We have these tools. We’re using them. What do we get out of them? Are we really that much more productive? That is the question on everyone’s mind, especially since so many of these tools are very costly. There are people token maxing under the belief that the more tokens that are used, the more valuable the output will be.Seth: People are going broke over the tokens. We discussed a universal token budget when Alex Imas was on — or maybe that was with Seb. What are people getting when they’re actually paying for them? The paper we read to look into this is called “Writing Code vs. Shipping Code: Productivity Effects Across Generations of AI Coding Tools,” from Demirer, Leon Musolff, and Liyuan Yang. So, butchered every name, as is common for us.Andrey: Mert is a co-author of the pod, so very excited to be… Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe

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… Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe

In this episode of Justified Posteriors, we host Alan Rozenshtein and Kevin Frazier — the law-professor duo behind Lawfare’s Scaling Laws — to take two of the most-discussed AI policy documents of the spring and subject them to an inquisition. Our disputors are probably not what Pope Leo anticipated: two lawyers, two economists, and probably 3/4ths Jewish. Talk about a crossover episode!First up is Pope Leo XIV’s 42,000-word encyclical (that’s Pope-talk for letter) on artificial intelligence. Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence lays out 5 principles of Catholic social teaching, and then explains how this should shape Catholicism’s approach to AI. We focus on two in particular. The first is subsidiarity, which Seth summarizes as Catholic federalism, the idea that most decisions should be made at as local a level as possible. We discuss both the economic argument for this, but also what the Pope adds to Hayek: Decentralization not for efficiency’s sake, but a kind of ennoblement, the dignity of deciding things locally. The second is the universal destination of goods, which the encyclical extends to “immaterial goods”. This leads to the positive argument of the Pope - that AI should be undertaken as a communal project with decentralized power and discussion, rather than a technocratic “Tower of Babel” that will lead to ruin and division. Much of our disputation focuses on whether these principles actually resolve the important questions. Is the Pope rightfully cautious about an emerging technology, or was this an opportunity to take a stronger stand on what constitutes AI Sin? Interestingly, the Pope’s strongest stand is against transhumanism, which would be a plausible resolution to the dialectic of “Butlerian Jihad” vs. worship of a new machine god. Then we pick up DeepMind’s “Positive Alignment” paper, and the economists get grumpier. Andrey complains that the paper is vacuous, failing to take a stand on actual practical goals or methods. But it sets us up for a good conversation about several issues: Such as liberalism of fear, a type of anti-utopian liberalism; whether “flourishing” is something you can A/B test towards; and where the ‘constitution’ metaphor behind Constitutional AI works vs. breaks down. We also tease a joint project, “SCOTUS Bench,” a new benchmark for evaluating AIs’ ability to predict appeals court outcomes. Watch this space for more on that soon. Related Links* Scaling Laws — Alan and Kevin’s AI, law, and policy podcast at Lawfare* Alan Rozenshtein on X: @ARozenshtein · Kevin Frazier on X: @KevinTFrazier* Magnifica Humanitas — Pope Leo XIV’s first encyclical, “On Safeguarding the Human Person in the Time of Artificial Intelligence,” in full, straight from the Vatican* “Positive Alignment: Artificial Intelligence for Human Flourishing” — the DeepMind-led paper (Laukkonen, Krier, et al.) arguing alignment should optimize toward flourishing, not just away from harm* Claude’s Constitution — Anthropic’s ~20,000-word statement of Claude’s values and character, released under CC0* “Claude’s Constitution,” with Amanda Askell — the Scaling Laws interview with the document’s primary author (the one we keep saying we’re jealous of)* The Moral Machine — MIT Media Lab’s crowdsourced trolley-problem experiment: millions of judgments on the grandma-versus-criminals ratio* Meta’s Oversight Board — the “Supreme Court of Facebook,” and Kevin’s cautionary tale in institutional design* Andrew B. Hall — Stanford political economist on deliberative democracy, platform governance, and what went wrong with the Oversight Board* The Anthropic Economic Index — the adoption data behind the “whole countries blacked out” point* Judith Shklar, “The Liberalism of Fear” — the cruelty-first, anti-utopian liberalism Alan invokes against thick conceptions of the goodTimestamps(00:00) Intro — two papers, four hosts (01:47) Paper 1: Pope Leo XIV’s encyclical, Magnifica Humanitas (04:00) Subsidiarity, or “Catholic federalism” (12:26) Does the Pope take AI seriously enough? Mind-body dualism and the ex cathedra problem (15:34) The coming religious schism over AI personhood — and the Butlerian jihad (18:06) Transhumanism and the dignity of human limits (20:59) When is using AI a sin? Best-man speeches and eulogies (25:05) The universal destination of goods — is AI access already universal? (33:37) Is AI a centralizing technology? Dignity vs. efficiency (36:37) Freedom vs. control, the labor market, and make-work (41:10) Chess, the centaur era, and living after we’re no longer the best (47:34) Sponsor: Revelio Labs (48:49) Paper 2: DeepMind’s “Positive Alignment” (49:17) The liberalism of fear and thick vs. thin notions of the good (53:53) Is positive alignment an empirical question? A/B-testing flourishing (56:29) What would a useful positive-alignment paper actually do? (58:09) Constitutional AI as a site for public participation (1:00:47) The Moral Machine and trolley problems at scale (1:01:08) Does the “constitution” metaphor hold? Virtue ethics and self-binding (1:10:02) Running every Supreme Court case through the models (1:10:53) Lessons from Meta’s Oversight Board (1:15:09) Wrap-upJustified Posteriors is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber. You’re also invited to our Discord community at: https://discord.gg/2r3pExumQOur sponsorThis episode is brought to you by Revelio Labs, the leading provider of labor-economics data, available to academics on WRDS.Transcript:Seth (00:00:00): [upbeat music] Welcome to the Justified Posteriors podcast, the podcast that updates beliefs about the economics of AI and technology. I'm Seth Benzell, always positive and always aligned, coming to you from the Pocono Mountains of eastern Pennsylvania.Andrey (00:00:23): And I'm Andrey Fradkin, coming to you from San Francisco, California. We're sponsored by Revelio Labs, fine purveyors of data products. And we're very excited to have Alan Rozenshtein and Kevin Frazier from the "Scaling Laws" podcast on the podcast today. Welcome.Alan (00:00:43): Yeah, thanks for having us.Andrey (00:00:45): Just for our listeners, why don't you tell us a little bit about "Scaling Laws?"Kevin (00:00:51): Sure thing. So our main goal here is to provide robust and timely analysis of all AI policy questions. And that's an expansive ambit, and it's one that keeps us really, really busy because if it's not an executive order, then it's some big new policy idea from one of the labs, or it's some new economic report. But really what we try to do is dive into the weeds of policy and legal issues that are emerging in the AI space, given our backgrounds as law professors. But Alan's the one with the brain, so I'll let him fill in the details on earth.Alan (00:01:30): No, that's a perfect description. Yeah. We just think that there's a lot of really interesting stuff happening at the intersection of AI, law, policy, especially around national security, which is the core focus of the publication that "Scaling Laws" is part of, which is Lawfare, and so we're trying to fight the good fight, and it's never a dull moment.Kevin (00:01:47): That out of the way, I think we can dive into our first paper. Although, I think by any podcast standards, our first paper is lengthy to say the least, dealing with the-Andrey (00:02:00): MmKevin (00:02:01): ... pope's encyclical at 42,000 words, or for all those listening, about two and a half hours on my stationary bike. I don't know what that says about my biking skill-Andrey (00:02:13): [laughing]Kevin (00:02:14): ... or my reading ability, but it was a very tiring afternoon. But a very extensive, very important read from Pope Leo. And this has been covered by a lot of folks, but I don't think it's ever been covered by two lawyers and two economists at once.Andrey (00:02:34): [laughing]Kevin (00:02:36): My hunch is that this wasn't what Pope Leo was anticipatin… Get full access to Justified Posteriors at empiricrafting.substack.com/subscribe