We study how firms built around AI capabilities-“AI-native” firms-are organized. Drawing on Y Combinator batches W20-F24 and U.S. venture-backed startups whose first financing closed between 2020 and 2024, we classify each firm’s AI-native status and link it to workforce microdata on team size, function, seniority, and hierarchy. Relative to non-AI startups in the same industry-cohort, AI-native firms are 25% smaller. Their share of engineers is 13% greater, and the shares of entry-level workers and managers are each roughly 15% lower. Their hierarchies are half a seniority level flatter-yet valuations are comparable, implying more value created per employee. We argue these patterns reflect two channels: a process channel, in which AI changes how people work inside the firm, and a product channel, in which AI capabilities are built into what the firm sells. Using text from product descriptions and job postings, we find that embedding AI into the product, beyond layering on AI tools into existing workflows, is a primary way startups are scaling “knowledge work” without large teams of knowledge workers.

The tweet storm on the new paper is especially useful.  Via Luis Garicano.  And note those results predate the very latest and best tools.

The post AI-Native Firms appeared first on Marginal REVOLUTION.

      

Related Stories

• Do teens regret their social media use?
• Can Online Activity Be Regulated? Evidence from Adult Websites
• General-purpose large language models outperform specialized clinical AI tools on medical benchmarks

 

With over 100 million daily active users and projected revenue bookings of $7 billion this year, it is one of the largest gaming economies in the world—and one that has made millionaires out of teenage developers in Argentina, South Korea, and everywhere in between.

Tyler and Dave explore why Roblox decided early against prioritizing advertising revenue, why Dave thinks the main competition of Roblox is its own execution speed rather than Fortnite, whether every mega platform inevitably becomes an everything app, how falling token costs will change the platform, why he insists all the games on Roblox are beautiful, whether Robux should have a floating exchange rate, why admitting you have kids under 13 on your platform turns out to be a competitive advantage, why he’s skeptical of blanket social media bans, what his son’s experience with bipolar disorder taught him about metabolic health, his two-year sabbatical between companies that involved a motorhome trip across North America and a stint hosting talk radio in Santa Cruz, why Mutiny on the Bounty remains one of his favorite books, what he’ll learn next, and much more.

Excerpt:

COWEN: What percentage of your games now do you feel are beautiful?

BASZUCKI: All of them.

COWEN: Some look just quite ordinary. They might be fun, but I wouldn’t say they’re beautiful, right?

BASZUCKI: Well, I was trying to go a couple levels out of the box on you there. The reason I feel they’re beautiful is when you said that, I immediately went to look and feel, but then I tried to imagine the 12-year-old or the 18-year-old or the 30-year-old struggling to build something wonderful and the human connection to those games. By that definition, I think they’re all beautiful. They are all the efforts of creation of real people trying to pour their hearts out to make something that other people love to play.

On an artistic basis, I think you could ask me what percent of paintings in the MoMA do I think are beautiful. I’d probably say 20 percent. If I had to look at 1,000 Roblox games, I wouldn’t name which is more beautiful to me because I think that’s less important than really the heartfelt work of all the creators.

COWEN: I’ve been struck when I look at gaming at how much people don’t seem to care much about the visual beauty of their games. I would have expected something different, say, 15 years ago, and they just want a game that engages them somehow. Normal standards of visual beauty seem to have fallen away. Is that incorrect? Would you correct that impression in some manner?

BASZUCKI: I think you’re absolutely correct. What I feel you may actually be describing, if we looked into other disciplines, the evolution of story from the campfire to written to audio to a movie, and the increasing fidelity; all of those stories, in a way, are beautiful, but at the time, for the vast majority of the creators, it may be that writing is just easier than producing a 4K Hollywood movie. I feel that’s a little bit like the metaphor you’re talking about right now in gaming.

For the vast majority of people, their story or their idea for their game is actually pretty beautiful. Whether it’s a fashion game like Dress to Impress or it’s a grow garden game, the games are arguably beautiful, even if they don’t look photorealistic. What I think we’ll see is, over time, as AI helps accelerate the ability to make games look really polished in any style the creator wants—could be photorealistic, could be anime, could be a Warner Brothers 2D cartoon look—you and I might say that looks more beautiful, but the core gameplay is still somewhat the original gameplay. I think we are going to see games arguably look more beautiful, even though I think they’re all beautiful.

The dialogue is a bit slow to get underway, but there are many interesting parts.

The post My Conversation with Dave Baszucki appeared first on Marginal REVOLUTION.

       

Comments

• Video games are still delivering unmatched value on a ... by Jumfrey Tuckins, President of Texas
• In reply to Slam. Graphically amazing does not equal ... by chava
• In reply to Slam. Speaking of beauty — gamers know that ... by Slam
• In reply to MikeP. I guess it is appropriate that Ready ... by M
• In reply to MikeP. If by the godfather you are referring to ... by rayward
• Plus 5 more...

Related Stories

• Do teens regret their social media use?
• Can Online Activity Be Regulated? Evidence from Adult Websites
• AI nationalism, Europe included

 

Latin America’s momentous fertility transition is now in the domain of history, allowing a cohort perspective on the decline of completed fertility. Using census microdata from 17 Latin American countries, we track female birth cohorts from the 1920s to the 1970s by subnational region to document the extent to which cohort fertility decline coincided with other demographic and socioeconomic processes. Across cohorts within subnational regions, children ever born fell one-for-one with mortality decline. Expansions in urbanization, multigenerational living, women’s and husbands’ education, women’s employment, and the non-agricultural sector all predicted declines in ever-born and surviving fertility, but women’s education and sectoral composition were the dominant forces after covariate adjustment. Fertility decline was not systematically linked with improvements in children’s outcomes, including school enrollment, literacy, primary completion, and non-employment. These cohort facts challenge theories of fertility decline centered on women’s work and children’s education but support others emphasizing women’s education.

I fear that means the women think they are finding better and more fun things to do?  Which is hardly bad per se, but…

That is from a new NBER working paper by Regina Calles and Tom Vogl.

The post A Cohort Perspective on Latin America’s Fertility Transition appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Tom Meadowcroft. My Dad started his Civil ... by bazza
• In reply to rayward. In the ancient world, young women had ... by rayward
• In reply to Dismalist. Why not sell more bonds faster than ... by rsm
• In reply to Dismalist. That’s one half but I think the ... by KRS
• In reply to So…. I know. It’s really weird when the zoo ... by The Scary Black Hundreder
• Plus 10 more...

Related Stories

• Do teens regret their social media use?
• Who Leads? Relative Age Effects on Social Capital
• AI-Native Firms

 

2. Apply to Coase workshop on institutional analysis in Mexico City.

3. How does a mechanical watch work?

4. New immigration debate video from Caplan-Garett Jones.

5. Do stolen French fries taste better?

6. New minimum wage results.

7. Supply is elastic, even in the shuttered Straits of Hormuz.

8. “We registered the AI agent with the SEC as an investment advisor.”

9. Carlo Ginzburg, RIP.  And the NYT obituary.

The post Wednesday assorted links appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Sure. I think the Israelis sold Trump on either ... by M
• In reply to RAD. If I can recall, Trump said a something last ... by M
• In reply to Sure. The psychopaths in the IRGC murdered 10K to ... by mikeInThe716
• In reply to So much for libertarianism. It's almost like ... by Hazel Meade
• In reply to bazza. I mean it's true that in the ideal state, ... by Hazel Meade
• Plus 10 more...

 

Much of what looks like changing marriage preferences over the twentieth century is actually demographics. Exploiting plausibly exogenous variation in sex ratios across U.S. birth cohorts (1870, 1930, 1950), we jointly identify preferences, match quality dynamics, and the costs of marriage and divorce. Demographics alone explain two-thirds of cross-cohort differences. Women’s premium for older husbands collapsed across cohorts; men’s preferences barely changed. Love that survives its early years becomes permanent, but the odds of surviving fell from 97% to 44%. Divorce costs fell six-fold and depend on life stage. A horse race across behavioral channels shows that the match quality process—not mate-age preferences—is the primary dimension of generational change. Declining divorce costs and fragile match quality are substitutes: either alone fits the data, but together they reveal two independent dimensions of social change. The model validates out of sample on the 1910 and 1970 cohorts.

That is from a recent paper by Jose-Victor Rıos-Rull, Shannon Seitz, and Satoshi Tanaka.  Via the excellent Samir Varma.

The post Facts about American men and women appeared first on Marginal REVOLUTION.

       

Comments

• In reply to dmcharette. Your logic makes sense, but I have ... by ndp_jwh
• I assume Tyler is betting on a very high divorce cost ... by So...
• In reply to Sure. “4,000 swipes to find a serious match” ... by Dylan
• In reply to &. Agreed! The gifts have the unintended ... by dmcharette
• In reply to &. Even on a small scale… Before he died over ... by Phil S.
• Plus 10 more...

Related Stories

• A Cohort Perspective on Latin America’s Fertility Transition
• Do teens regret their social media use?
• Can Online Activity Be Regulated? Evidence from Adult Websites

 

A new study by Irish researcher Eoin Whelan attempts to answer this. Dr. Whelan told me he was specifically inspired by Haidt’s 2024 claims and sought to examine them rigorously and in the context of other regrets. This is a great use of science…testing dramatic public claims. So…do they hold up?

In Dr. Whelan’s study, 389 young adult participants (20-24) who were social media users as teens were asked about their regrets regarding their teenage years. A list of 20 possible teenage regrets was asked of all participants, with degree of regret marked on a 7-point Likert scale. This is an interesting design…testing social media regrets against other possible regrets, putting them in better context than the crude survey Haidt relied on.

So how did social media regrets hold up? Out of 20 possible regrets, too much time on social media ranked 13^th. The top regrets were 1.) not sticking up for oneself, 2.) being too self-conscious, 3.) not documenting memories, 4.) not learning practical life skills and 5.) not getting help with mental health. Girls were slightly more likely to regret time on social media than boys (ranking 11^th vs 13^th) though this effect was very small (I estimated it at about r = .11) so hardly the big “vulnerable girls” narrative some have peddled.

Further, regrets over time spent on social media as a teen did not predict current young adult life satisfaction for either boys or girls. Thus such regrets may be more a symptom of current panics over social media than anything of actual life importance2. Of the regrets, only not working harder in school and not exercising negatively predicted young adult life satisfaction. Interestingly, having regrets over socializing with friends positively predicted life satisfaction.

As Dr. Whelan noted in his study, “The objective of this study was to critically examine the commonly held belief that social media use during teenage years is a significant source of regret and a predictor of diminished well-being in early adulthood…Contrary to dominant narratives in the public domain, our results suggest that regrets over time spent on social media are not among the most potent regrets reported by young adults…As such, these results align with prior research indicating that the harmful effects of social media may be overstated.”

Here is the full Chris Ferguson Substack.

The post Do teens regret their social media use? appeared first on Marginal REVOLUTION.

       

Comments

• Standing up for oneself. … Hmmm…. I'm not sure what to make ... by Neurotic
• In reply to steve. “… the worst are full of passionate ... by Earweigh Phenup
• In reply to Wayfaring Stranger. Except Haidt's claims/studies ... by steve
• In reply to rayward. Remote work is dangerous to one's mental ... by rayward
• In reply to JFA. TV and Social media are the only ones ... by David C
• Plus 10 more...

Related Stories

• My Conversation with Dave Baszucki
• A Cohort Perspective on Latin America’s Fertility Transition
• Can Online Activity Be Regulated? Evidence from Adult Websites

 

The consequences of online regulations depend on the extent to which users can circumvent restrictions or substitute toward noncompliant platforms. Since 2023, 25 U.S. states have implemented age verification laws that caused prominent adult websites (including Pornhub) to restrict local access for all users. We study how these restrictions affected browsing activity using individual-level panel data. Access restrictions reduced overall time spent on adult sites by roughly 10%. Specifically, for every 100 hours spent on top adult sites before restrictions, about 50 hours remained accessible at noncompliant sites that never restricted access, 30 hours persisted through VPN-based circumvention, 10 hours were substituted from compliant sites to noncompliant sites, and 10 hours were no longer spent on adult sites.

That is from a new NBER working paper by Matthew Brown, Emily J. Davis, and Devin G. Pope.

The post Can Online Activity Be Regulated? Evidence from Adult Websites appeared first on Marginal REVOLUTION.

       

Comments

• In reply to b. Where have I shown a blindness to violence? I ... by SpeculativeDiction
• In reply to An Aussie. You referred to Elon, he has faced ... by b
• In reply to b. And the arrests for social media posts? Where ... by An Aussie
• In reply to rsm. No. Median voter is something like 47 and ... by Sure
• In reply to Sure. Is the median voter having fewer and fewer ... by rsm
• Plus 10 more...

Related Stories

• AI-Native Firms
• My Conversation with Dave Baszucki
• Do teens regret their social media use?

 

2. A treasury of book dedications.

3. Abdullah Ibrahim, RIP (NYT).

4. One part of the Iran equilibrium.

5. Economists report on Fable 5.

6. Another estimate of the productivity gains from AI.

7. Until the 1970s, Algeria was the world’s largest wine exporter.

8. Mosquito drone for surveillance?

The post Tuesday assorted links appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Action Figure. Early on, there was some talk ... by Gilligan
• In reply to M. Iran would only need a handful of medium-range ... by B.B.
• In reply to Hazel Meade. Supposed to be sub-thread reply to ... by Hazel Meade
• In reply to Guy. Reading these chilling status reports, you ... by Action Figure
• I'm arguing, if you're going to do it, do it right, and if ... by Hazel Meade
• Plus 10 more...

 

What’s remarkable is how China is winning: deregulation and capitalism. It’s faster and easier to set up a clinical trial in China than in the United States. China is even experimenting with the peer approval model I’ve long advocated. The Medical Tourism Pilot Zone on Hainan island lets medical institutions import and use any pharmaceutical or device approved in the EU, US, or Japan — no separate Chinese approval needed. China is using our own regulatory judgments to get treatments to its patients faster than we do.

The core problem is that our clinical trial and drug approval system is slow and expensive. Getting a new drug to market in the US takes billions of dollars and a decade or more of clinical trials — and all of that before a company earns a single dollar. The consequence is drug lag and drug loss and also learning loss. Innovation is a dynamic process. You must build to build better.

It’s not over for the United States, however. Montana’s SB535, signed into law in May 2025, is the most important regulatory innovation in drug approval in my lifetime. The law authorizes investigational drugs and therapies that have cleared Phase I trials to be prescribed and sold — bypassing the traditional FDA approval pathway. It makes Montana the first state to license experimental treatment centers, “one stop shops” for otherwise hard-to-access care.

This is a very big deal.

SB535 makes Montana the only state in the nation where firms can move more quickly from a successful Phase I trial into limited commercialization. This positions Montana as a highly attractive location for biopharma, biotherapeutics, and other life sciences companies that want to accelerate time-to-market while continuing the federal FDA approval process.

Montana’s regulatory system creates the possibility of a self-funding clinical pipeline: companies using early commercial revenues to finance the path to full FDA approval. You get treatments to patients faster, and you keep companies alive long enough to prove their treatments work. Experimental treatments are not for everyone–these treatments are cash based–no Medicaid or Medicare and probably no private insurance either–but after conventional treatments have failed experimental treatments should be available for some patients, both for their benefit and for ours.

Montana is not alone. Florida now allows non-FDA approved stem cell therapies:

A new law in Florida, CS/CS/SB 1768, allows physicians to market and administer stem cell therapies that have not been approved by the U.S. Food and Drug Administration (FDA) for orthopedic conditions, wound care and pain management.

These experiments in regulatory federalism are vital and not just for patients but also for geopolitical competition. I am thrilled China is pursuing medical innovation (I predicted and applauded this in my TED talk) but I also don’t want to see America falling behind.

The Trump administration has been supportive. I would like to see HHS and the FDA working with companies operating under state right-to-try frameworks — sharing data, clarifying federal-state boundaries favorably, and treating these experiments as the biotech competitiveness infrastructure they are.

The FDA approval process has long been treated as the only legitimate path to market. The cost of that orthodoxy is measured in companies that never reached viability, innovations that never got off the ground, and patients who died when they didn’t have to. I have spent thirty years trying to get people to see the invisible graveyard. That’s hard. Most remain blind. But China’s bursting pipeline of new drugs is visible — could this be a Sputnik moment for biotech?

An American biotech renaissance — driven by AI, federalism, and regulatory innovation — is possible. The path forward is to double down on what makes America great: the laboratories of democracy are working, and in Montana and Florida, so are the labs.

The post Montana’s SB535 and a Potential Biotech Renaissance in America appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Brian Villanueva. What about making sick people ... by qetzal
• In reply to Can't Have Any Pudding. If the patients have not ... by superdestroyer
• Federalism. It gets things done. by Brian Villanueva
• In reply to Dan. Efficacy should be a stamp of approval by ... by Brian Villanueva
• In reply to RAD. Doubtless just a temporarily embarrassing ... by Anti-Gnostic
• Plus 10 more...

Related Stories

• The Nationalization of American Science
• The Cultural War is a Civil War
• Here Comes the Sun(screen)

 

Justin Helman didn’t get his dream acceptance from the University of Florida. But that isn’t stopping him from pursuing the classic college experience there.

The recent high-school graduate from Park Ridge, N.J., is set to move into a private apartment right by campus. He is enrolling in a UF online program for the first few semesters and paying an extra fee package to access services like the campus gym and student-section football-game tickets. He plans to study at the library, join clubs and might rush a fraternity.

The post Educational arbitrage? appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Gilligan. This is true, but those schools that ... by Slocum
• In reply to Wayfaring Stranger. The margin is about the same. ... by anonprof
• In reply to Wayfaring Stranger. @Gilligan: Where I ultimately ... by Wayfaring Stranger
• In reply to Cicero. Last I heard on-line students also pay ... by Wayfaring Stranger
• In reply to Liface. For sure. In my times we had guys that ... by Bluto
• Plus 10 more...

Related Stories

• My Conversation with Dave Baszucki
• A Cohort Perspective on Latin America’s Fertility Transition
• Do teens regret their social media use?

 

There is yet another huge problem behind all these first-order problems. Let us say, for instance, that France’s Mistral AI develops very nicely and serves as an EU counterpart of Anthropic and OpenAI. Well, then the other European countries will become highly dependent on the French. That may seem okay today, but it will be much less fun for the Germans if the French really do have all that extra power and leverage.

As for the French themselves, they would be highly dependent on a private company. France may end up with one such company, but it is unlikely to have three of them. So Mistral will in turn have high leverage over France, French politics, and French foreign policy. Let us hope they are up to that. The simple point is that being influenced by someone in your home country, even if it sounds more appealing rhetorically, is not always better than being pushed around by foreigners. Sometimes the foreigners are less oppressive and intrusive, if only because they care less about you.

Worth a ponder.  I am hearing good things about the new Mistral model, so these questions may become relevant sooner than I had thought when writing this.

The post AI nationalism, Europe included appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Kaz Stanczak. For all equivalent other utility ... by M
• Given that France is a nuclear power, doesn't it already have ... by Nik
• In reply to M. “Tyler might think (and maybe I’m only 50% ... by Kaz Stanczak
• In reply to Kaleberg. I don't know if this will happen, but ... by M
• In reply to Wayfaring Stranger. Thank you for your comments. ... by Kaz Stanczak
• Plus 10 more...

Related Stories

• Sometimes it is hard to solve for the equilibrium
• Can Online Activity Be Regulated? Evidence from Adult Websites
• Montana’s SB535 and a Potential Biotech Renaissance in America

 

Peter Thiel, Ross Douthat, telephone!

And yet I have plenty of quibbles.  It was a little too long.  The acting is entirely serviceable, but none of the characters are excellent or memorable.  The portraits of America are below the level of charm and insight we have come to expect from Spielberg.  And any time a character makes “a speech” it is pretty mediocre.

Cinematic influences are numerous, starting with E.T. and Close Encounters of course.  Not to mention Sugarland Express.  I was surprised to see the references to The Magic Flute, including the Bergman cinematic version.  Perhaps Spielberg had Jacob’s Ladder in mind as well?

The Freudian interpretation of the film I will not articulate, but it surfaces near the end and never quite goes away.

But who here was the Antichrist anyway?  That is up for grabs.

I had no problem sitting through the movie and enjoying it, but the problem of excess hodgepodge worsens as the exposition continues.  So I will grade this one as misunderstood, but nonetheless no better than an interesting failure.

The post *Disclosure Day* (doubt if there are net spoilers in this post) appeared first on Marginal REVOLUTION.

       

Comments

• I am surprised that you say anything positive about this movie ... by Fahad Hasin
• In reply to Pedro. I love science fiction, but I am not drawn ... by It Ain't Necessarily So
• It was a horrible movie. Some of the action scenes were simply ... by Pedro
• I watched it in theatres, I was surprised at how similar it was ... by Nicolas Krause
• Didn't much like it. Script was bad. The “will Christians ... by Ted
• Plus 10 more...

Related Stories

• *The Pressure* (no spoilers)
• My excellent Conversation with Katja Hoyer
• Hayekian Literary Criticism

 

Champions of a European AI model should ask themselves if a European effort would be more effective than Meta, which this year will spend more on chips ($125 billion) than Germany spends on defense ($114 billion) and offer salaries of over $100 million to attract the best researchers, and is still failing to catch up.

Here is more from Pieter Garicano and Simon Grimm.  Via Jesper.

The post Germany fact of the day sentence to ponder appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Gilligan. They haven't invested into necessary ... by Marian Kechlibar
• In reply to Old Programmer. IBM had this problem in the 1960s ... by Kaleberg
• In reply to Okay. Gee, and I thought that terrorists ... by Kaleberg
• In reply to M. The Europeans also don't like being locked ... by Kaleberg
• In reply to Aaronn. It sounds like one could write an app for ... by Kaleberg
• Plus 10 more...

 

2. Jason Furman on Social Security (NYT).

3. Markets in everything, customized water edition.

4. AI progress in Rio de Janeiro.

5. Satya Nadella does Oliver Williamson.

6. A shared feed of my guest appearances.

7. Results on the Great Recession and U.S. fertility.

8. Nice words about me.

The post Monday assorted links appeared first on Marginal REVOLUTION.

       

Comments

• In reply to M. More like aquaoir, I think. by mkt42
• In reply to ads. you just can't hate boomers enough. ... by Mark Bahner
• In reply to Bill Allen. So, no, the trust fund isn't a fraud. ... by Mark Bahner
• In reply to Bill Benzon. Whoops. NWP=numerical weather ... by Bill Benzon
• In reply to Dismalist. So, I did a little digging with ... by Bill Benzon
• Plus 10 more...

 

In the most recent census (2022), 1,017,437 people in Ireland were born abroad. Even if you classify people from Taiwan as “foreigners”, there are 845,697 + 157,886 = 1,003,583 immigrants to China. There are now more foreigners in Ireland than in China in absolute terms, despite having a population that is 260 times smaller.

The post Republic of Ireland (China) fact of the day appeared first on Marginal REVOLUTION.

       

Comments

• In reply to Anti-Gnostic. I don't think this sentence is ... by M
• In reply to John Mansfield. “Born here” in this article ... by John Mansfield
• In reply to Peter. How many non-British/non-EU immigrants, or ... by Anti-Gnostic
• When you talk about immigration to Ireland, it is important to ... by Marty Murphy
• In reply to Hoosier. The culture of the USA certainly changed ... by Florida Man
• Plus 10 more...