Here’s what caught our attention over the last week:

1. Legalizing manufactured housing requires legalizing starter homes — Alex Armlovich

2. The cascade of rigidity — Nisha Austin

3. Willingness to pay for faster clinical trial approval — Matt Clancy

4. Should more of us be theoretical biologists? — Saloni Dattani

Willow is on holiday this week.

Legalizing manufactured housing requires legalizing starter homes — Alex Armlovich

Emily Hamilton at Mercatus has a new policy brief on the key zoning reforms needed to make HUD Code building code reform work: if a town has zoned out starter homes, it has zoned out manufactured housing. Twelve states have passed laws since 2020 to make HUD Code homes easier to site, and the cost advantage is real: a single-section home runs 65% cheaper than comparable site-built housing, a double-section 40%, and a CrossMod 27%. But in the three states with the strongest equal-treatment laws, manufactured housing's share of new construction showed no visible jump afterward, in substantial part because nondiscrimination against HUD Code starter homes is perfectly compatible with bans on all starter homes. Pairing National Zoning Atlas data with FHFA land values, Hamilton finds that land cheap enough to support a single-section home ($47,000 for the lot under a $94,000 house) covers 0.77% of zoned land in Middle Tennessee, 0.89% in Hampton Roads, and 0.02% in Northern Virginia under current minimum lot sizes. Cut minimums to 1/16 of an acre and those shares rise to 12.8%, 5.3%, and 1.8%. The ROAD Act's chassis fix widens the HUD Code cost advantage…but only if those homes aren't also bundled by zoning law with an unaffordably large parcel of land.

The cascade of rigidity — Nisha Austin

I was excited to see America.gov launch yesterday, and happened to be reading How to peel a banana by Marina Nitze this week. There’s a nice parallel between being able to access the right information and the information itself being right.

Nitze writes about the “cascade of rigidity” which she defines as what happens when “each layer of government interprets the layer above it a little more stringently than the one before… rules only ever get added and calcified, never taken out or relaxed”. She uses the foster care licensing standards as her primary case study and talks about how HHS published model foster care licensing standards in 2018 with the explanatory context stripped out, states copy-pasted them, and created situations where care providers were disqualified under rules that didn’t make sense (e.g. a grandmother on tribal land being disqualified because she didn’t meet a recycling requirement but the land she lived on did not have recycling service). Nitze and her team figured out how to work with the system: they worked with HHS to make separate standards for kin legally permissible, and then published better model standards themselves, letting the same copy-paste machinery carry them; 39 states and 6 tribes had adopted them so far.

Willingness to pay for faster clinical trial approval — Matt Clancy

One important reason why accelerating clinical trials is a good idea is that we want people to get access to new medicines sooner. But another reason is that there is a robust literature that shows pharma companies do more R&D when it is more profitable to do so. Shorter clinical trials mean firms can make money off their drugs for a longer time before their patents expire, and all else equal, this should induce more research that leads to new drugs.

Just how much does faster approval increase profits? Priority Review vouchers provide one line of evidence on this question. Priority review vouchers allow the holder to get an expedited review of a drug, shortening approval time by about four months; they’re issued to induce organizations to develop certain qualifying drugs. Drug sponsors can sell these vouchers to each other, and the price a pharma company is willing to pay for a voucher gives a window into how valuable a four-month speedup in drug approval is. Writing for Nature Reviews: Drug Discovery, a recent short article by Ridley and Xu find the typical price firms are willing to pay for a review voucher was about $100mn between 2019 and 2024, but has climbed to about $200mn in recent years. For the drugs that obtained vouchers, they also look at sales data and attempt to model the value this speed up provided to the holder. The average value, in terms of increased sales from faster market entry, is a little over $500mn. But the returns are highly skewed: three drugs probably made more than $1bn in additional sales from accelerated approval, but several others earned less than $100mn. Even so, they estimate more than half of drugs with a voucher earned an additional $200mn from the voucher, double what the going price for a voucher over the last few years.

Should more of us be theoretical biologists? — Saloni Dattani

‘Biology is more theoretical than physics’ was the provocative title of an article several years ago that laid out some examples of biological theories that preceded or predicted the discovery of important entities in advance, including genes, receptors, enzymes, tumor suppressors. Entire processes in biology were also theorized before their discovery, including how neurons function and carry out computations; how the inheritance of genes is sometimes linked to that of others; and ‘clonal selection’, the process by which immune cells multiply after matching with antigens to respond to an infection.

In a new piece for Asimov Press, Ulkar Aghayeva writes about why theoretical biology hasn’t had as much recognition, whether it’s harder for theories to be developed and tested due to the complexity of biology, and about some of the impacts that this theoretical work has had, including inspiring entirely new fields of experimental work. It seems to me that a huge amount of biology traces back to ideas that had previously only been theoretical, extrapolated from a couple of observations here and there, before being tested, refined or overthrown. And sometimes, even theories that were once cast aside have made a comeback decades later, like innate immunity, which was basically forgotten for about eighty years. Perhaps more of us should be theoretical biologists.

Here are a couple other highlights from our team:

• Matt appeared on Macroscience to discuss science policy and the returns to R&D investment.

• Saloni published an interactive clinical trial bottlenecks game, letting readers walk through the tradeoffs and constraints that slow down drug development.

Here’s what caught our attention over the last week:

1. This week at Climate Week — Alex Armlovich

2. OMB's intrusion into NIH — Matt Clancy

3. The great American property tax freak out — Dylan Matthews

4. Your flying car will still have AM radio — Dylan Matthews

5. Hiring impossible: FDA fallout — Saloni Dattani

6. The state of think tanks — Nisha Austin

Willow and Alex are at Climate Week, and Jordan is on parental leave and will be away from the blog for the next few months. Congratulations to Jordan and his family!

This week at Climate Week — Alex Armlovich

I'm writing this from Climate Week, where Coefficient Giving is a sponsor of “The Urban Climate Solution” series, ten free sessions with the Parking Reform Network, Pew, and WRI's Ross Center, making the case that urban policy is the most neglected form of climate policy for climate funders. Wednesday's track is of particular interest to housers: The opening panel's framing device is the 2007 McKinsey global cost curve for climate abatement, which ranked all kinds of decarb interventions…and never included land use at all. But infill housing cuts VMT and carbon while running on infrastructure that already exists, and the policies that legalize it (ending exclusionary zoning, permitting infill by-right; ending parking mandates) cost governments nothing in fiscal terms, and generate large social benefits by breaking the tradeoff between economic growth and emissions reduction. The spiciest session might be "Build More, Electrify Faster," which takes seriously where housing abundance and building decarbonization could conflict if tradeoffs aren’t managed carefully. Transportation sessions run Monday, Tuesday, and Thursday, including a congestion pricing retrospective now that even New York’s heavily-compromised launch edition of congestion pricing has strong numbers: 11% less congestion, 22% cleaner air, 7% more transit ridership.

OMB’s intrusion into NIH — Matt Clancy

When an agency proposes a significant regulatory action, the Office of Information and Regulatory Affairs (OIRA) has the power to first evaluate it to ensure alignment on policy and the minimization of regulatory conflicts. This review step slows down policymaking, but that might be worth it if the review helps avoid costly regulatory errors. Given that the review is intended to be used for significant regulatory action, it is no surprise that it has not often been used at the National Institutes of Health, which primarily fund research rather than regulate.

But as Christopher Marcum notes in a guest post for the Good Science Project, something has changed in 2026. We’re not even all the way through 2026, but 27% of NIH documents ever reviewed by OIRA were documents from 2026. Even more puzzling, since this specific OIRA review power was established in 1993 (under Executive Order 12866), it has never before bothered to review NIH requests for information; now it is reviewing all of them. The upshot is a slowdown in the ability of the NIH to make policy changes.

NIH seems to be the subject of special attention here - there has been no uptick in review of documents at NSF, the other large federal science grantmaker. Meanwhile, there are other signs of conflict between NIH and OMB (where OIRA is housed). The Washington Post reports that NIH director Jay Bhattacharya refused a recent request from the president to cut funding to Harvard, and in response OMB Director Russell Vought proposed the creation of a board that would oversee NIH grantmaking. According to reporting from Politico, the proposed board would include Vought and need to reach unanimous agreement for all grants, essentially giving OMB a veto over NIH grant decisions. As we’ve written about before (see Jordan’s sections of this post and this post), this follows OMB’s other attempts to expand political control over the scientific grantmaking process. So far those efforts have been put on pause.

It’s worth noting that despite the interference, NIH in 2026 is close to being on track to spend the funds that had been appropriated to it according to Jordan Dworkin’s science spending tracker.

The great American property tax freak out — Dylan Matthews

Building state capacity in the US is hard, but a good starting principle is to not do any harm. Alas, US states appear committed to doing harm.

In a nice new paper, Yale law professor David Schleicher tracks the ongoing post-Covid property tax revolt, which threatens to starve states and localities of desperately needed revenue in coming years. Last year, Texas adopted $51 billion worth of property tax cuts; Florida legislators put on the ballot a cut that would exempt most homeowners from paying any non-school property taxes. Several other states have adopted large credits and exemptions meant to exclude large shares of homeowners from paying property tax.

Property tax is the main way state and local governments pay for themselves in the US. They generate up to 30 percent of combined state-local revenue, and 70 percent of local revenue specifically. As Schleicher and Nick Bagley have argued elsewhere, the state capacity problem in the US is disproportionately state/local in nature. State and local governments employ the vast majority of government employees, are responsible for most infrastructure for services like transportation, water, sewage, and electricity, and handle permitting for most private construction projects.

The failures of capacity at the state/local level are not solely matters of insufficient funding. Indeed, on infrastructure the problem is more that most projects are too expensive on a per-unit basis. But there are key areas of underfunding, and the property tax revolt is likely to make those problems all the worse.

Your flying car will still have AM radio — Dylan Matthews

A lot of abundance and growth discourse can be summed up as, “some people really don’t like change.” Case in point: the US House of Representatives has passed a measure that requires all new cars to include an AM radio, and the Senate is expected to follow suit. Now, you may be asking, “who the hell still listens to AM radio?” I remember being five or six and asking my dad about the “AM” option in our car, and him saying that AM radio was a bit older and lower-quality than FM. I am thirty-six years old, and AM radio was old-fashioned when I was a child.

AM is actually worse than old-fashioned, though; AM receivers can interfere with the electrical currents in electric vehicles, and EV manufacturers have argued that it will cost them billions of dollars to develop technology to overcome this problem and keep AM radios in cars.

But the National Association of Broadcasters, representing radio stations, is still pretty powerful, and despite the vast majority of Americans getting emergency notifications these days on their phones, the broadcasters have successfully pushed the idea that AM radio is an essential backup if the phones, FM radios, TV broadcasts, terrestrial internet, satellite radio, and all other alert mechanisms all fail.

The AM Radio for Every Vehicle Act would not be the most egregious wasteful act of overregulation the US has adopted, and it might not even be the worst this year. But it’s a great example of how short-term political considerations can lock in bad, costly policy for a long time. We could be zooming around in our self-piloted flying cars in the year 2100, and those flying cars will still, by force of law, have to have AM radios.

Hiring impossible: FDA fallout — Saloni Dattani

This week I’ve been reading about the FDA’s hiring struggles. For context, last year, as a result of cuts led by DOGE, the FDA laid off nearly 1,900 staff and another 1,200 staff took early retirement packages – combined, that’s roughly one-sixth of its workforce. The lay-offs seemed to have reversed last October, when the agency started rehiring. That rehiring process, according to STAT, is still ongoing and is bottlenecked, partly because the FDA’s HR teams also lost staff and because DOGE consolidated multiple administrative teams into one office that's understaffed.

It’s worrying because greater staffing levels at the FDA allow drugs to be reviewed more quickly, as well as inspections of facilities, investigations of safety problems, and allows staff to work more proactively on new initiatives, while shortages force them to focus on essential tasks. There’s also been pushback over whether the cuts could jeopardize the FDA’s targets under PDUFA, under which it's committed to review applications within set targets.

The state of think tanks — Nisha Austin

On Think Tanks published its State of the Sector report last week, surveying 299 think tanks across 94 countries. Two thirds of the respondents named funding as a challenge this year, and almost half expect the environment to get worse, which is surprising to read when many are expecting a huge surge of philanthropic money.

The report describes this as being a structural issue, which makes sense to me. 40.5% of these organizations hold less than three months of operating reserves, and 22.1% have none. This is mainly because they run on short-term project grants that don’t cover overhead (only 18% are predominantly core-funded, and over a third work on contracts of a year or less). It tracks that organizations with no reserves overwhelmingly expect funding trouble, while those with a year’s cushion mostly don’t.

I also found it surprising that pessimism is worst in the wealthiest regions. The Anglosphere has the deepest reserves and the most core funding in the survey, and also the largest share of organizations expecting mergers or closures. I’m still figuring out what to make of that!

Here are a few other highlights from our team and grantees, and announcements worth sharing:

• California YIMBY was named a Fast Company 2026 World-Changing Ideas winner.

• The Institute for Replication launched a newsletter tracking their work on verifying key findings across the social sciences.

• Willow published What to expect from energy abundance on the AGF blog earlier this week.

• The OpenAI Foundation announced funding for CTD Commons, a project to preserve and publish regulatory data from failed drug development programs so future teams can learn from them.

There’s a strange tension in the conversation on energy. On one hand, converts of energy “superabundance” claim that energy is economically critical, worth making so readily available that it’s practically too cheap to meter. On the other, textbook growth accounting holds that since energy is a fairly small share of the economy in developed countries1 (about 5 - 7% of US GDP in recent decades, depending on how you measure it), increasing energy supply will have a fairly moderate effect on GDP. How to resolve this tension, and more pointedly, how to value energy’s contribution to US growth?

Fortunately there’s at least a partial resolution to this dichotomy, one that respects both the intuition that energy can matter a lot and the evidence that it’s not a large share of most developed economies. Put simply, additional energy is very valuable when it’s scarce, since it becomes the binding constraint on all growth. When it’s already broadly available (although not necessarily too cheap to meter), the returns to additional supply drop back towards the unspectacular figures that stingier neoclassical models would suggest.2 Importantly, since energy demand in developed countries is on the rise, we may be entering a period when energy is scarce relative to our ability to make use of it, and therefore highly valuable for growth.

There’s another sub-argument that even scarcity is only a temporary phenomenon. Since energy is very difficult to substitute over the short term but gets a lot easier to replace over time as firms adapt, a sort of “keep calm and wait for efficiency” perspective argues that even shortfalls can and will be mitigated by improved technology. However, this doesn’t account for how economically costly that adaptation could be, potentially seriously understating the value of avoiding scarcity in the first place.

A clearer best guess at how energy drives growth is a road map of what policies and interventions are likely to work best (should we prioritize short-term supply bumps, or long-term structural change?) Since policymaking is an exercise in tradeoffs, it’s important to have an educated guess here, even as we look to new evidence to inform our understanding. The following discussion is a starting point for how we’re thinking about the energy-growth relationship at the AGF.

Why energy is missing from mainstream growth models

In an affront to the superabundance theory, most of the classic economic growth models (Solow 1956; Romer 1990; Aghion and Howitt 1992) don’t even include energy. Okay, this is a bit of an oversimplification – technically, energy purchases are netted out as an intermediate input – but broadly, we can’t independently observe or manipulate the role of energy using these models. And while energy and GDP clearly move together, efforts to prove a causal relationship have been largely inconclusive (Stern 2019).

The 1970s oil shocks, which showed that energy could behave in unexpected ways with outsized impacts even in developed countries, began to change this – somewhat. There was a rough split in the treatment of energy: while mainstream growth theory stuck with improvements to the classic, energy-less workhorses, some did start to add in an explicit energy component.3 Among economists that added energy, there was another schism, with some models starting to explore the idea that energy and capital could be complements, not substitutes (Berndt & Wood 1975), and others arguing that energy’s small cost share, and the possibility of capital / energy substitution over time, would, for the most part, keep its effects moderate (Hogan & Manne 1977).

Since energy typically isn’t a very large portion of GDP in developed countries – spending on electricity, for example, is only around 2% of US GDP – using this share to extrapolate a GDP effect from changes in energy supply yields a modest assessment of energy’s economic impact. However, this simplification remained a widespread best guess of energy’s role in developed economies as supply and demand moved mostly in sync (for instance, this is roughly the shorthand we at CG were using ourselves in late 2025).

It’s not hard to see why that’s an incomplete answer, though. If electricity supply drops dramatically, production doesn’t drop by 2% – since you can’t (quickly) substitute energy, you can’t produce much of anything. This is intuitive enough that it’s its own metaphor, but to belabor the point, energy is like water in Adam Smith’s classic paradox – while you might not spend much on it day to day, its absence would be distinctly noticeable. This relationship doesn’t just apply to a loss of supply that’s already online (and blackout losses are unique in several ways), but also to new potential growth on a fixed energy budget. Once you account for the possibility of a shortage, a fixed cost share shorthand is clearly missing something.

The complementarity literature continued to develop exactly this missing piece. While this literature is extensive, I find that Stern & Kander (2012) came up with a particularly convincing, lean articulation of the energy-growth relationship. Their modified version of the classic Solow production function frames energy and capital/labor as separate inputs to growth that can’t be easily substituted for one another. As a result, as energy becomes scarcer, it becomes an increasingly binding constraint on all growth.

Stern & Kander’s model formalizes the intuition that the energy-growth relationship is non-linear: energy’s effect on growth could be very different in periods of energy scarcity versus periods of abundance. If energy is a bottleneck, relieving the constraint is very valuable; if it’s not, returns will be much more muted. This resolves the contradiction between the apparently central role of energy in certain periods and geographies but not others, or the economic havoc apparently wreaked by shortfalls. It also explains why it so often goes missing in classic growth economics: in times of “normal,” unconstrained supply, since energy is a small and stable fraction of firms’ spending, valuing it at that share (or even simplifying it away) doesn’t cause too much of a problem.

It’s probably fine to meter your energy

So far, we’ve mostly argued that energy matters more than a naive application of mainstream growth economics would suggest, since it acts as a bottleneck during times of scarcity. But that’s a two-sided coin, with implications for the limits of energy abundance as well. Stern & Kander’s model implies that although energy is a limiting factor on growth, sufficiency is enough – once demand is met, additional energy supply will have strongly diminishing returns. The growth function converges back towards the energy-free version (with energy valued at its relatively modest cost share) once energy stops being scarce.

“Not scarce” in this case doesn’t mean free, or even radically cheap: under this binding constraint model, energy scarcity is a spectrum, defined in relation to the effective labor and capital available to employ it. Effective energy can become the binding constraint either due to supply issues (e.g. the 1970s oil shocks or the current Hormuz situation), or if the labor/capital deployed in energy-using industries increases faster than supply can respond (e.g., due to AI and electrification). However, once the bottleneck resolves and energy is again a small share of GDP, available labor and capital again become the binding constraints.

To be clear, this model doesn’t disprove the “too cheap to meter” thesis – it’s just a different conceptualization of how energy and labor/capital inputs relate. The superabundance case holds that if energy prices drop far enough, we’d see an substantially changed demand regime, with cheaper energy supply inducing a large crowd-in of capital and labor to energy-hungry industries. The binding constraint case is much more restrained on the marginal contribution of energy once energy is already abundant.

Beyond Stern & Kander’s own empirical evidence4, we have some real-world indication that a binding constraint model is a better approximation for how energy really works (although this is such a tricky question precisely because it’s hard to test given the data we have). First, alternative ecological economics models (e.g. Ayres & Warr 2005), which envisioned a more central role for energy in driving growth, fit the data noticeably less well after 1970 (Stern 2019), suggesting that a simple “[useful] energy drives growth” model doesn’t quite work.

Additionally, more recent real-world examples of positive shocks to energy supply don’t seem to bear out the argument of a transformative boom. While it’s admittedly hard to pull out a single effect at the GDP scale, we didn’t see a step-change in productivity as a result of the shale revolution. This doesn’t mean that the shale revolution didn’t have positive growth outcomes, including some industrial effects (and on fronts like inflation mitigation, which I don’t investigate here) – just that the clearest test of whether cheaper energy would kick-start a boom in industry or innovation, absent the latent capital/labor demand to effectively employ it, yielded a relatively modest result.

Even the role of energy in the industrial revolution, probably the clearest argument for an energy transition driving economic transformation, is debated. Although Allen (2009) and Stern & Kander themselves actually do argue for (in S&K’s case, scarcity-relieving) coal as a trigger of modern growth, others (Clark & Jacks 2007; Madsen et al. 2010) are much more skeptical of the effect of coal on British productivity.

However, to (at last) address the obvious and critically important point — there’s a good chance that we are staring down the initial stages of an energy shortfall, in this case driven by a combination of growing demand and artificially constrained supply. We therefore likely are entering a period where energy will act a bottleneck, and where the economic returns to unblocking it could be very significant. While it’s important to try to accurately forecast how long the shortfall will last if we expect returns to energy investments to drop sharply once it’s over, it’s equally important to be clear about the short-term costs – and the long-term implications – of entering a scarcity regime in the first place.

The catch: adaptation over the long term

Now that we’ve at least somewhat reconciled energy’s ambiguous relationship with growth, we can return to the more nuanced critique of an energy-forward strategy: namely, that we shouldn’t worry too much about an energy shortage over the longer term because the economy will adapt to scarcity, either by improving efficiency or through other means.

This short vs. long-term distinction was somewhat glossed over in the older complementarity literature, which assumed constant, exogenous substitutability between energy and labor / capital. However, later efforts found that effective substitutability could actually change over time in response to energy scarcity – even if energy is hard to substitute over the short term, it becomes increasingly (although not fully) substitutable over the long term as firms find more energy-efficient ways to produce goods and services. As a result, scarcity-driven prices themselves cause the energy constraint to become less binding over time, blunting the shortfall cost that a naive fixed-technology model would predict.

At first glance, this sounds great! Technological improvements make it easier for us to use less energy relatively painlessly, firms shave their energy usage to bring their energy cost share back down, and the economy basically stabilizes, despite the energy constraint. This yields a neat mental model of efficiency as a way to get most of the outputs associated with more energy, without actually needing more energy.

Unfortunately, this is only half of the story. The second half is that firm-level adaptation, whether it’s increasing efficiency or building out alternative generation, is costly – if firms could just use more cheap energy, they’d do that instead. This is because adaptation doesn’t spring into being from nowhere; rather, firms have to re-direct brainpower and budgets to make those scarcity-adaptive improvements. Assuming R&D budgets are relatively fixed,5 a redirection towards efficiency crowds out other productivity-enhancing technology. The opportunity cost of that crowding out – in addition to the growth directly foregone before adaptation gets up and running – should be added to the tab for an energy bottleneck, not ignored.

Adaptation can have lasting costs even if a bottleneck turns out to be temporary. First, once the constraint lifts and energy is no longer scarce, the knowledge of how to produce more with less is much less useful. Second, because knowledge compounds as future research builds on existing stocks (Acemoglu 2002; Aghion et al. 2016), the long run effects of misdirected R&D can compound – each year of foregone general R&D also leads to a weaker foundation for future general innovations (relative to the non-scarcity counterfactual). Scarcity drives which giants’ shoulders future research will stand on.

Hassler, Krusell, and Olovsson (2021) came up with a model for this process, estimated on postwar US data with the 1970s oil shocks as the key episode, indicating that firms substitute energy-saving technology (R&D targeting energy efficiency) for labor- and capital-saving technology (the more classic productivity-enhancing improvements) when energy is scarce. As a result, they attribute part of the post-shock productivity slowdown not just to the direct growth-constraining effects of scarcity, but also due to the adaptations firms made in response. They find that from a productivity perspective, technology-driven efficiency is a necessity if scarcity is unavoidable (firms are optimizing for the scenario they’re dealt), but it’s very far from a free lunch.

Overall, this is a kind of counterintuitive way to think about efficiency. But I think some of the dissonance this might provoke is due to two increasingly outdated heuristics that don’t fit cleanly onto today’s (or tomorrow’s) energy system.

Reassessing the efficiency - abundance tradeoff

The first of these is the nature of scarcity. Firms aren’t making an irrational substitution when they redirect budgets towards adaptation – they’re intentionally maximizing their future profits based on their expectation of the binding constraints, now and in the future. So if we expect that energy scarcity will last a long time, it makes perfect sense to invest in efficiency; conversely, if you’re facing a short-term constraint, you might decide not to invest in efficiency and just accept a period of constrained output.

For most of the waning 20th and early 21st century, we lived in a world where the predominant energy supply was derived from fossil fuels and therefore finite. In the context of a “peak oil” world, accounting for scarcity was not only rational but unavoidable – the long-term decline of economically viable fossil resources was fundamental to the way we thought about energy usage. In practical terms, this meant that efficiency would be an (eventual) economic necessity, even before any climate arguments.

The second, of course, is the climate benefit associated with using less energy. In a fossil fuel world, nudging the path of innovation towards efficiency has meaningful climate benefits – less energy used means less emissions produced, and as a result efficiency itself could be a useful knowledge stock. In that world, it’s tricky to weigh foregone productivity improvements against the efficiency improvements that crowded them out (especially if the replaced knowledge stocks were in fossil energy).

Both of these frameworks, fully reasonable in their contemporary context, helped to stack the deck in favor of efficiency. However, neither of them applies with quite as much force today, and it’s likely that they’ll become increasingly unhelpful ways to think about the tradeoff. To start, under a bullish clean energy / electrification scenario, the climate benefits of reducing energy use are severely undercut.

Second, to put my abundance hat back on, you’re not supposed to inflict scarcity on yourself. When models like Hassler’s find efficiency to be an economically optimal response, that’s assuming that scarcity is an unavoidable fact of life. In contrast, one of our most important constraints on energy supply in the US is permitting and regulatory systems that delay new energy projects, preventing new supply from coming online fast enough to meet demand. Few would argue that such a system should be permanently enshrined in its current form. Especially in an era of rapidly falling solar and battery costs, there’s no reason to think that costly firm-level adaptation is an economically optimal, or even a particularly effective way to meet our goals for climate and growth.

In effect, we’ve imposed a constrained optimization problem on firms. Rather than deciding how to allocate R&D resources based on their best guess of real energy availability over the next decades (which to be clear, could absolutely involve efficiency or other forms of adaptation), we’re forcing firms to make those decisions based on artificial scarcity. It’d be better to take on those institutional handicaps directly – by streamlining permitting and interconnection, for example – than to force firms to work around them at what might be an exceptionally high societal price.

When we think about efficiency as costless, or judge its net cost against that of facing an unavoidable energy constraint without adaptation, we’re not using the right counterfactual for an artificial scarcity world. The right way to measure the costs of adaptation to artificial scarcity is not against the costs of not adapting, all else held equal – it’s against the potential growth we would have enjoyed in a world where that redirection never had to happen.

Conclusion

Are we headed into an entirely new growth regime, where the relationship between energy and GDP will be fundamentally different and none of this will apply? Maybe! However, even if we are, we should still ground our discussion in our best guess about how the mechanics of energy and the economy actually work. That doesn’t mean making wildly conservative assumptions, but disciplining the assumptions we do make, without having to abandon our empirical grounding.

I also want to be clear that there are plenty of ways I could be wrong here. Radically available energy could guard against inflation, helping bring down the chances of recession. Cheaper electricity could make the electric tech stack more competitive, with long-term implications for the direction of American innovation. The question remains, though, whether energy prices are really the main, or most direct, channel to the desired outcome, or whether you mostly just need better battery energy density for that flying car.

1. Where is housing slow to build? — Alex Armlovich

2. Giving up on homeownership — Matt Clancy

3. Taste vs. cost — Nisha Austin

4. Why biotech companies go to Australia — Saloni Dattani

5. The real math was the dead-ends we found along the way — Jordan Dworkin

6. Data center announcements and property values — Willow Latham-Proenca

Where is housing slow to build? — Alex Armlovich

Evan Soltas has a big first-of-its-kind paper measuring permitting and construction durations for standardized housing projects across 60 US cities from 2000 to 2025: A standardized 20-unit apartment building takes 4.6 years from permit submission to completion in coastal cities like LA, New York, and San Francisco, but just two years in Sunbelt cities like Raleigh, Orlando, and Phoenix. The fastest cities finish entire multifamily projects in the time the slowest cities take to merely approve a permit. Across cities, each additional year of permitting time is associated with 39% higher rents, an economically and statistically huge relationship that survives controls for income, density, and neighborhood characteristics.

Tragically, permitting is getting slower: the probability that a multifamily project gets approved within six months fell from 71% in 2000 to 44% in 2025, holding project characteristics fixed. Construction time fluctuates with the business cycle but shows no secular trend…in other words, the problem is permitting, not building. This paper is the knockout blow the YIMBY movement has been missing: systematic evidence that permitting generally, and permitting delay specifically, does indeed belong in the “growth control toolkit” hall of shame as a first-order regulatory driver of housing costs alongside zoning. Beyond the paper’s findings, I like to note that obvious out-of-pocket carrying costs during development aren’t the only source of permitting time project savings: Because of the pure time value of money, moving the exact same cashflows forward in time mechanically increases returns and project viability even if carrying costs are zero.

Giving up on homeownership — Matt Clancy

As I’ve written before (see item #3 here), the housing shortage is disproportionately affecting the young, with first time buyers and renters paying unusually large shares of their income to housing, compared to long run trend. A 2025 working paper by Lee and Yoo looks at what happens when owning a home becomes out of reach. They argue that when people decide that a home of their own is unattainable, they give up on home ownership which leads to a cascade of behavior changes: less saving, less hard work, and more interest in riskier investments.

Lee and Yoo present a variety of evidence to back up their argument. The share of Millennials who report they expect to rent forever has risen from under 15% in 2018 to 25% in 2022, and the share who say they currently have $0 saved for a house down payment is up from 48% to 67% over the same period. Lee and Yoo also conduct several comparisons of renters and homeowners, focusing on people with a low to medium net worth: this group has higher monthly credit card spending, is more likely to say “ Working hard: always giving my best effort” is not important, and more likely to invest in cryptocurrencies. These behaviors can be reinforcing; those with an income just below the threshold needed to buy a house end up with much less wealth and income than peers whose income is just above the threshold. The paper is a bit of a grim complement to the set of papers I recently wrote about for my post The non-material consequences of growth, which looked at how slow growth is associated with less trust in your government, less trust in democracy, and more zero-sum thinking.

Taste vs. cost — Nisha Austin

My career began in architecture, so reading Robert Kwolek on whether people prefer traditional buildings and Samuel Hughes on ugliness and housing, I wasn’t surprised that architecture students want different things than everyone else. We do! Part of it is that we are trained to imagine ourselves standing in the room rather than looking at the facade, and that is hard to unlearn. And part of it is that we think “modern” is the harder problem. Classical at any scale is often reproducing something already designed, or scaling up something that already exists, and I’d argue plenty of it isn’t really designed at all, certainly not to the degree a context-appropriate modern building is. Also, a lot of what these surveys call modern is actually postmodern, which is a different conversation.

Kwolek ends on the question worth pointing out here: “Why are we commissioning styles that we say we dislike?” I think the answer is mostly that almost none of us are commissioning anything. Architecture and engineering came to 1.5%, or $6,480, of the $428,215 it cost to build the average American home in 2024. Only about one in five new single family homes is built for a particular owner on their own land, and the ten largest builders account for 44% of all new single family closings.

I have a related personal anecdote: we are swapping our home oil system for heat pumps, which means sealing the envelope, which means replacing the siding, which is about as close to a free opportunity to change how the house looks as we will ever get. Both the material and the design costs of going more contemporary would have been significantly higher, so we’ve landed on a nominal material update and effectively no aesthetic change (same color, scale, etc), though I would have readily gone the other way if the cost had been comparable. I assume a lot of these types of decisions land here, and for the same reason, and that most of them are about houses that already exist rather than houses anyone is commissioning.

Of our collective affordability problems this is very much the least of them. But if a dread of ugliness really is hanging over our cities, maybe the lever is cost rather than taste. About forty jurisdictions now pre-review house plans and publish them, which either solves the problem or hands everyone the same house, and I don’t actually know which. How to make good design cheap and widely available is a question for another day!

Why biotech companies go to Australia — Saloni Dattani

This week I read what I wished I was reading last week: a detailed explanation of how Australia’s clinical trial system compares to that of the United States. The key difference is in their phase one trials, which are the first test of a new drug in humans. They’re used to get an initial sense of the drug’s safety, but can also provide biological clues as to its efficacy, and be used to learn, iterate and improve, before drugs progress into the later and more expensive stages of clinical trials. In a recent report for IFP, Ruxandra Teslo and Adam Kroetsch explain why Australia does these trials faster: in short, they use more streamlined pathways and make asks that are proportional to risk.

I found the specifics quite interesting and sometimes surprising: in the US, the FDA decides whether a phase one trial can proceed, while in Australia, this is left to independent ethics committees. Drug developers in the US generally get only one formal meeting with the FDA before submitting their application to start a phase one trial, while Australian committees can meet with them weekly, which helps them learn more precisely what to include. FDA reviewers typically ask for months of stability testing for the drug, even if it is only used for hours, days or weeks, while Australia asks for testing that covers the dosing period. Australia also accepts research-grade components for the laboratory tests in an application, rather than pushing developers toward using commercial grade components, as the FDA tends to do. Australian developers can also submit different components as they become ready, rather than waiting to compile everything at once. And while US applications face a fixed 30-day waiting period before the trial can proceed, Australia’s do not. All that results in a more streamlined process that’s, in my view, more reasonable, predictable and, on average, several months shorter.

This all matters because delays have a range of consequences. One, for learning: the process of iteration and learning happens faster if scientists can start running trials faster, which hopefully improves a drug’s success rates at later stages. Another is for patients in the trials themselves: if they have a terminal disease, months of delay can mean it becomes too late for them to receive the drug. And this effect applies more broadly to patients outside trials too, if the overall timeline (from research to an effective drug becoming available) is reduced. It’s a great example of how, sometimes, saving more lives isn’t about improving a new drug’s efficacy, safety, or accessibility, but just getting it to patients faster.

The real math was the dead-ends we found along the way — Jordan Dworkin

Back in May I wrote about David Bessis’ essay “The fall of the theorem economy”. In it, he argued that theorem proving has historically been a reliable proxy for producing conceptual innovation and insight in mathematics, but that this coupling was contingent on the manner and pace with which mathematicians conducted, shared, and canonized their work; AI – by breaking that coupling, hyperoptimizing on the proxy itself, and making an end run on the socially rewarded output – could thus dramatically speed up result-generation without a corresponding increase in knowledge.

The past few weeks have brought these concerns back into the spotlight. On September 8, OpenAI announced a resolution of the Navier-Stokes existence and smoothness problem (one of the Millennium Prize problems) produced by an internal model running roughly 10,000 coordinating agents. A few days before the announcement, Terry Tao wrote a thoughtful and prescient thread laying out why scenarios like this troubled him, somewhat eerily using Navier-Stokes as his animating example. His concern was an extension of Bessis’: problems like Navier-Stokes are valuable primarily for the insights thrown off in the process of trying to solve them, rather than in the solutions themselves. The failed paths, the new obstructions, and the unexpected connections can add more value to the field than a regularity result itself adds to our knowledge of fluid dynamics. An agent swarm that runs that entire iteration internally in a day and then emits only the solution is therefore producing a fraction of the value that a comparable human-led solution might provide, even if the proof itself is easy to understand and canonize. If the underlying process is shared, it may be possible (with a lot of help from AI agents) to sufficiently parse the preceding work and recover some of that insight; but not when the process is obscured by the labs or researchers. A few days after the OpenAI announcement, 25 Fields Medalists released a signed declaration titled A Severe Misalignment of AI in Mathematics. The declaration makes a series of related arguments, warning against rushed, benchmark-driven production of proofs and highlighting the misalignment between the goals of AI companies and the math community.

Do the growing concerns of mathematicians generalize to other domains of AI-for-science? On one hand, “the real treasure was the dead-ends we found along the way” is a dynamic that applies fairly broadly across basic research. On the other, many fields have made enormous progress using something approximating a “discover first, ask questions later” approach (drug discovery being a prime example), and it would be a great day if an agent swarm spat out the molecular structure of an Alzheimer’s preventative even if it came with no mechanistic explanation or stepping stone knowledge. Upstream of cures, however, there are many questions for which it is not obvious whether the answer or the search would provide more value. It will be increasingly important to figure out how to tell those types of questions apart.

Data center announcements and property values — Willow Latham-Proenca

A new working paper by Garrison Schlauch and Simon Greenhill last week tries to estimate the property value impact of large data centers, based on a new sample of nearly 350 data center announcements since 2003 (drawn from a dataset of nearly 1000). For each announcement, the authors trace surrounding home prices for 5 years before and after and compare with matched comparison areas, first with the announcement and then with the opening of a new data center.

Consistent with increasing public salience, the authors find that before 2021 data center announcements barely moved prices, with most of the discount concentrating after the center opened. Starting in 2021, discounts became significantly more front-loaded, with the price drop shifting almost entirely to the announcement. Unfortunately, there’s not enough data to conclusively say much about the trend in the total discount over time (the current data doesn’t find much of a change to the sum of the post-announcement and post-opening discounts, but it’s too small a sample and too short a window for a definitive answer), which would be an interesting check on how public opposition translates into home prices. Larger data centers and those in areas with fewer existing data centers, though, do see higher discounts.

The effect is, unsurprisingly, higher for homes closest to the data center – while the authors find a roughly 3% average discount for homes within a 3-mile radius of a large center, it jumps to about 7% for homes within a mile (this is, as the authors point out, net of the capitalized benefits of tax revenues or other community amenities). This relatively tight geographical concentration of property value impacts aligns directionally with the wind farm impacts we discussed earlier this year. Both cases illustrate that we’re still clearly short on targeted compensation mechanisms that align with the geography of impact.

Here are a few other highlights from our team and grantees, and announcements worth sharing:

• Matt published "The non-material consequences of growth," a review of the academic literature on how slowing growth shapes trust in government, support for democracy, and zero-sum thinking.

• Arnold Ventures and IFP launched US Energy Data, a clean, searchable interface for EIA energy data covering prices, generation, demand, and grid reliability across all 50 states.

• Ruxandra Teslo published an op-ed in the New York Times on clinical trial reform, a nice complement to Saloni’s blurb this week on the IFP report Ruxandra co-authored.

• The Inclusive Abundance Institute published their September newsletter.

• The Urban Climate Solution is running during Climate Week NYC next week (September 21-25), with ten free sessions on housing abundance, parking reform, and transit investment as climate strategies.

• John Mangin was appointed to NYC's Board of Standards and Appeals, a win supported by Open New York.

• Asimov Press is back! They are returning from a six-month hiatus with a sharpened focus on intensive reporting in biotechnology.

• CleanEcon launched Energy Frenemies, a new podcast co-hosted with ExxonMobil’s Vijay Swarup and CEBA CEO Rich Powell.

This week we’re drawing some inspiration from Jorge Luis Borges and writing about articles and books that don’t exist. We didn’t read any of the following pieces because, so far as we know, nobody has written them yet. But we wish they had!

Matt Clancy: This week I wish I was reading “What’s a PhD cost and what’s it worth?” a new NBER working paper. The paper (would) trawl through the last few decades of NSF and NIH PhD training programs to identify quasi-random surges in funding for PhD students. It would use these to estimate both how much it costs to “create” a new PhD, and what the impact of a 10% increase in the PhD labor force is on the rate of innovation and economic growth. On a dollar-for-dollar basis, it would find the marginal impact on growth of funding a new PhD is [redacted] compared to funding original research. The paper would also find [redacted] programs tend to produce PhDs at the lowest cost, suggesting we might want to do more of them.

Jordan Dworkin: Four years ago, in “A Vision of Metascience”, Michael Nielsen and Kanjun Qiu asked the reader to imagine they were “a science fiction writer depicting a scientific discovery made by an alien species.” It seems likely, they say, that “such aliens will have radically different social processes to support science. What would those social processes be? Could they have developed scientific institutions as superior to ours as modern universities are to the learned medieval monasteries?” This week, I wish I were reading a series of vignettes creatively answering that call, exploring how scientific processes and institutions might look under the stewardship of dramatically different societies, cultures, minds, and technologies. Duly inspired by related endeavors from David, and Scott, and Scott, and Corin, and Scott again, these yet-to-exist vignettes have a flair for the humorous and bizarre, while also giving us valuable new frameworks for thinking about what is and what could be.

Willow Latham-Proenca: Maybe more Calvino than Borges, but I’d like to be looking at a constellation of new papers following a huge geographic and sectoral expansion of (something like) the Critical Mineral Project dataset, a cradle-to-grave project tracker. Unlike most public data, this version not only shows how long a project spent in permitting, but time spent in pre-permitting discussions and/or litigation (two common ways a project can end up dead by attrition). With the grunt work of compiling datasets out of the way, future researchers can focus on parsing out which provisions really drive up time and costs (like this and this), and investigating which are most cost-effective in protecting public welfare (like this) – a roadmap for evidence-based future policy.

Nisha Austin: This week I wish I were reading something about which ~five numbers tell you a policy field is growing rather than just getting louder. I'd like to see some metrics around things like how many paid jobs there are in the field, how many organizations reach year three, how many people leave advocacy for jobs in government, and what share of the field's money comes from someone other than the funders who started it. Some of this is already countable: there is a jobs board with roughly 280 listings on it today (worth noting this is an aggregator of adjacent orgs and not necessarily a single field), and the best public count of pro-housing groups is still more than 140, assembled off social media in January 2022 by two researchers who admitted they had probably missed some. Someone please go count them again!

09/10 Edit: Sightline reached out to say they've done a version of this. They tracked down the pro-housing groups that were not purely volunteer operations and asked them what they spend on housing work. The answer: “at most $40 million in the United States in 2023," against an estimated $4 billion for the nonprofit climate movement the same year. It's a spending tally rather than a headcount, it leaves out all-volunteer groups so it covers fewer organizations than the Brookings inventory, and it's now three years old. But it's a good example of the type of information I'm eager to see more of, and it turns out we helped build it!

Alex Armlovich: This week I wish I were reading an analysis of the consequences of the regulatory cliff between America’s single-family “International Residential Code” (IRC)1 and our multifamily “International Building Code” (IBC). One hypothesized reason for the 20th century disappearance of “missing middle” multiplex buildings–duplexes, triplexes, quadplexes, etc–is that beyond the direct zoning bans on them on most land in most US cities after WWII, these buildings also trigger stricter IBC rules, raising costs enough to make multiplexes uncompetitive compared to attached rowhouses built under looser IRC rules2. Architects in the New Urbanism movement, predating the modern YIMBY movement, have long advocated for allowing multiplexes similar in overall size to a single family mansion to be able to use the IRC rules available to mansions, but face opposition from the legacy fire protection lobby, who argue the IRC isn’t safe enough.

Unfortunately, there is no research on the fire safety increment at the IRC-IBC cliff3. The old-fashioned fire officials who oppose IRC rules strangely don’t bother collecting fire safety data by building code type, making it impossible to decisively assess the debate. Detailed fire outcomes data would allow quantitative social scientists to perform a social benefit cost analysis of the sort typically required by federal rulemaking4. America’s building codes have added provisions written by vibes, hunches, and industry vendors for so long, and with so little outcome data collected, that even the most basic questions are debatable.

Thus far YIMBYs have pursued a “no-regrets” comparative approach to building code reform, pushing for egress and elevator rules that have already been proven safe in Europe and elsewhere. But since Europe has no equivalent codes for America’s IRC-style light wood frame buildings, the champions of the international comparative approach (like the Center for Building in North America) have been silent on the IRC-IBC multiplex question. So that’s what I wish I were reading: If not a full-fledged federal RIA-style analysis, then at least a basic fire safety analysis to demonstrate whether IBC rules even meaningfully affect life safety in the first place (let alone whether they do so cost-effectively). All we know right now is that fire death rates are extremely low in all forms of new construction–there is not yet publicly available evidence, nor do fire regulators even attempt to collect usable evidence, that there are meaningful safety benefits from any of the incremental regulatory cost triggers in the IRC, in the IBC, or between the IRC and IBC5.

Saloni Dattani: Last week I thought, “I’d love to read more about Australia’s clinical trial system and how it compares with the United States!”, since Australia is known for having a much faster process for starting early stage clinical trials. But then, on Friday, Ruxandra Teslo and Adam Kroetsch published a report for IFP on exactly that topic, which was great, and I thought “Wow, that was fast!” So now I’ll suggest that I’d like to read about the institutional history of TIMI, the academic research organization that has coordinated more than 80 trials to test treatments for heart attacks across 50 countries, and has massively influenced cardiovascular medicine globally. A whole book-length treatment would be the dream, but I’d also be satisfied with a detailed report or book chapter. Hopefully that will be published next week too.

Finally, here are some announcements from our team and grantees:

• One bittersweet announcement: Dylan Matthews is leaving the Abundance and Growth Fund to join Caleb Watney’s new US AI Policy team at Coefficient Giving. Dylan will also continue to be a guest contributor on the AGF blog, so we'll only miss him some of the time. I know he’ll do great work with Caleb.

• Coefficient Giving is now on Substack! Subscribe now, and think of it as your one-stop shop for the best of CG thought across the internet.

• CG also launched Launchpad Funds, an RFP for new funds organized around one outcome and led by a “general manager” accountable for it. Expression of Interests are due October 2.

• 1DaySooner is hiring for a few roles, including a research fellow for their CTA team and a project lead to shepherd CTD Commons through its initial planning period and to create the momentum and proof points needed for a successful public launch.

1. California’s year of fees — Alex Armlovich

2. Observations from Europe — Nisha Austin

3. Can antibody explain these images? — Saloni Dattani

4. The future of the university — Jordan Dworkin

5. The antimicrobial resistance problem — Matt Clancy

6. Texas transmission and the Permian basin — Willow Latham-Proenca

7. The public housing miracle that never happened — Dylan Matthews

California’s year of fees — Alex Armlovich

California housers informally dubbed 2026 the “Year of Fees”: On the heels of the SB79 win that will decisively legalize midrise transit-oriented development statewide when it takes full effect in the early 2030s, YIMBYs and legislative champions took a breather on big zoning bills and turned to fee and exaction reform to boost construction feasibility.

Still, a CAYIMBY pure zoning play just passed: AB 2074, Matt Haney’s Downtown Revitalization Act, requires California’s seven largest cities to designate regional transit hub districts with a 150-foot height baseline (and at least 25% of each district allowing 450 feet) where highrise housing gets streamlined ministerial approval. The bill imports the SB 35/423 labor mandate framework requiring prevailing wage on all projects over 10 units, “skilled-and-trained” workforce above 85 feet. Downtown high-rises are already heavily union-built (though not necessarily always paying full closed-shop wages above market rates for every trade). It’s not clear what to expect in production outcomes because downtown highrises in SF have been mysteriously sluggish of late, with even entitled projects citing extreme construction costs as a reason for delay. AB 2074 originally included a $500M revolving loan fund to help subsidize financing to offset the cost of the labor mandates; the enrolled version downgrades it to a CalHFA study. But surging rents, up at least 26% in SF proper over the last 12 months, may yet wake SF-skeptical bank loan committees out of their dogmatic slumber. It would be ideal if this bill successfully unlocks feasible “missing massive” zoning on top of a half dozen of California’s underutilized rail systems, but at the very least this is still a nice solidarity effort by YIMBYs to repay organized labor’s support for SB79 last year.

Observations from Europe — Nisha Austin

My family vacationed (or as they might say, “vacationed”) in Spain over the last couple of weeks, and we walked a portion of the Camino de Santiago from Sahagún to León. We had our three and six year olds in tow, so there were complaints, though many of them were most clearly articulated by the adults in the party.

A huge part of the draw of the Camino was to do something simple together, to appreciate and minimize, which we certainly did. But the things that were missing also stood out. It turns out a lot of the complaints have active policy literature behind them.

1. You can’t eat at a “normal” hour. Spain sits on London’s meridian and runs on Berlin’s clock, because Franco moved it in 1940 and nobody moved it back. Solar noon in Madrid falls after 1pm in winter and after 2pm in summer, so lunch at 2 and dinner at 9 are really lunch at 12 and dinner at 7. Spain did ask the EU in October 2025 to stop changing the clocks twice a year, but nobody is proposing to fix the hour that actually moved.

2. Everyone stays up too late (because of said late dinner), no wonder productivity is bad. Spanish productivity really is low, but not because Spaniards work too little. For example, they work more than Germans, not less: about 300 more hours a year, which is seven and a half extra working weeks. Spanish output per hour is about 63% of Germany’s.

3. These places feel empty. The Sahagún comarca has lost over 30% of its people and holds 6.1 per square kilometre. Four of Spain’s eleven emptiest comarcas are in the province of León.

4. You can’t buy a banana, and we’re two hours from Madrid. 4,632 of Spain’s 8,132 municipalities have no food shop at all.

You would think a potential answer is that more people move to Madrid, but increasingly they can’t. A new IMF working paper finds that without the price run-up in Spain’s high-productivity regions, roughly 63,000 more working-age people would have moved to them between 2017 and 2023. The observed net inflow was 1,700. Empty Spain and unaffordable Spain turn out to be one problem. Meanwhile the Camino set a record in 2025 with more than 530,000 pilgrims and 242,000 of them walked the route that runs straight through some of the emptiest parts of Spain where we were; a study commissioned by the Camino municipalities concluded, politely, that this does not fix it.

It’s worth pointing out that Spain is not short on capacity (windmills were a favorite for “I spy”). Between December 2019 and June 2025 it doubled its wind and solar, adding over 40 GW, more than any EU country except Germany, and its wholesale power price in the first half of 2025 ran 32% below the EU average.

To be clear, none of this made the trip worse (some low key suffering is good for personal growth!). The people still living in these towns are the reason the Camino works at all, and they were very good to us. It did send me looking for the research once we got home.

Can antibody explain these images? — Saloni Dattani

Back in May, science sleuth Sholto David came across a suspicious image from the major life-sciences supplier Thermo Fisher. It was a Western blot showing the performance of one of their antibodies that is commonly used in cancer research. On the blot, four apparently separate protein bands were copies of the same shape, duplicated and flipped or rotated to produce the others.

Since then, Reese Richardson, a metascience and computational biology researcher and other sleuths uncovered many more by increasing image contrast and searching suppliers’ catalogues for repeated visual fingerprints, partly using image-search tools. According to Nature, they have now found over 18,000 questionable images across 15 commercial suppliers, with digital alterations such as repositioned bands, painted-over regions, and reused background noise. For the last category, Elisabeth Bik (who some say has forensic laser eyes) told Nature, “That particular type of image alteration or manipulation suggests that these antibodies were never tested at all in the lab.” Weirdly, identical images were sometimes used by multiple suppliers, which suggests that they came from the same third-party manufacturers.

The images don’t necessarily mean the antibodies themselves are flawed, but since other research has found that commercial antibodies are commonly unreliable, I wonder whether this is just the tip of the iceberg, and how much of scientists’ time has been wasted chasing false leads. One takeaway is that researchers should avoid relying on suppliers’ images and independently validate their antibodies. But given how time consuming and repetitive it would be for individual scientists to validate them, a wider effort seems worthwhile. For example, the Canadian open-science company YCharOS tests some commercial antibodies head-to-head and publishes their results openly, and the University of Leicester’s ‘Only Good Antibodies’ hosts this data on a searchable database, though neither are comprehensive. The sleuths themselves also host a Zenodo repository with their findings.

The future of the university — Jordan Dworkin

There’s been a lot of writing lately on what AI is going to do to the university. The most recent entry is from Kevin Munger, who writes in Asterisk that the threat isn’t that AI replaces universities or makes them obsolete (rather they will survive, as they have for centuries, by reinventing themselves), but the possibility of academics reacting defensively, banning or policing its use and turning academic work into “a sort of dockworkers union of knowledge.” He proposes a middle path, in which rather than using AI to churn out yet more papers, academics let it absorb the grunt work and redirect freed-up time to things only universities have the time horizon and incentives to do well: maintaining decades-long datasets that underpin research fields, reading and carefully engaging with others’ work (an activity he positions as to some extent a return rather than a continuation), and acting as a check on a rising tide of amateur research from a newly empowered class of curious non-experts.

The piece pairs well with other recent work engaging with the future of universities from different angles. In June, Nils Gilman argued that the current form of the teaching university is finished, because the set of activities holding it together (“the book, the lecture, the problem set, the written examination”) are disrupted and dissolved by AI. Back in early 2025, Ben Reinhardt wrote a long essay positioning the modern university’s cacophony of roles as an issue that predates and exists largely independently of AI (in fact, he mentions AI in the piece only once, in the title of a new Princeton initiative). Ben posits that the university’s problem is that it combines too many unrelated jobs into one institution (credentialing, research, teaching, job training, technology development, etc), and that each would benefit from being housed in a more specialized institutional form. Despite the disparate diagnoses and recommendations in the pieces, all agree that universities aren’t going anywhere; they predate many modern nations and may outlast many more (for more on that history, you can read this new NBER working paper).

The antimicrobial resistance problem — Matt Clancy

Antimicrobial resistance poses a challenge to normal incentives for invention. It’s driven by the overprescription of antibiotics, and so any new antibiotics we discover we might prefer to keep in reserve, to be used only in cases where existing treatments fail or when there is a pandemic from a mutant bacteria. But with clinical trials costing hundreds of millions, how do you incentivize a pharma company to invest in a drug that isn’t going to be used?

This would seem to be a good use-case for innovation prizes, which are well-suited to a situation where you want to encourage invention and discovery, but scaling up isn’t necessarily critical. Abi Olvera writes about a recent failed attempt in the USA to create such a mechanism, the PASTEUR Act, which would have enabled the government to pay pharma companies for access to new antibiotics, whether or not they are used. A similar program already exists in the UK, where the NHS can pay for access to new drugs, regardless of whether they end up being used (they call it Netflix for antimicrobials). And the EU is considering a related idea, wherein companies that develop new antimicrobials are rewarded with a data exclusivity extension that can be transferred (or sold) to another qualifying drug, staving off generic competition and raising profit. Olvera writes that the US’ PASTEUR Act was last defeated by skepticism about paying industry for drugs whose efficacy was uncertain.

I can see the argument for not overpaying for a drug whose efficacy in a pandemic is uncertain. But at the same time, neither would it be ideal to find ourselves in an antibiotic resistance pandemic without many options. I wonder if we could find a way to incentivize companies to build a pipeline of drugs that haven’t gone through the expense of a big clinical trial, and so can’t be used at this time, but where initial evidence is compelling. If we later find ourselves in another pandemic, as Saloni has written, it may well be easier to run a high-powered clinical trial fast.

Texas transmission and the Permian basin — Willow Latham-Proenca

Texas has long been an unlikely darling for the renewables industry, rocketing to leading positions in wind and solar deployment driven by a combination of a laissez-faire attitude towards permitting and ERCOT’s connect-and-manage system (where generators can connect directly to the grid without waiting for network-upgrade studies in a lengthy interconnection queue, but accept the risk of curtailment until the system can reliably integrate that new supply). The last leg of the stool is the trickiest – “managing” new generation requires planning and executing major transmission improvements.

Texas has been exceptionally successful in rolling out one of the nation’s most ambitious transmission projects, the roughly 3,600 miles of high-voltage lines that connected West Texas’s vast wind (and later solar) resources with cities further east under the CREZ initiative. Now the state has another ambitious transmission plan – bringing high-voltage lines back to West Texas, this time to supply oil and gas production (and likely data centers) in the Permian basin. Like CREZ, it’s faced opposition from landowners, in part because the state legislature streamlined its transmission review process in the interim.

Another part of what is making this challenging, beyond the streamlined siting process, is who’s benefiting. Part of what’s traditionally made it easier to plan transmission in Texas is ERCOT’s islanded grid – since the grid (roughly) shares the state’s borders, there are no interstate fights over how to split the cost of new infrastructure. However, state borders aren’t the only fault line on cost allocation. Under Texas’s current PUC rules (which the legislature has required the PUC to update by December), transmission costs are currently socialized across all ratepayers (including residential users, who have a harder time shifting their use to avoid charges), another hurdle for the planned rollout.

The public housing miracle that never happened — Dylan Matthews

I like Jerusalem Demsas’s writing on basically any topic, but she’s at her best when she can dive into housing policy. Case in point: this devastating takedown of a deeply wrongheaded op-ed and report from two Vanderbilt law professors who tried to argue that land use and zoning rules are a distraction from building out massive quantities of public housing the way the US did in World War II.

Demsas’s response is twofold: first, any way you look at it, the primary cause of high housing costs in the US is the plethora of laws and regulations that reduce the supply of housing available. Second, that public housing buildout during World War II? It mostly built purposefully sub-standard homes where families had to share bathrooms and kitchens. Because it was literally World War II, people accepted the sacrifice. But after the war, when we needed permanent homes, it was time for the private sector to rip.

Here are a couple highlights and announcements from our team:

• Alex published a spotlight last week on building code reform in Chicago, covering how a 2019 building code change (not the more famous Connected Communities zoning reform) appears to be the bigger driver of Chicago’s five-story building boom.

• Our CG colleague Oliver Kim appeared on Planet Money to discuss South Korea’s entrepreneurial culture.

Let’s start with what Daniel found in Chicago: five-story apartments are back for the first time in double-digit annual volumes since a parking-centered downzoning in 1971, and building code liberalization is probably the biggest driver. At Chicago DOH, Daniel helped pass the Connected Communities zoning reform that relieved costly parking mandates for these and other buildings in 2022. For several years he had casually assumed that Connected Communities was the key reform that unlocked the new five-story buildings he saw around the city. But upon using Claude to compile Chicago’s permit data (instead of relying on the flawed and highly aggregated Census Building Permit Survey data), he found that five-story buildings actually spiked in 2021, a full year before Connected Communities!

Go read Daniel’s post for the answer to the mystery:

So what happened in 2021? Well, Chicago’s 2019 building code reforms went fully into effect. And Chicago’s 2019 building code reforms included a change that allowed developers to use a form of construction that had previously been capped at four stories, at five stories […]

[…] Which means that when the City allowed Type III to be used for five-story buildings in the 2019 building code, developers leapt at the opportunity.

Five-story buildings went from 100% Type I construction before the implementation of the 2019 building code to 83% Type III construction after the 2019 code allowed it. Suddenly, because Type III offered a significantly cheaper way to build, developers could afford to build five-story buildings profitably much more often, and the number of five-story developments jumped by more than 100% overnight.

Daniel had a prominent DOH legislative affairs role in passing Connected Communities, while building code reform was handled in a different city department. That could have biased his first guess at the cause of the recent five-story building boom towards the zoning reform he knew best. He now suspects building code was probably more important at this margin for these projects in these locations. But before the anti-YIMBYs crow “it’s not all zoning”1, be sure to read his whole post to see that Daniel found that zoning reform to ease costly parking mandates still did in fact help juice these projects.

Land use regulation is full of one might call gating bottlenecks (a fully-binding constraint, for which any one gating reform produces no observable housing production results until all gates are lifted) and throttling bottlenecks (partially-binding, but not wholly prohibitive of housing production). Gating bottlenecks in zoning are still the first-order problem: Where five-story buildings are simply illegal, there’s no way to even test building code changes that lower construction costs for five-story buildings. But throttling bottlenecks in building codes (like Type I materials mandates where they’re not necessary) can sometimes have larger effects on building feasibility than throttling bottlenecks in zoning (like parking mandates, at least where the mandates are expensive but not wholly prohibitive). Of course the nice thing is that lifting throttling bottlenecks has an interactive effect: Specifically, these reforms complement rather than competing with each other, and the easiest ones are worth fixing for free, or at least fixing very cheaply in terms of city staff time and capacity for implementation.

My final note on all this is the implicit upshot for AI and expertise: Daniel was thoughtfully skeptical, then intrigued, and now empowered by these new AI tools. These tools helped him evade a typical bias of human expertise (assuming the partial solution we know best is the most important single driver of that solution) by testing his hypothesis against the primary source data. He used new AI tools that give him direct and customizable access to go straight to the city’s formerly-unwieldy primary permit data to test and then revisit the hypothesis, all while ensuring the AI research assistant isn’t missing offline context that is common knowledge for a senior DOH executive and accomplished urban history author.

Manually collating the permits (tragically necessary to avoid the problems with legacy Census BPS data) is a painstaking effort. Before AI processing tools, a project like this used to carry a data cleaning and analysis workload equivalent to an undergraduate honors capstone thesis. Now that analysis can be performed under expert oversight for a quick (yet revelatory) blog very quickly. I had that very experience drafting the recent Transit-Oriented Development post on this blog: The data was all free and public, it just takes a lot of time to process it manually. Blogs aren’t peer-reviewed research, of course, and neither Daniel nor I are representing these as such. But as a serious first draft for a publishable white paper, these blogs are great examples of the future of “casual causal inference”, first-draft event studies, and simple but powerful policy-scoping analyses by expert practitioners assisted by (and adding value to) well-supervised AI tools.

1. Personalized cancer vaccines are now real — Saloni Dattani

2. Downweighing review scores at NIH — Jordan Dworkin

3. Housing supply laws across the country — Alex Armlovich

4. Housing and innovation: evidence from Germany — Matt Clancy

Dylan, Willow, and Nisha are all out this week!

Personalized cancer vaccines are now real — Saloni Dattani

Researchers have spent decades trying to develop personalized cancer vaccines, and last week, the first one succeeded. It’s called intismeran autogene (mouthful), was developed by Moderna and Merck, and involves sequencing melanoma patients’ cancer mutations, creating a custom mRNA to target them, and delivering the vaccine to stop the cancer recurring. The news came as the drug passed its interim goal in a phase three trial, where it was given to people with high-risk resected melanoma.

It led to Moderna’s stock jumping 177%, which was apparently the largest one-day gain by any S&P 500 stock this century. I found the size of the surge surprising because you could have already been fairly confident that the drug would succeed since at least June: as I wrote in an earlier post, the company presented data from their phase 2b trial at an oncology conference almost three months ago, which showed a 49% reduction in the risk of recurrence or death (p = 0.0075). But, according to financial analysts, the huge market reaction was probably down to a short squeeze and the belief that mRNA could be a more widely useful platform to develop other cancer vaccines too.

That makes sense, but it’s worth knowing that melanoma is a much easier disease to develop cancer vaccines for than others. It carries a very high number of mutations, so you have thousands of possible ‘neoantigens’ to select from and can choose those that look strong on many measures, but this isn’t the case for many cancers. Many cancers have far fewer mutations to select from, and it’s currently hard to predict whether their mutations will be displayed or measure which ones actually are displayed.

But very few people thought this attempt would work either, until recently. According to a very interesting new STAT article detailing how this cancer vaccine came about, Moderna’s own executives didn’t expect it would work until 2023; it first entered clinical trials in 2017. In case you’re interested in more about personalized cancer vaccines, I’ll once again recommend this great podcast episode, How to design a cancer vaccine, hosted by Owl Posting and featuring Alex Rubinsteyn and Benjamin Vincent, two scientists whose work focuses on cancer immunology and vaccines.

Downweighing review scores at NIH — Jordan Dworkin

Last week the NIH released an RFI proposing a change to how it uses and reports peer review scores. Instead of receiving a numerical impact score and percentile, applicants would learn only which of three bins their proposal landed in: “most competitive” for the top 25%, “competitive” for 26-50%, and “not discussed” for the rest. Overall impact scores and percentiles would also no longer be shared with program staff and advisory councils, who would instead see the relevant bin, the written critiques, and the specific criterion scores.

The stated rationale is that the numerical scores falsely suggest a high level of quantitative precision, and are middlingly predictive of downstream outcomes; thus, more qualitative buckets can give applicants and program staff a higher-level sense of relative quality while mitigating the likelihood of either actor overindexing on score-based rankings in their assessment of proposal quality. Indeed, I have heard from NIH institute directors that program officers’ decisions often hew too closely to percentile scores, rather than incorporating reviewer scores, reviewer comments, and PO discretion into a more holistic portfolio creation process. It is reasonable that removing the opportunity to simply rank and fund to the cutoff could encourage more consideration of written comments and disciplinary priorities (NSF already approaches their review and decision-making in something closer to this manner). But many scientists and science policy practitioners are reasonably wary, given the broader context of the OMB proposed rule, ongoing grant terminations, and politicization of review.

In my view, the most useful piece on this proposal is from DrugMonkey, a longtime pseudonymous blogger on biomedical science. They generally support a balanced approach, and point out that many institutes have historically deviated from the strict percentile order; as a result, this change mostly forces institutes that have historically funded more strictly on paylines to adopt the practices of the more balanced institutes. By DrugMonkey’s lights, the reason to be concerned is not that peer review scores are sacred and deviating from a strict order is itself harmful to science, but that the proposed change could give political appointees more authority to override scientific review while simultaneously obscuring that process. Comments are due October 13.

Housing supply laws across the country — Alex Armlovich

Mercatus Center’s fourth annual state YIMBY legislative tracker is out: 350 housing supply bills introduced across 40 states in the past year, 89 enacted in 28. The headline wins are familiar: California’s SB 79, New York’s SEQRA reform, North Carolina’s parking minimums, Virginia’s YIGBY package. But Mercatus also covers a long tail of state reforms: Idaho passed one of the strongest minimum-lot-size reforms in the country plus ADUs, single-stair, and third-party trade inspections in a single package. Kansas capped residential lot size mandates at 3,000 sq. ft with bipartisan support. One striking trend is geographic diffusion: the YIMBY legislative frontier has moved into moderate-cost states (Indiana, Kansas, Michigan, Pennsylvania) where postpandemic price increases turned housing from a coastal problem into a local one, and there’s some tractability argument for acting before prices get too far out of control. On the minus side, Connecticut repealed its single-stair law, the first outright repeal of a YIMBY-era reform. Hawaii, Minnesota, and Illinois all stalled despite high costs or strong gubernatorial pushes.

Housing and innovation: evidence from Germany — Matt Clancy

There’s an old idea that cities are engines of innovation in large part because they cluster together lots of innovators; these innovators make each other more productive via a variety of mechanisms. Across US cities, a 10% increase in the population of inventors is associated with a 2% increase in the per-person patent output on average (see Wiebe 2026, Table A12, column 7). We’ve always been interested in this idea, because it (partially) informs how valuable housing is: if zoning reform increases the housing stock of a city by 10%, do its inventors eventually become 2% more productive?

Unfortunately, correlation is not causation and pinning down just how real this relationship is has been tricky (see item #4 in this post for an example). So I was interested to find a 2024 working paper by Brian C. Fujiy at the US Census Bureau which looks at this question afresh. Fujiy looks at the unification of East and West Germany as an event that created a large amount of sudden and unanticipated cross-city movement from inventors. Using a variety of approaches to try and control for potential confounders, he eventually estimates that a 10% increase in the inventor population leads to a 4% increase in the productivity of the population’s inventors.

Here are a few other announcements from our team and things that caught our eye this week:

• Patrick Collison, Mario Draghi, and Luis Garicano announced the Rhine Group, a forum to convene European leaders from government, academia, and business to discuss possible answers to get Europe to compete, build, and grow again.

• Matt wrote a piece for Asterisk’s newsletter titled Why can’t technologists and economists agree on AI?

• Jordan wrote a roundup on all things AI and Science last week (alas, already out of date!)

• Works in Progress hosted the second iteration of its Invisible college.

AI is doing math

In the past few weeks, OpenAI announced 10 new (apparently impressive, according to mathematicians) results achieved by an internal model; Anthropic followed with a Claude-discovered Riemann-related result (not a proof, but a substantial improvement on the lower bound).

In light of these results, Fernando Borretti refutes some common arguments for why AI might not transform mathematics.

“Computers are superhuman chess players, yet we don’t care, and continue playing as normal. Why should mathematics be different? The main reason, I think, is that chess is self-contained: results from chess don’t help us understand the orbits of the planets or the binding of drugs to protein surfaces. But mathematics, famously, is the great dynamo of science, the best language and method for understanding the world. A machine that can replace a human mathematician, but better and faster and cheaper, is materially useful; a better chess engine is not.”

Noah Smith, while maintaining that the role of the mathematician will continue, argues that a specific form of heroic individual contribution is likely on its way out.

“We normally think of AI as something that replaces a single human worker, but I think that’s the wrong way to think of it. AI is really a new world-mind — a technology that takes the accumulated findings of individual humans (or robots, or other sensors) and integrates them into a general picture of the world… [E]ach time one of us reports a new scientific finding or expresses a new perspective, that adds to AI’s knowledge and understanding of the world… But unlike human society, AI doesn’t need heroic human geniuses. Its ability to understand complex ideas is not limited by the ability of a single human brain to apprehend, intuit, or communicate those ideas.”

Paata Ivanisvili recounts the pace of mathematical discovery from the front lines.

A small episode from the AI–math battlefield:

Aug 4: a paper appears on arXiv https://t.co/eJyFl3cfAG claiming partial progress on the Bourgain–Brezis Sobolev conjecture.

Same day: I ask AI, “can you try solving the full conjecture?”

Aug 6: AI says, “done.” (The proof attempt and all iterations are in my Overleaf history.) I start asking experts if they would like to check the argument.

Aug 8: another paper appears on arXiv https://t.co/stCwYd7ik7 claiming a full resolution of the conjecture (also with help from AI).

The pace is getting wild.

Tim Gowers explores the question of what aspects of mathematics, specifically, LLMs are good at, and the implications of potential answers.

“If it is true that current models are particularly good at finding examples, that is probably not because they have a particular affinity for existential statements, but more because the proof-discovery methods that are appropriate for finding certain kinds of examples play to the obvious strengths of LLMs: wide knowledge and the ability to explore many paths of the search tree that humans would judge to have a low probability of success.”

Terry Tao gives a lecture on mathematics in the age of AI, and the need for the field to take a more active role in examining its own role.

“What are the precise goals, objectives, and values of our mathematical community, and the enterprise of mathematical research? Not just the explicit goals that we communicate to the public (or to funding agencies), but also the implicit goals that we actually seek in practice?”

And Iris Shi writes a beautiful, difficult-to-summarize-but-very-worth reading meditation on the present and future of math as her PhD comes to an end.

“My physicist friend once asked me what the point of doing research was if someone like Terence Tao could have figured out everything in my dissertation in a tenth of the time. I answered by pointing out that Terence Tao didn’t. Terence Tao did not find a small open problem posited by my advisor and publish a bite sized result making incremental progress. He has only so much time and so many other fish to fry. And in his absence, I was given the chance to touch the edge of knowing and experience the unmatched exhilaration of discovering structure in the labyrinth of everything.”

Is AI doing other science?

There’s also been a flurry of activity in AI for non-math-science, with the first AI-designed virus, breakthroughs in cyclone forecasting,1 a shake-up at DeepMind, and a new Jeff Dean-led AI for science company.

To figure out what we can actually say about AI’s capabilities in drug discovery, an all-star team takes a careful look at the evidence and finds it wanting (if promising).

“The focus of AI in drug discovery must shift from doing what can be done - such as modelling data that is readily available, but that is unlikely to move the needle - to doing what should be done, even if this requires, for example, substantial data generation.”

“‘AI in drug discovery’ can mean two very different things: advanced, rapidly developed methods that lack thorough validation, as well as mature, established methods that have only just become practical to use at scale in production environments. What is common to both situations though is that the translation from ‘AI’ to ‘drug discovery’ has not yet been observed in an efficient and sufficiently durable manner to lead to clinical impact.”

Noah Olsman posits that many AI-for-bio efforts implicitly view biology as chess, when it is actually more like Kriegspiel.

“There is a variant of chess, called Kriegspiel, where each player can see only their own pieces. The game is played with three boards: two players sit back-to-back and move their own pieces while a referee keeps track of both players’ movements and mirrors them on a third board. The players go back and forth taking turns and, each time, the referee tells them whether that move is legal, and if they took a piece. The identity of the taken piece is only revealed if it’s a pawn.”

“Recently I have come to think that the differences between chess and Kriegspiel serve as a good metaphor for the gap between theory and practice in the life sciences. I entered biological research as a theorist who believed that enough mathematics could turn study of life into a game of chess: difficult, but ultimately legible. After spending years as an experimentalist, though, I have come to think that biology is much closer to Kriegspiel. We make moves (experiments) with only partial information, infer the hidden state of the board from sparse feedback (data), and often learn what mattered only after the consequences have unfolded, sometimes weeks or months later.”

Meanwhile Carlos Outeiral lays out the case that the gap in performance between AlphaFold 2 and AlphaFold 3 is because both rest on modeling evolution rather than physics, which works better for shape (AF2) than binding (AF3).

“The thesis of this essay is that the root cause of the slowdown is reliance on coevolution. The fundamental breakthrough in AlphaFold 2 was its ability to mine the evolutionary history of a protein to rapidly explore conformational space. Within a single protein, residues that touch in the folded structure (“contacts”) are constrained to evolve in concert… So if you align a protein’s sequence against millions of its relatives in genomic databases, these correlated changes capture a faint statistical shadow, left by evolution, of which residues sit close in space. … Put another way: the model appears to have learned the physics of biomolecular interactions, when it has largely learned protein genealogy.”

A team at DeepMind discusses both the promise of AI agents for speeding up scientific ideation, and the yet-unsolved gap in speeding up the validation of promising ideas.

“Most scientists have no shortage of ideas. The challenge is knowing which ones to pursue. This is where agents are starting to help. An agent can digest a field’s accessible literature, making connections across disciplines that no single researcher would have time to trace.”

“At the moment, however, the validation gap in most disciplines is widening, not closing. An agent can propose a novel genetic lead to reverse cellular ageing, but cannot say definitively whether it actually works. This explains why companies like Google DeepMind, Ginkgo Bioworks and Lila Sciences are investing in automated labs. But they only suit some fields, are expensive to build and are still early in development. And even automation cannot rush nature’s clock.”

Daphne Koller breaks down the misalignment between where AI-for-drug-discovery efforts focus and where the actual bottlenecks in drug discovery are:

“AI will certainly generate new and better molecules at an unprecedented rate, but will it generate drugs that unlock diseases for which there is currently no meaningful treatment? … [T]he biggest step functions in our ability to drug the undruggable have historically come not from better molecular design tools, but from expanding our repertoire of therapeutic modalities: first biologics, then siRNA and antisense oligonucleotides, then gene editing. Each new modality opened a class of targets that was simply inaccessible before.”

And the CRUX team, in a study led by Peter Kirgis, Sayash Kapoor, and Arvind Narayanan, look beyond AI agents’ ability to make progress on clean, verifiable tasks, and test whether they could conduct open-ended, independent research. For now, no.2

“Our key finding was that while agents could solve the engineering problems necessary to do the research, they failed to produce original research at the caliber of a top ML conference. Our main takeaways:

1. The agents lacked the judgment to identify when a problem was adequately solved….

2. The agents lacked awareness about the resources available to them and the timeline for the project…

3. The agents did not creatively respond to feedback about poor research design…

4. The agents did not effectively backtrack from unpromising approaches….

5. The agents did not follow concrete instructions....”

AI is reviewing (and submitting) science

Some aspects of the validation gap (at least in computational or theoretical fields) may be taken on by AI review. This possibility has caused two stirs on twitter in the past week or so, one when Refine announced its partnership with AEA and Econometric Society (discussed here), and another today based on this post from Itai Yanai.

Efforts to evaluate AI review tools often face the issue of mediocre ground truth (i.e. did a human or an LLM think the review was good overall? Did they think the individual comments were substantive?). Paul Litvak instead inserts errors and checks if tools can spot them (spoiler: they seem to do better than humans).

“Most systems work by ensembling, having the LLMs judge each other’s output. That, as we’ll see, is a good strategy, since different LLMs are surprisingly uncorrelated as far as the errors they can catch. Nonetheless, this isn’t satisfying since we don’t know whether even pooled models miss some categories of errors. Using human peer review as the gold standard, as any professor will tell you, is also unsatisfying, since human peer review varies widely in quality. Of Claude’s suggestions, one stood out — inserting errors into papers and seeing whether a given AI system can spot them.”

Beyond reviewing papers, advances in AI-based reproduction and replication of data+code continue apace, this time with a discipline-agnostic tool from Liu, Tjiaranata, and Tan.

“We present VERITAS, the first end-to-end domain-agnostic replication framework built around CLI coding agents. VERITAS supports flexible input modes, extracts the paper’s claims independently of the original authors, and withholds those claims from the replicating agent to mitigate leakage. The pipeline spans claim extraction, replication, and verification, and aggregates the per-claim verdicts into an importance-weighted Replication Score for the paper.”

But what about replication of findings in new data? From an ambitious new benchmarking effort:

“Several benchmarks have been proposed to develop and evaluate LLM agents on research reproduction tasks, including CORE-Bench, PaperBench, and REPRO-Bench… Existing benchmarks operate under the assumption that the new data sample is readily available to the agent. There are no benchmarks designed to assess agents’ capability to replicate research claims in a setting where a new data sample must be retrieved in advance... [W]e introduce ReplicatorBench, a benchmark for replicating published research claims in social and behavioral sciences.”

Let’s close on a would you rather: (a) have an LLM review your conference submissions, or (b) review dozens of LLM-authored conference submissions? I don’t know for sure which they’d pick, but I can say that Caleb Robinson and Isaac Corley did not enjoy the conference review circuit this summer.

“Between the two of us, we reviewed 22 paper submissions this summer, spread across NeurIPS, WACV, and TerraBytes (a geospatial workshop at ECCV). Fifteen of the 22 (68%) contained entirely fabricated citations, fabricated author lists for existing papers, and/or were clearly LLM-generated (e.g. hallucinated technical jargon, nonsensical writing, irrelevant citations). This is called being in the ‘slop trenches.’”

1. The Growth Effects of Energy Shortages - Willow Latham-Proenca

2. America Makes The Wrong Steel For Its Transformer Supply Chain — Dylan Matthews

3. Building New Homes - Alex Armlovich

4. GDP makes you better off, but it’s not the meaning of life - Matt Clancy

Nisha and Saloni are off this week. There will be more from Jordan later this week!

The Growth Effects of Energy Shortages - Willow Latham-Proenca

Energy-rich countries may not be all alike, but energy-constrained countries are also often each unhappy in their own way. Taiwo Hassan Odugbemi, Meron Tesfaye and the Energy for Growth Hub break down the monolith of unreliable power supply in Africa, showing how the grid can struggle in a variety of ways with distinct economic consequences.

Nigeria is the poster child for grid dysfunction, with chronic outages leaving the grid so unreliable that self-generation (usually with diesel generators) exceeds the total capacity of Nigeria’s grid. Transmission is a big part of the problem, likely a main culprit in keeping nearly 70% of the grid’s theoretical capacity offline. Beyond the direct cost of outages, the cost of buying and maintaining expensive backup generators increases firms’ production costs (potentially to 4x that of industrialized peers), and probably keeps firms that can’t afford the cost out of the market. As the authors flag, the “missing growth” cost here is hard to estimate (although apparently easy to invent?), but almost certainly in the billions.

In Ghana, in addition to outages and self-generation, long-dated power contracts rushed through in response to an energy crisis continue to undermine utilities’ financial stability. South Africa illustrates the continuing economic drag of past energy outages, even after (partial) recovery, with an estimated 1.5% GDP loss in 2023 yet to be recovered. Egypt’s case is perhaps the most interesting analog for the US right now – (barely) sufficient grid capacity and limited self-generation for the moment, but an uncertain grid expansion path when staring down the barrel of rapid demand growth.

America Makes The Wrong Steel For Its Transformer Supply Chain — Dylan Matthews

One of my favorite subgenres of business journalism / analysis are articles deeply exploring one particular, hyper-specialized product or resource that turns out to be strangely crucial to the whole economy. A great example is this piece in The Ecomodernist by Ryan Alimento on the production of “grain-oriented electrical steel” (GOES).

A specific alloy of iron and silicon, GOES is used to make cores for transformers. Transformers, an essential component in the grid that allow voltage to be adjusted for different types of transmission, are in extreme shortage, with order waiting times stretching to three years or more as the recent surge in electrical demand in the US forces more and more transmission buildout.

The US only has one steel plant making GOES, however, and its GOES is generally lower-quality and produces less efficient transformer cores than the GOES from Japanese and Korean mills. To make matters worse, the Trump administration briefly imposed steep tariffs on foreign GOES last year until outcry by the power sector forced it to reverse course.

Alimento gives a great breakdown of the problem, and an appealing suggestion for solving it: rather than rely on protectionism and import barriers, offer financing to enable US steel firms to develop new GOES-producing plants. He makes a compelling case that this is one specific area where industrial policy might be highly effective.

Building New Homes - Alex Armlovich

Brian Potter’s reading list has a good housing subsection this week:

First, there’s a striking divergence in existing home price data versus transaction prices at the largest homebuilders: Lennar has cut average selling prices 25% since 2022, overall median sale prices on FRED are down, but the Case-Shiller repeat-sale index for existing homes hasn’t followed. I see one unintended implication of this: Even the largest homebuilders can rapidly pivot to cheaper starter homes wherever they’re in demand, as soon as such homes are legalized in a jurisdiction.

Second, a new CalMatters piece documents that California cities have approved their Regional Housing Needs Allocation (RHNA) housing plans but are building almost nowhere near their targets. “Fair Share”-style housing frameworks like RHNA place a massive implementation burden on localities to proactively plan for growth from scratch, leaving the details wholly under local control. Decades of zero or weak results in these frameworks are beginning to draw more expert skepticism. State laws (like SB79) that prescriptively modify local zoning by default, subject to a local equivalent plan if a jurisdiction has staff capacity to make such a plan, increasingly look like a promising way to bypass the pain, paperwork, and massive staff time losses to RHNA-like processes.

GDP makes you better off, but it’s not the meaning of life - Matt Clancy

GDP per capita is positively correlated with a lot of desirable things, like health and subjective well-being (even though self-reported well-being in the United States has been falling for the last decade, it remains roughly at the level one would expect given our GDP per capita!). But GDP is not everything - it turns out reports of meaning in life actually decline with GDP per capita! Is this because “The meaning of life is a problem for people who are not desperate, people who can count on survival, comfort, security, and some measure of pleasure”, as social psychologist Roy Baumeister once wondered?1

Maybe not. A 2014 paper by Oishi and Diener argues this effect is actually driven by religiosity. Religiosity is positively correlated with meaning; it’s just that religiosity is also negatively correlated with the GDP per capita. If you try to predict meaning with two variables - religiosity and GDP per capita, religiosity carries the day and GDP per capita no longer carries any additional predictive value. Indeed, when the data was collected, the US was actually an unusually religious high income country, and also a leader among high income countries in terms of how meaningful its residents find their lives. US religiosity has dropped substantially since 2007 though, and so I suspect a repeat of the study may not find the same results.

Here are a few other highlights and announcements from our team and grantees:

• Last Friday, Alex wrote about the vast potential for transit-oriented development to end America’s housing problems.

• You can now watch Saloni Dattani’s TED talk!

• California YIMBY’s newsletter, The Homework, gave an update on the legislative session in all things California housing.

• Renaissance Philanthropy has opened applications for the Biosecurity version of its Big if True Science Accelerator

Watch the TED talk

I’ve also adapted it into a longer, companion article for Works in Progress.

I think we’re at a point of divergence when it comes to medical innovation. Our tools are improving rapidly, but our ambitions to use them have fallen behind. Science funding has slowed down, aid budgets have been slashed, and immigration restrictions mean we’re driving away expertise. Even when things have come back online, the disruption and loss of expertise will make it hard to return to our previous trajectory. We could be doing so much more. Medical innovation might not seem urgent in the abstract, but most of us know someone whose life it could change, and eventually I think it affects us all. We can change that.

That’s part of the motivation for my work on clinical trial reform on the Abundance and Growth Fund, but the talk is also larger than that: it’s a call to arms for more people to work on medical innovation, and explains why you don’t have to be a scientist to do it.

I hope you enjoy it!

Reading list

The talk covers lots of topics I’ve previously written about or covered in podcast episodes. Here’s a list if you’d like to read more:

• The Golden Age of vaccine development The first vaccine was a lucky accident. Now we can design new vaccines in weeks, atom by atom. [Works in Progress]

• Death rates from cardiovascular disease have fallen dramatically Over a century of progress in surgery, drugs, prevention, and emergency response has driven down death rates from heart disease and stroke. [Our World in Data]

• Childhood leukemia: how a deadly cancer became treatable Before the 1970s, most children affected by leukemia would quickly die from it. Now, most children in rich countries are cured. [Our World in Data]

• Why we didn’t get a malaria vaccine sooner (written with Rachel Glennerster & Siddhartha Haria) Hundreds of thousands of people die from malaria each year, but it took 141 years to develop a vaccine for it. [Works in Progress]

• Lenacapavir: the miracle drug that could end AIDS (episode co-hosted with Jacob Trefethen) Saloni and Jacob talk about the development and prospects for this new drug, as well as the history of HIV, the initial discovery of retroviruses, and how HIV was transformed from a death sentence to a manageable condition. [Hard Drugs]

• Should everyone be taking statins? (episode co-hosted with Jacob Trefethen) Heart disease is the leading cause of death worldwide, but it’s also one of medicine’s biggest success stories. [Hard Drugs]

• Medical breakthroughs in 2025 An end-of-year round up of the biggest news in biotech and medicine that I wrote last year. [Scientific Discovery]

Robot cars are coming

Waymo last reported a service area now growing to more than 1,400 square miles across 11 cities. We have finally reached the point where “self-driving cars are five years away” is no longer a joke.

To be clear, this is still good on balance. Robot drivers that never drive drunk, respect cyclists, and crash less often than a human driver would be a major public-health improvement. Not having to park is a real urban amenity, especially given the difficult political economy of charging market prices for curb access. Beyond the safety benefits, autonomous vehicles (AVs) could, with sound policy choices like congestion pricing in urban cores and High-Occupancy/Toll (HOT) lanes on highways, complement transit for trips and riders that fixed-route service handles poorly.

But AVs will probably also compete with transit. Ride-hailing offered a premium-priced but convenient door-to-door substitute for discretionary bus trips during the 2010s, while rising incomes and increased household car ownership diverted even more transit ridership in some markets. (Uber became the focal point for the wider phenomenon of urban cars outcompeting urban mass transit during the 2010s, even though it’s surprisingly hard to prove to a peer-reviewable standard that Uber has actually been an important culprit in the transit decline.)

Figure: National Transit Database ridership file rendered by system and mode. 80 out of 115 US rail systems in the NTD file peaked in Unlinked Passenger Trips in 2018 or earlier, long before COVID.

Now, a transit expert might balk that this concern about competition with mass transit concedes too much to both existing ride-hailing firms and to future autonomous vehicles. The first objection is empirical: the stagnation of American mass transit systems in the 2010s is not mostly attributable to ride-hailing alone.4

The second objection is conceptual and longer-term: the densest urban cores cannot realistically be served by single-occupant vehicles. As the well-worn “vehicle geometry” problem frames it, cities just cannot squeeze enough single occupant vehicles on the roads for the volume of people moving through the highest-density urban cores. But in-motion vehicle geometry is not the only problem; there is also a bottleneck in the mechanics of loading and unloading. If you’ve been to a concert or sporting event in San Francisco, you’ll note the vast pool of Waymos trapped in clumsy, low-throughput pickup and dropoff lines while people fumble with the doors and slowly clamber inside. High-throughput mass transit stations with level boarding and multiple wide large-vehicle doors are necessary at the human geometry level to be able to embark and disembark quickly on limited urban core curb frontage.5

But AVs do not need to eliminate rail to create a problem. They only need to skim enough trips to weaken fare revenue and political support, producing service cuts that drive away still more riders — driving yet another wave of higher taxes for worse service. In potential cases where voters balk at higher taxes for worse service, urban rail systems will collapse on a scale unseen since the 1970s, unleashing congestion and serious particulate matter pollution even if the cars are all-electric.6

A physically realistic benchmark goal for national TOD: 40 homes per gross acre

Density alone is not the only predictor of travel behavior: A dense neighborhood with no sidewalks and zero transit service definitionally cannot have any transit ridership.7

Academic caveats aside, residential density is an important and nonlinear predictor of transit ridership. The Transit Capacity and Quality of Service Manual, a policy guidebook published by America’s national academy for transportation science, advises transit and land use planners that the ridership returns to density near transit are steeply increasing with rising density. See the figure below, a copy of Exhibit 3-9 in the Manual:

Figure: TCQSM nonlinear relationship between density and transit demand

The national manual for transit service policy and planning is not omniscient, but it does propose a convenient and long-accepted national density threshold at which transit demand inflects to its most transit-friendly level: 40 dwelling units per gross acre. For federal policymaking purposes, a person living in an urban-density neighborhood above 39 units per acre is more than 41.5x more likely to use mass transit than a person living in an exurban, roughly half-acre zoning neighborhood at 2.35 units per acre.8

This 40 units per gross acre density threshold will form the analytical backbone of our TOD scoping and benchmarking exercise. There is no single correct answer about how many transit trips any given new apartment will produce, because service frequency, mode travel time competitiveness, tolling and parking policy, and other variables can also affect mode choices. But a reasonable lower bound estimate of this hypothetically built-out housing exercise still yields enough new transit trips to fully reverse the national post-COVID transit ridership losses (see below).

A deliberately mechanical benchmark

We’ve established that transit-oriented development matters for saving America’s rail systems from a nasty mix of continued operational decline, lower social returns on higher transit subsidies, and some lower-probability tail risks of tax revolts and total transit system collapse. But is the TOD opportunity space big enough to matter for the housing crisis?

If America is “short” somewhere between 4 million homes and 20 million metropolitan homes, the unused growth potential on the land near our rail systems has to be in the millions to be relevant. And it can’t just be any old land near any old train; it has to be where people are voting with their wallets to live: high-wage, high-amenity places where housing prices are above marginal housing supply costs (meaning that growth-control regulations are already the only thing stopping more people from building and living there).

The short answer is yes: physically and financially realistic TOD housing potential is a big deal, and is underrated as a reform opportunity on a surprisingly massive national scale. To be sure, one should treat this particular rail analysis as an order of magnitude rather than an exact count. But America’s bus networks are much vaster than our rail networks, and cover many more cities. Beyond this rail exercise, TOD in reality includes opportunities around buses and all the neighborhoods within walking and biking (or other micromobility) distance of downtowns. Without including those opportunities, this rail-only figure still yields strong-market capacity to build homes on a scale that is large relative to America’s short-term housing needs.

View Live Analysis & Maps

Finally, I want to push back on any hyper-literal or “overly engineering-brained” interpretations of these rule-of-thumb forecasts and planning heuristics for complex social and urban economic systems with endogenous or even self-regulating feedback loops that often do not behave like linear engineering problems. The elevation of rigidly simple traffic engineering and growth control planning rules of thumb, prioritized over social science and observable preferences revealed in market prices and behavior, created the very transportation and housing bottlenecks this analysis hopes to help solve. The point of this exercise is just to show that the mix of physically realistic growth opportunities currently banned by land use regulations, combined with revealed consumer preferences for transit-adjacent living observable in existing rents and home prices, together yield a reasonable expectation that relaxing the worst forms of central planning near our transit systems would plausibly yield voluntary actions on a large enough scale that is likely to solve our country’s most pressing housing undersupply and transit ridership sustainability problems.

Desirability and first-cut feasibility: Filtering physically realistic capacity from 31 million to financially plausible capacity for 13.5 million homes

America’s Census block-level capacity for allowing up to 40 homes per gross Census acre within a half mile of America’s rail systems is 31.3 million additional homes.9 This exercise thus answers one narrow question: How many more homes would these station areas contain at a half-mile-area average of 40 homes per Census land acre? First, some caveats:

• This is an illustrative gross-density gap representing the physical realism of development potential (it is not zoned capacity or a production forecast).

• A straight-line half mile is not always a half-mile walk.

• Some land should not or cannot be residentially developed.

• The calculation assigns the same benchmark to tidy residential blocks and to blocks dominated by office buildings or other nonresidential uses.

  • However: I excluded all Census blocks that now contain zero housing units or population, eliminating almost a quarter of the housing potential from this forecast to substantially account for airports, parks, and heavy industrial areas (see methodology appendix).

Even with those caveats, 31 million usefully establishes the physically realistic scale. It is larger than any conventional estimate of the national housing shortage. And remember, it’s still just a rail-only exercise: America’s bus networks cover much more land area than our urban rail networks, and high quality car-competitive bus networks can be scaled up in less than a decade.

The next question is where additional market-rate construction is at least plausible, financially. In much of America, especially outside high-wage metro areas, existing homes sell for less than it costs to replace them with new apartments. Upzoning those places is harmless, and could facilitate sparks of subsidized community development, but it will not unlock much construction in the current market.

I divided the station-area blocks into broad market tiers tied to my own rough estimates of inflection points in housing construction costs and first-cut market feasibility. These are binned using American Community Survey (ACS) median home values, with ACS contract rent as a fallback for blocks that didn’t have home value data.10 The “cool” category ($150k to $300k) approximates the roughly $250,000 “minimum profitable production cost” per single family home in Glaeser and Gyourko (2018).11 For the next bin, the “hard” construction cost to stack an extra 1,000 square foot home onto a Type I high-rise tower in NYC, America’s highest-hard-cost construction market, averages roughly $534,000, and San Francisco is the only other US city over $500,000, making $500,000 a relevant rough-cut threshold for the leading edge of high-rise multifamily construction feasibility. $800,000 is the threshold where even a $500,000 marginal hard cost of multifamily construction, with a soft cost multiplier around 30%, could together yield a ~20% gross margin to the builder on those hard and soft total marginal development costs.12 In other words, $800,000 per home is a plausible threshold for where incremental units added to (even high-rise Type I) skyscrapers in very difficult construction environments, at heights triggering all of the most hostile building code mandates beyond 7 stories, should be widely feasible if allowed cleanly by-right.

I label two tiers as high-demand:

• Hot: median home values from $500,000 to $800,000, or monthly rents from $2,360 to $3,780. These blocks account for 7.8 million homes in the gross-density gap.

• Red-hot: median values above $800,000, or rents above $3,780. These blocks account for another 5.7 million.

Together, the hot and red-hot blocks account for physically realistic capacity for 13.5 million additional homes under the 40-per-gross-acre benchmark in the highest-demand station areas where home prices are very likely to be above minimum profitable construction costs. This price screen still does not directly convert theoretical headroom into truly “market-feasible capacity.” I have complained at some length about planners making related mistakes. A serious soft-site analysis also needs parcel characteristics, the value and condition of existing structures, land assembly, local construction costs, financing, taxes, fees, and the actual density increment allowed by zoning. My screen tells us where demand conditions indicate zoning is most likely to bind. It does not identify the specific buildings likely to be demolished next year.

To be sure, one would also need to zone much more liberally than a 40 units-per-acre cap to achieve a realized production outcome averaging 40 units per acre within the next few business cycles. SB79’s zoning for 120 units per acre on high-quality rail sites (allowing seven-story buildings) is a good way to think about what should be allowed near rail, even in places where only the cheapest IRC rowhouses or three-story IBC Type V-B walkup apartments will pencil in weaker land markets.

Figure: State-by-state counts of current housing units and physically feasible capacity. Try other cuts of the data for yourself, and view every covered Census block on this live map.

What 40 units per gross acre actually looks like

“Forty units per acre” may sound more visually dramatic than it often is because density gets discussed through the most noticeable building rather than the land average around it. New York City offers a useful example. Among New York’s Census blocks near rail, I find the median block has about 29 homes per acre, and 62% of the blocks fall below 40. Yet the median home near rail sits on a block with about 63 homes per acre.

Most blocks are relatively low-density, but most people live in the smaller number of blocks containing apartment buildings.13 In many blocks, reaching the benchmark could mean one small or medium apartment building: a 24-home single-stair walkup, perhaps, or a 100-home mid-rise on a larger parcel.

Figure: Forty gross units per acre doesn’t entail Manhattanization on average, just rowhouses and the occasional apartment building. Even in New York City, 92% of residential structures are still three stories or fewer.

The largest gross-density gaps appear in the New York, Chicago, Philadelphia, Boston, Los Angeles, and Washington metropolitan areas. Commuter rail looms especially large: The station areas of Metra, SEPTA, NJ Transit, and the MBTA each contain more than two million homes of “gap” at the 40-per-gross-acre benchmark. Much of the opportunity is not in downtown skyscraper districts. It is along mature rail corridors where the public sector owns the expensive infrastructure and local zoning still treats an ordinary apartment building as an exotic industrial use.

A benchmark ridership scenario:

Having warned above about the dangers of taking mechanical “traffic engineering-brained” forecasts too seriously as a tool to justify coercively mandating or banning innocuous choices in housing and transportation policy, let’s indulge in one last mechanical forecast to estimate a lower bound of how many riders our 40 units per acre gross building envelope would yield.14

Figure: Calculated annual unlinked passenger trips from our land use exercise if new units are occupied at 2.2 people per home, with those people taking mass transit only as frequently as the average regional resident in earch rail system’s home region. 2024 has fewer trips per capita than 2019; calculations from both years are shown.

At 2024 transit ridership rates per person in each of the urbanized areas with rail systems, the nearly 6 million red-hot homes in our exercise would yield 1 billion new unlinked trips per year, increasing nationwide transit ridership by about 13 percent over current levels. Adding the nearly 8 million “hot” homes bumps that up to 30 percent. (There were roughly 9.9 billion unlinked transit trips on all modes in the US in 2019, and about 7.8 billion unlinked trips in 2024.)

The upshot: Physically and financially plausible rail-only TOD would be sufficient to completely reverse the national-level decline in transit ridership after COVID. If we expanded this analysis to high-demand blocks around frequent bus routes and neighborhoods walkable to job centers, the results would transcend mere pandemic recovery, and venture into outright reversal of America’s long-term declining relationship with mass transit. The benefits to economic growth, congestion, climate, land conservation, and the amenity and livability benefits of allowing most of America’s next 30 million homes to be built in the most extremely desirable neighborhoods in the country near mass transit (if people in the aggregate choose to do so voluntarily, without eminent domain) are immense.

Legalize a rider base

Will allowing homes near transit in high-demand places actually work? The closest thing American housing policy has to a natural experiment is Long Island City. LIC was rezoned to allow housing in 2001, with development taking off sharply after 2010 as LIC went from roughly 35,000 to 63,000 residents — 78% neighborhood-level growth in a decade, most of it in market-rate apartments a subway ride from Midtown. That is what clean transit-oriented upzoning does when the zoning constraint is actually released. Large reform case studies like NoMA in Washington, DC; Hudson Yards in Manhattan; Downtown Brooklyn; Gowanus, Brooklyn; and Austin’s University Neighborhood Overlay are all-star examples of rapid financially feasible growth commencing immediately after “clean” by-right upzoning.15 Multiply that fast-acting neighborhood level land use reform pattern across the hottest neighborhoods on America’s rail systems and we get 6 million potential red-hot homes.

Again, this is just a draft analysis looking at the development potential of rail systems only. Widening the analytical scope to alternative-transportation development opportunities beyond rail would include cheaply scalable bus rapid transit corridors, high-frequency local bus corridors, and any land areas within walking distance of downtowns but not yet served by a frequent fixed-route transit line. Someone should do that wider analysis but this rail-focused minimum viable product shows rail alone is big enough to matter. The rail analysis also conveniently centers the most above-cost large metro areas and the overpriced, undersupplied neighborhoods targeted by the Abundance and Growth fund (AGF)’s urban economics methodology.

This is not to say that the only place new housing should be legal is next to America’s trains: AGF supports broad upzoning to allow freedom of location choice. The goal of this analysis is only to demonstrate that TOD is not a marginal or boutique idea and should be a much bigger part of the near-term housing reform conversation.

Climate funders in particular should take note: Even if housing abundance is not an organization’s main goal, anyone who cares about climate has an opportunity to work with AGF grantees and the whole YIMBY ecosystem to prioritize land use reform wins on the scale of California’s SB79. That means cheaply decarbonizing transportation for millions of new homes simply by rescuing existing transit systems while reducing per-capita energy consumption all while increasing economic growth. American voters and volunteers are already solving the housing crisis with all-of-the-above solutions; climate funders now have the opportunity to make sure TOD is at the top of the housing reform conversation in every state with a big rail system.

America’s large rail systems are probably not at mortal risk from autonomous vehicles: The prewar urban core corridors of the biggest transit systems move too many simultaneous people for low-throughput, car-based vehicle geometries to fully shut down their transit systems. (Multiple parallel autonomous Bus Rapid Transit lines could combine enough corridor throughput to be able to replace most rail US systems, but autonomous cars alone cannot). Voters are more likely to vote for higher subsidies to keep the systems running even at lower utilization levels, unless we take this opportunity to solve both our housing and our transit problems. We already built the trains! Just let the riders live near them.16

Subscribe now

Methods appendix

To start, I overlaid 2020 Census blocks with the Bureau of Transportation Statistics’ National Transit Map. After consolidating duplicate platforms and station entrances, the data contain 5,672 rail stations and 152,105 populated or developed Census blocks whose centroids lie within a straight-line half mile of one. Those blocks already contain about 9.2 million homes.

Then I asked how many additional homes these blocks would contain if each reached 40 homes per full Census land acre. I chose 40 homes per gross acre because it is the floor of the top transit-supportive density tier in the TCQSM’s density-demand relationship, which is measured — conveniently for a Census-block analysis — in households per gross acre.17 I also excluded Census blocks that currently have zero households or residents in the Census data, reflecting the fact that many transit-adjacent Census blocks are airports, parks, heavy industry, or other places where residential redevelopment may be unlikely.18 Those excluded Census blocks are 2,041 water-only blocks covering 34,309 acres, and 46,026 completely vacant blocks with 267,406 acres of land. This is 24 percent of all Census blocks and 22% of all land near rail stations, enough to hold 10.7 million homes at 40 units per gross acre. This creates some risk of an underestimate: Converting unoccupied industrial land to residential land is the source of some of the nation’s most successful recent upzonings like Gowanus in NYC and NoMA in DC. But on the other hand, some of the land I did count includes commercial central business districts with small but nonzero resident populations, like Midtown Manhattan, where it’s not obviously likely that office conversions and new infill will reasonably be able to achieve 40 gross units per acre amid massive office towers in a land market where office usually remains the highest and best-yielding use. On balance, excluding 22% of all station-adjacent blocks for vacancy reasons gives me high confidence that this exercise is at least not an underestimate.

Finally, the gross versus net land area distinction is an important difference in what a “units per acre” number means. Forty homes on an acre of buildable residential land is approximately one attached home per 1,100 square feet. Forty homes per whole-block gross acre must fit the homes and the streets and other nonresidential uses on the same acre. Unless the street has become one of the homes, the latter requires smaller lots, apartment buildings, or a mix of the two. So 40 at the Census-block scale is more demanding than either 40 on residential parcels or Colorado’s 40-times-non-exempt-acre capacity goal.

Full replication files are available on request (and if there’s wider interest, we will edit this post and host them openly for public use, localization vetting, and extensions to further analysis suggested herein).

Predictably, there have been a lot of takes.

On the pro side, the arguments I find most compelling are: (a) Peer review has been buckling under its own weight for years, with editors even at top journals struggling to find reviewers. That problem gets much worse if AI drives a substantial increase in submissions, as it seems to be doing. Something’s got to give, and AI is likely to be part of the answer, so I would much rather it be via well-designed and publicly-validated tools than reviewers pasting manuscripts into a chat (note: we covered Refine’s own evaluation a few weeks back, and Paul Litvak has a new AI review eval that I will discuss more in a future post).

And (b) There are several subtasks that human reviewers have never been especially good at — like catching math or software errors, checking consistency between tables and text, and verifying that citations accurately represent cited work — but that we’ve fully relied on them for nonetheless. Offloading those subtasks could let reviewers concentrate on the tacit and taste-based components of review, like assessing the novelty of results, their impact on the field, how they fit (or don’t) with existing literature, etc. Peter Hull, Paul Novosad, and Kevin Bryan each proposed related versions of what the future of combined AI and human review could look like.

On the con side: (a) At equilibrium, there will need to be ways to avoid creating capture for for-profit companies – i.e. if journals pay subscription fees for AI review on one end, that heavily incentivizes authors planning to submit to those journals to pay for those services up front as well. That is less of an issue for this pilot, where Refine’s use does not affect publication decisions and is being offered to journals for free, but if it ever shifts upstream then this concern would need to be resolved.

And (b) There is also a possibility that offloading technical checks could make human reviewers less careful/thorough, which could reduce overall review quality on net. This is something Refine and the journals could formally study, and I hope they do.

Overall, props to Ben Golub for his work on Refine, and to the AEA and the Econometric Society – two of the publishers who have also been on the front lines of replication packages and data checks – for being forward-looking and experimental about research evaluation.

1. China tightens the reins on clinical trials — Dylan Matthews

2. A keyhole view of the next pandemic — Saloni Dattani

3. Europe’s scaling problem — Matt Clancy

4. PJM’s large load experiment — Willow Latham-Proenca

5. ROAD Act implementation comes to the states — Alex Armlovich

There will be more from Jordan later this week!

China tightens the reins on clinical trials — Dylan Matthews

One of the big biotech stories of the 2020s has been the rise of China. A recent survey of industry leaders found broad agreement that the US now lags behind China in clinical development of new drugs, and two years ago China oversaw more clinical trials than the US for the first time. This has led a lot of observers, like Congress’s National Security Commission on Emerging Biotechnology, to urge emulation of some of China’s regulatory approaches to drug development, like letting investigators initiate phase I trials without needing regulator approval first.

But Stat’s Brian Yang reports that China may be moving in the opposite direction, toward stricter regulation. Essentially, two deaths in gene therapy trials, and the bungled attempt by their investigators to cover up the deaths, pushed the Chinese government last October to advance Order 818. The order represents an across-the-board tightening of Chinese trial regulation. Now, investigator-initiated trials are restricted to the highest-ranked subset of hospitals, limited to certain types of drug designs, and required to meet “good manufacturing practices” standards akin to those the FDA requires of commercially available drugs in the US.

Yang finds that the rules may already be slowing down the famously fast-paced Chinese biotech center: “Since May, the China National Center for Biotechnology Development, an organization under the National Health Commission, has approved 80 or so [investigator-initiated] projects, but a STAT tally showed nearly all of them were retroactive registrations of work that had already been underway, and all of them are testing cell therapies. Only one is genuinely new.”

A keyhole view of the next pandemic — Saloni Dattani

I’ve been following the Ebola outbreak in central Africa, which has just surpassed 2,000 reported deaths and 4,381 confirmed cases. (The real numbers are likely higher due to underdetection.) What’s surprising about the Ebola outbreak is that we actually have an Ebola vaccine, but it was developed and tested against the Ebola Zaire virus during the 2014–2016 outbreak, while this outbreak is caused by Ebola Bundibugyo virus. The two viruses have similarities and some laboratory and animal data suggested there may be cross-protection, although the evidence is still unclear.1 But until last week, the WHO hadn’t decided to test that vaccine in this outbreak; the focus was on developing and testing new Bundibugyo-specific vaccines and antivirals instead.

In a recent STAT opinion piece, Krutika Kuppalli and Placide Mbala argue that pandemic preparedness shouldn’t be so narrow. Rather than focusing so narrowly on each virus, we should invest in broad-spectrum diagnostics, vaccines, and treatments, and should have tested the effects of the Zaire vaccine on Bundibugyo earlier; I agree.

Interestingly, the Ebola Zaire vaccine came about through innovative trials and coordination. It was developed through funding by BARDA and the DoD, tested using ring vaccination trials led by the WHO, incentivized through an advance market commitment, and stockpiled by Gavi, WHO, UNICEF and the Red Cross for future outbreaks. That stockpile has been useful in the years since then to contain outbreaks of Ebola Zaire quickly. But now that a larger, harder to contain outbreak has arrived, we’ve lost expertise and funding to tackle it, and we don’t know if the existing Ebola vaccine protects against this virus too. We could be doing a lot better.

Europe’s scaling problem — Matt Clancy

In 2008, the total market capitalization of US-listed firms was about a third larger than the market capitalization of European-listed firms. In 2023, it’s about triple. What happened? A new NBER working paper by Becker, Benmelech, and Monteiro tries to answer that, first by ruling out some potential explanations: it’s not a few lucky superstar firms; it’s not driven by differences in the relative size of different sectors; it’s not about how many firms get listed; it’s not about successful European firms re-listing in the states; and so on.

Instead, they argue Europe’s smaller market cap is downstream of the fact that successful European firms struggle to scale. Part of the explanation here is about financing. While the difference in the leverage of European and American firms did not change much between 2008 and 2023, the venture capital supply gap widened by a substantial level over the same time period. Since VC money supports new high-growth firms, this difference shows up most acutely there. While the largest European and American firms face similar costs of capital, the terms are much less favorable for small firms in Europe than in America.

The other problem is European firms have a harder time tapping into continent-wide markets than American firms do. The size of European firms is more closely related to the size of the national market they are based in, than the size of an American company is to the size of the state-wide market it is based in. The greater integration of American markets means it doesn’t much matter whether a company is based out of Idaho or Texas for its ultimate size, but it does matter in Europe if you’re based in, say, Germany versus Denmark.

PJM’s large load experiment — Willow Latham-Proenca

PJM – the country’s largest regional transmission organization by population – recently filed plans for two big steps on large load integration. As covered by Maeve Allsup in Latitude, the plan is two-part. First, PJM will try to procure enough new generation to cover the gap left at its latest capacity auction (where PJM procures enough power to ensure a reliability margin to cover unexpected peaks or outages) through a generator-load matchmaking process and an auction for the remaining shortfall. Second, PJM will require data centers that don’t bring or purchase their own supply to accept “curtailment” requirements – essentially, cutting the power they draw from the grid during emergency conditions.

PJM sidestepped the limits of its own authority (and the political fallout) by delegating – states and utilities will be responsible for implementing and enforcing curtailment, as well as the final cost allocation for generation capacity procured through the auction.

This is explicitly not a long-term solution – first of all, the capacity procurement is targeting a single year’s 7 GW capacity auction gap, a fraction of the 50-60 GW shortfall PJM itself expects over the next decade (others have widely varying projections here). Second, “matchmaking” pairs generators with the wealthiest suitors (probably the data centers), crowding out access for other types of new load. A long-term version of this type of program could undercut the viability of the capacity market itself, making it harder for cost-sensitive electricity users (like industrial loads) to come online.

ROAD Act implementation comes to the states — Alex Armlovich

IFP and Pew have a fresh state and local guide to ROAD to Housing implementation, complementing the NAAHL federal guide featured in our July posts. Will Poff-Webster, Alex Horowitz, and coauthors provide at least one sentence on every key provision in the 300+ page ROAD Act that requires or rewards action by states, localities, and tribes. The jurisdictions that move first on zoning, building codes, and permitting will capture the new federal dollars, and the ones that don’t will eventually lose some CDBG funding under the Build Now Act’s conditionality mechanism. The most immediate action item is manufactured housing: states need to certify to HUD that their laws treat chassis and chassis-free homes equally, which means most will need to amend statutory definitions that currently specify a permanent steel chassis. The piece also flags preapproved building plans, single-stair grants, and the HOME program overhaul that finally replaces per-unit cost caps with area-adjusted limits. Pair this with the NAAHL guide for a full introduction to what has to happen next.

Here are a few other highlights and announcements from our team and grantees:

• Adam Kroetsch joined IFP to help advance their work on FDA policy and clinical trial abundance.

• Our Coefficient Giving colleague Rafa Latham-Proenca published The Staggering Welfare Consequences of Catch-Up Growth After AI, estimating how much faster AI-enabled growth could benefit developing countries that adopt AI later, and finding the welfare gains are enormous even for latecomers.

• Center for Building, The Kelsey, and Sightline Institute were all quoted in a Washington Post piece on why the US doesn't have the small, affordable elevators common in Europe and China, a building code barrier that limits both accessibility and density.

• Michigan adopted the four-story single-stair rule statewide, building on Center for Building’s ICC win earlier this year. Jurisdictions with their own fire departments can go up to six stories.

1. Can beauty solve the housing crisis? - Alex Armlovich

2. Can cheaper inputs solve the housing crisis? - Alex Armlovich

3. How much of the housing crisis can we blame on endangered species? — Dylan Matthews

4. A new kind of problematic regulation — Willow Latham-Proenca

5. Cutting the Future STEM Workforce — Matt Clancy

6. It doesn’t take two trials — Saloni Dattani

7. What we talk about when we talk about institutional design — Jordan Dworkin

Can beauty solve the housing crisis? - Alex Armlovich

Samuel Hughes has a new Works in Progress essay asking again whether beauty can solve the housing crisis. He distinguishes NIMBYs (locals who oppose development of all kinds nearby) from NITBYs (”Not In That Backyard”): people who live nowhere near a project but casually oppose it because it looks ugly to them. Hughes argues that NIMBYism predates modernist architecture by a generation (first-wave restrictive zoning swept Europe and the US while new buildings were still classically ornamented), so beauty alone won’t dissolve local opposition. But the historical conservation movement, which froze European city centers and large swaths of American downtowns starting in the 1960s, could plausibly be a NITBY response to modernism: thousands of years of demolition-and-rebuild stopped almost overnight once the replacements got “ugly”. For infill development, it appears true that sometime after WWII, American city residents lost faith in the future of what their cities would look and feel like.

But competing stylized facts remain: cinderblock strip malls and vinyl-sided “mcmansion” sprawl continues to be the central feature of the most rapidly-building metros in the US (a sharp difference from the UK’s strict century-old ban on “ribbon development” sprawl). Beautiful projects like I’On in Charleston and the Duchy of Cornwall’s Faversham still get NIMBY’d to a standstill, and Robert AM Stern’s traditionalist neoclassical limestone revivals brought favorable reviews but few permitting wins. Finally: beyond the uncontroversial option of preapproved “pattern books”, nobody has an obvious policy lever to make new buildings beautiful at scale.

Can cheaper inputs solve the housing crisis? - Alex Armlovich

One of Brian Potter’s latest Construction Physics installments asks whether cheaper labor or materials can meaningfully cut housing costs: Looking at national average data in the US, he mostly answers no. Construction can’t chase cheap labor the way garment manufacturing did because most work happens on-site; modular builders that ship from low-cost locations (Idaho, even China) typically claim schedule certainty, not hard-cost savings.

From AGF’s perspective, however, it remains a puzzle how HUD Code manufactured homes have achieved huge hard cost savings over site-built homes, while modular construction has not done so. (Potter, too, has cited >50% lower costs per square foot for HUD Code manufactured homes). Some of the cost advantage for traditional HUD Code homes is related to entry-level finishing and materials: more upscale HUD Code CrossMods, built to middle class tastes and finishes, are “only” 20% to 30% cheaper per square foot than site-built homes. If multifamily modular could match the 20% savings of CrossMods, that would be a meaningful improvement. It may be unwise to expect miracles from multifamily modular policy or innovation, particularly in the national average data that covers wide swathes of the US where construction costs are already low, regardless of any code reform or innovation. But I also don’t think it’s crazy to explore whether some mix of code reform and innovation could help bring, say, NYC’s current Type I high-rise hard costs down closer to Chicago’s current Type I high-rise hard costs.

How much of the housing crisis can we blame on endangered species? — Dylan Matthews

You may have heard about the effects of the Endangered Species Act on big construction projects — like the famous case of the tiny “snail darter” fish and its role in delaying the Tellico Dam in Tennessee. But what about the effects on more everyday construction?

Harvard econ grad student Maxwell Tabarrok has a nice Journal of Public Economics article (official, free PDF, blog summary) estimating how housing permits change in areas that see a nearby species listed as endangered under the law. Listing of animal species leads to a large drop in permitting rates (0.5 permits per thousand residents, or 10 percent of an average area’s permitting rate); de-listing a previously endangered species leads to a similar-scale increase in permitting.

Tabarrok’s methodology is to treat each listing/delisting as a policy change and then do a “difference in differences” comparison of permitting rates in places that have the species nearby (the treatment group) and those that don’t (control group). This is about the best you can do in the absence of an actual experiment or some source of truly random variation in species listing, though it leaves some room to question whether the Endangered Species Act is causing this slowdown in permitting.

Tabarrok’s accompanying blog post suggests that the Trump administration’s new interpretation of the ESA could greatly weaken its effects on housing construction. Specifically, the Trump team is junking the current interpretation that includes damage to habitats as a form of harm to endangered species, and embracing a new interpretation where the law only applies to literally, directly killing or injuring animals. Under the now-final rule, set to take effect in September unless courts block it, the ESA would go from a law that bans construction that might harm endangered species to one that bans hunting and poaching endangered species. (The exception is projects on federal land or with a “federal nexus,” like the famous Tellico Dam.)

I like housing construction, but conserving endangered species is also valuable. Tabarrok has nicely illustrated the costs of the law, and I’d love to hear more about the benefits and how to balance the two optimally.

A new kind of problematic regulation — Willow Latham-Proenca

A forthcoming paper lends some structure to the intuition around regulated utilities (these are the vertically-integrated kind, with a local monopoly and their own generation facilities – they’re “regulated” because a state Public Utility Commissions, rather than the market, sets their rates). Gowrisankaran, Langer, and Reguant look to explain why coal has hung on so much longer in regulated markets than deregulated ones during the start of the natural gas boom – 26% of coal capacity in deregulated states was retired between 2006-2018, versus 17% in regulated areas.

They find that the standards Public Utility Commissions use to set rates are likely part of the explanation – to be included in a utility’s rate base (the pool of capital investments on which a utility can earn a rate of return), generation facilities must be “used and useful.” This creates a perverse incentive for utilities to keep expensive legacy facilities operating even if a cheaper option (in this case, natural gas) becomes available.

Interestingly, emissions don’t really factor into the decision over the long term; the authors find the cost advantage alone would be enough for a cost-minimizing utility to retire practically all coal generation over a 30-year horizon. However, the authors make an important caveat on reliability – reducing the rate base too far, without other compensation, could hamstring utilities’ ability to invest enough in the infrastructure that keeps the lights on.

Cutting the Future STEM Workforce — Matt Clancy

Last week, the Wall Street Journal reported that the Trump administration is considering implementing a $100,000 fee on international students who want to work in the United States after graduation via the optional practical training visa, one of the most common pathways to immigration to the United States. This policy, if implemented, would be the second recent blow to US legal high skilled immigration: earlier this month, the administration changed visa duration rules to add new burdens and uncertainty for international students. Previously, common student visas lasted for the duration of a course of study. Now the visas last up to a maximum of 4 years, with extensions possibly but uncertain and requiring application.

(The trouble is that four years is too short for a PhD. I did mine in five years, which put me on the fast end of my Economics PhD cohort, and the typical STEM PhD takes between 5 and 6 years.)

Collectively, these policies (along with the heightened uncertainty around funding for research) are likely to reduce high skilled immigration from international students. A new report from Amy Nice, Michael Clemons, and Jeremy Neufeld tries to quantify the size of the impact, looking at survey evidence and the historical share of the STEM workforce that follows affected pathways. They estimate that, if these policies had been in place for the twenty years between 2003 and 2023, the US STEM PhD workforce would be 11.5% smaller than it is today. That would mean lower US economic growth (and probably no significant benefits to native-born STEM workers; more here).

It doesn’t take two trials — Saloni Dattani

In June, the FDA issued new draft guidance stating that drug makers no longer needed to run two trials to demonstrate the efficacy of a new drug before it could be approved; rather, “one adequate and well-controlled clinical investigation with confirmatory evidence” could satisfy the evidentiary standard. It was a controversial move, although others pushed back saying the two trial guidance isn’t commonly followed in the first place and was an unnecessary burden.

In a recent post, the biostatistician Kaspar Rufibach shared a characteristically thoughtful take. When drug developers want to maximize their chances of succeeding in the second trial, they often try to design it identically. But, for a drug that “just” showed significance in the first trial, its chance of succeeding in a second, identical trial is just a coin toss: 50%. Better, he suggests, would be to run a single, larger trial, or run a second trial that’s different and gathers evidence for the drug’s effect on another disease, in another region, or a different route of administration, for example.

What we talk about when we talk about institutional design — Jordan Dworkin

If you’ve been following the conversation around the design of scientific institutions and funding strategies, you may have come across a table breaking down how different characteristics map onto different mechanisms.

Last week Hiya Jain, Stuart Buck, and Aishwarya Khanduja published “A Taxonomy of R&D Orgs”, an excellent longform contribution to this tradition that distills the multidimensional differences between research organizations into nine key parameters: (1) optimal project timeline, (2) level of user input, (3) size of discretionary research budgets, (4) career incentives, attracting talent, and personnel models, (5) revenue strategy, (6) degree of focus on a single mission, (7) size of organization and/or research teams, (8) governance and project selection mechanisms, and (9) openness and IP strategy.

They go deep on each, exploring the range of possibilities and laying out case studies of institutions that sit at different points on the spectrum. It serves as both taxonomy and history (e.g. I did not know that staff scientist/engineer roles in part emerged as a result of early ARPA block grants to CMU, MIT, and Stanford), and will be a useful resource as the scientific ecosystem tests out a wide range of new (and old) models.

Here are a couple of highlights from our grantees and team:

• 1Day Sooner is still looking for a clinical trial abundance policy director

• Saloni and Jacob Trefethen released a new episode of Hard Drugs on drug repurposing: Are miracle drugs hiding in plain sight?

• Dylan Matthews wrote about our chance to fix the Jones Act

• A paper about the Atlas of Innovation was published as an NBER Working Paper

Here in the US, at least one 2026 policy change has been a clear winner for economic growth. Starting on March 17, 2026, the Trump administration has waived the Jones Act, America’s cabotage law barring non-US built vessels from conducting trade between US ports, for certain oil, gas, and fertilizer shipments. The waiver was extended in April; it’s currently set to expire on August 16.

This move shouldn’t have been overly surprising. Kevin Hassett, Trump’s Council of Economic Advisers director in his first term and National Economic Council head today, is a vocal Jones Act detractor, and wrote in a 2021 book that Trump hates the law too (“This Jones Act, it doesn’t make any sense,” he quotes the president saying). Casey Mulligan, another economist who served during Trump I and is now chief economist at the Department of Health and Human Services, reported the same in his memoir You’re Hired!: “President Trump hates the Jones Act. It is the type of harmful regulation that he has succeeded in ending in health insurance, telecommunications, farming, and many other industries.”

Of course, Trump World contains multitudes and other advisers (notably Peter Navarro, Trump’s “senior counselor for trade and manufacturing”) are vocal Jones Act defenders. Mulligan and Hassett, in their books, recall being defeated by Navarro and allies in Congress in term one when they fought for a Jones Act waiver for liquid natural gas. But that a waiver actually went through this year, and has continued to be renewed, suggests the law’s critics are winning the intra-administration fight this time around.

This makes 2026 perhaps the single most promising moment for meaningful Jones Act reform in decades. Durable reform, though, has to go through Congress, and will have to be bipartisan. It will also have to grapple seriously with what transitioning to a new regime looks like.

The case against cabotage

Defenders of the Jones Act like to call it “America’s oldest law,” noting that rules limiting which ships could use US ports are as old as the first Congress. But the “Jones Act” generally refers to section 27 of the Merchant Marine Act of 1920, a provision authored by a Seattle shipping lobbyist and added to the law by Washington Senator Wesley Jones. This is still the most important single statute regulating shipments between US ports.

As it exists today, the law requires that ships be…

• Captained by a US citizen, with all US citizen officers, and US citizens making up at least 75 percent of the rest of the crew

• If owned by a corporation, owned by a corporation with at least 75 percent US citizen ownership

• Assembled in the US, with the “major components” of their hull, main deck, and other “superstructure” all built in the US

Cabotage laws like this are common internationally, but they often only restrict trade to vessels flagged in the country in question, or owned by nationals of that country, or crewed by staff in that country. The Jones Act is unusual in requiring that ships moving between US ports be built in the US. Most other countries with large cabotage trade, like China, Japan, and India, don’t require domestically-built vessels, despite the former two having massive shipbuilding industries much more successful than that of the US. As a result, only the 92 Jones Act-compliant US-built ships currently in operation can move between US ports when the Act is in effect.

The economic case against the Jones Act is simple: there are many useful trips to be taken by ships within the United States; the Jones Act, plus the tiny size of the Jones-compliant fleet, means many trips are not taken when the Act is in operation; allowing those trips to take place represents new, beneficial economic activity that we’re currently blocking.

Perhaps the best examples of these Jones-prevented trips are the ones happening right now, under the waiver. The Cato Institute has been tracking waiver-enabled voyages, and there have been 195 to date as of this writing, on 161 different vessels, shipping some 50 million barrels of various fuels and fertilizers (most of the volume is gasoline and crude oil). Those 161 vessels represent a more-than-doubling of the US cabotage fleet, relative to the Jones Act’s 93:

As you can see from their maps, these journeys typically originate in the Gulf Coast, the center of US oil refining and export, and then make their way either to the northeast US, to Puerto Rico, or to the West Coast via the Panama Canal.

If you’re not especially interested in boosting oil shipments for climate reasons, don’t worry: the Jones Act is harming your priorities as well. A major effect of the Jones Act is to transfer goods from freight ships to freight trucks, which produce vastly more carbon emissions per ton-mile traveled than ships.

The Jones Act is also a major factor behind the stagnation of the US offshore wind industry. Installing a wind turbine in the ocean requires specialized ships called wind turbine installation vessels (WTIVs). Until March of this year, there were literally zero WTIVs built in the US. That meant that installation required sailing WTIVs from ports in Canada to as far as Virginia with turbine components, because foreign-built ships couldn’t be based in the US and do US port-to-port hops. As of March, there’s exactly one Jones Act-compliant WTIV, first installing turbines years behind schedule and far over budget, and we need at least four WTIVs to meet the Biden administration’s old offshore wind targets.

More formal modeling gives us some sense of the magnitude of the economic harms from the Jones Act. A 2019 paper from the OECD’s Karin Gourdon and Joaquim J. M. Guilhoto conducted a simulation estimating the economic costs of the Jones Act and an analogous domestic-content rule in Brazil. They suggest that total US economic output will grow by between 0.1 and 0.4 percent annually. That’s between $31.8 billion and $127.4 billion a year, per the most recent GDP numbers.

Here at Coefficient Giving, we often evaluate grant opportunities by estimating their impact in “Coefficient Giving dollars,” which we define as “the amount of welfare generated by a $1 gift to someone with an annual income of $50k,” roughly the US GDP per-capita when we started this framework. More precisely:

CG $v = $50k * w * ln(1+z%) * y.

Where v is the value of an intervention, w is the number of people affected, z is the percentage change in the income of the people affected, and y is the number of years they’re affected. The math is less important than the overall idea: we place a great deal of value on increasing people’s incomes, with longer-lasting changes affecting more people being worth more.

Jones Act repeal looks very, very good under this framework. Suppose we take the low-end figure from the OECD researchers (0.1 percent added to annual output); that’s z. Imagine for the sake of argument that we have a grant opportunity that has a 1 percent chance of leading to the US repealing the Jones Act ten years earlier (y = 10) than would otherwise happen. The US has 340 million people or so (w = 340 million). Do the math and you get about CG$170 billion in impact. Right now, our Global Health and Wellbeing teams have a 1000x bar for grants: they have to produce CG$1000 for every US$1 we spend. That implies we’d spend $1.7 million to generate a 1 percent chance that the Jones Act gets repealed ten years early, before even considering other factors like environmental benefits.

Obviously, no one can guarantee an advocacy campaign will even be that effective, and the nature of advocacy is that most people you fund have a low, sometimes under a percentage point odds of success. But when the value is this high, it’s still worth funding longshots. And with a Jones Act waiver currently in effect, serious reform looks like less of a longshot.

We need bipartisanship and creative thinking

The battle lines around the Jones Act are very familiar at this point. On one side are the US ship-building companies and allied unions, organized as the American Maritime Partnership, which push very hard against any weakening of the law. On the other side are anti-protectionist groups like Americans for Prosperity and the Cato Institute, the latter of which has done yeoman’s work documenting the law’s harms.

My sympathies are more with the latter, but the current ecosystem could use shaking up. For one thing, there are strong reasons for Democrats, both moderate and left-wing, to oppose the Jones Act. Behind its huge environmental costs, it particularly raises costs for consumers in Hawaii and Puerto Rico, which are entirely dependent on shipping for most goods but are forced to use the small Jones Act fleet for shipments from the US mainland.

As a result, there’s a reasonably large group of Democratic Jones Act critics in Congress, largely consisting of members with Puerto Rican ancestry. Alexandria Ocasio-Cortez (D-NY) is a vocal critic, having tweeted even before she entered Congress, “Congress must repeal the law that cements PR’s status as 2nd class citizens.” Nellie Pou (D-NJ) repeatedly urged repeal during her time in New Jersey’s state legislature. Ritchie Torres (D-NY) has proposed legislation exempting Hawaii, Alaska, Puerto Rico, and Guam from Jones Act requirements, just as the US Virgin Islands is currently exempt. While most of Hawaii’s Congressional delegation has sided with the shipbuilders over consumers and backed the Jones Act, Ed Case (D-HI) has been a welcome exception in supporting the waiver this year and urging permanent reform.

Conversely, there are Republicans like current House Speaker Mike Johnson who are vocal fans of the act. The Jones Act isn’t an obviously partisan issue, which is good: it means there could be bipartisan support for reforms, which is helpful even under unified government — but will be essential if, as expected, Democrats take the House this fall.

Realistically, any progress will involve real compromise. There aren’t the votes for an outright repeal of the Jones Act, and there are not likely to be the votes any time soon. There might be votes, though, for smaller measures:

• An exemption for Puerto Rico, Hawaii, Alaska, Guam, and Northeast states reliant on heating oil, or perhaps just one of that set

• Repealing the build requirement of the act but not the crewing requirements, so that labor unions representing US seafarers continue to benefit, and perhaps even gain as the new foreign-built ships hire their members

• Pairing adjustments to the act with buyout offers to current US-based shipbuilders, so those affected have resources to choose a new career path and the transition is truly win-win

• Pairing changes to the act with subsidies to US-based shipbuilders akin to those in the SHIPS for America Act, so US firms face more competition but also have more resources with which to face it

None of these are my first choice of policy regime, and I don’t suspect any of them are the first choice for any player in this debate. But the current Jones Act waiver makes compromises like these uniquely possible. Will pro-Jones Act members of Congress choose any of these over an in-effect Jones Act? Of course not. Might they be willing to budge if it means an indefinite Jones Act waiver comes to an end and they get at least some of the law back? Potentially.

Right now there simply aren’t many groups trying to work through what a good compromise could look like. If you are a think tank, advocacy group, or independent policy researcher interested in this area and crafting a compromise that reduces the Jones Act’s harms while providing some kind of win to its promoters, please get in touch. The waiver means the time for action is now.

1. Ignoring the price signal - Willow Latham-Proenca

2. The Hostile Turn in Administrative Procedure, and the Ad Hoc Exceptions — Dylan Matthews

3. Actually Existing State Capacity Reform — Matt Clancy

4. The Two McNamaras - Saloni Dattani

5. The Effects of U.S. Public R&D on Global Growth — Dylan Matthews

And Alex, Jordan, and Nisha are all traveling this week.

Ignoring the price signal — Willow Latham-Proenca

How much do prices affect how much electricity we use? It’s an increasingly relevant question, since changing when people and firms use electricity – usually to try to spread out short-term peaks – is often one of the first proposed responses to the current energy crunch. It would also seem to have an obvious answer – if electricity prices go up, from an Econ 101 perspective, use should go down. Some go so far as to say that exposing customers (and their devices) to real-time prices should even be enough to organize meaningful demand response, without virtual power plants or other middlemen.

However, it turns out people really like electricity. A new paper this month finds that prices matter less than you’d think – Kudela and coauthors find that price increases do little to change demand (a 10% rise in the electricity price cuts consumption by under 2% on average, over the short term). Even more surprisingly, the most careful studies reviewed find prices have next to no effect at all on electricity use. They also find that elasticity is higher over the long run than the short term – consistent with users adapting to higher prices through capital improvements, rather than behavioral change (think a more efficient dryer, not waiting until midnight to dry your clothes).

The findings here are not just surprising but also policy-relevant, implying both that price transparency won’t do much to smooth demand without paired technology improvements, and that firm capacity (generation that can be called on at times of peak demand) will likely continue to be needed, especially over the short term. To be clear, this isn’t an argument against peak-shifting as part of a larger energy abundance strategy (particularly when intra-day price-sensitivity can be programmed into automated devices), but it is a reality check on how most people think about and use energy.

Speaking of paired technology improvements, Massachusetts launched a residential vehicle-to-grid pilot program this month – exactly the sort of demand-smoothing technology the Kudela paper implies we’d need to actually get electricity demand to follow supply.

The Hostile Turn in Administrative Procedure, and the Ad Hoc Exceptions — Dylan Matthews

“Abundance” as a movement has a set of diagnoses of what’s wrong with US public policy, and how to fix it, but it also has a historical narrative behind it, which goes something like this: the US used to be able to build stuff, fast. Starting in the 1970s, we lost our ability to build stuff. Zoning got tighter, permitting energy projects got harder, red tape got thicker all around.

Alex Mechanick, a professor at Johns Hopkins and veteran of the Biden regulatory shop, calls this shift “the hostile turn,” and has a useful new post outlining what it actually consisted of. Drawing on Paul Sabin’s book Public Citizens (for my money, the single most important text on American abundance politics) he argues that the slew of new substantive rules in the ‘70s did not, by itself, lead to this sclerosis. The Clean Air Act and the Vehicle Safety Act did not, on their own, make it impossible to build. The hostile turn, instead, consisted in a set of procedural obstacles, sometimes set up in substantive laws like the National Historical Preservation Act, and sometimes in purely procedural measures like the National Environmental Policy Act (NEPA), “statutes that tied the hands of agencies through more and more onerous procedures, rather than a focus on regulating on the basis of scientific and economic expertise.”

That much of the story is probably familiar; NEPA in particular is a common target of criticism from abundance folks. But for Mechanick, the lesson is that procedural problems demand a procedural solution, and that begins by rethinking and revamping the Administrative Procedures Act (APA), the so-called “constitution” of administrative law in the US. This is a move that most commentators don’t make, and Mechanick takes it to some interesting places.

Actually Existing State Capacity Reform — Matt Clancy

Compared to some of the other areas we work on, improving “state capacity” can seem a bit nebulous. The end goal of improving the productivity of government is laudable, but what, specifically, does that mean? With that question in mind, I enjoyed Jen Pahlka’s rundown of North Carolina’s recently passed Public Modernization Workforce Act.

One concrete way of making government more productive is via civil service reform. Prior to the new bill, finding new job opportunities in the North Carolina government was byzantine, giving calibrated feedback to workers wasn’t happening, retaining high performers with pay increases was challenging, and letting go of poor performers was extremely protracted and difficult. The Public Modernization Workforce Act streamlines much of this process, and introduces a lot of new innovations to help recruit, retain, and effectively manage civil service talent: a track to more easily make temporary workers permanent, a pilot testing upward feedback (where reports rate the performance of their managers), and setting in motion a review of the underlying classification and compensation system. As Pahlka notes (and we agree), hopefully the Federal Government and other state governments can learn from North Carolina’s experiment.

The Two McNamaras - Saloni Dattani

I enjoyed reading Oliver Kim’s article on the two lives of Robert McNamara: one as US secretary of defense and chief escalator of the Vietnam War, the other as World Bank president and crusader against global poverty. He suggests that though it’s tempting to contrast these as different sides of the same man, they shared common threads: a drive for quantification, a bureaucratic command approach, and the ambition to reorganize large institutions. In Vietnam, he used the Hamlet Evaluation System, which graded 12,000 hamlets each month and fed the results into an IBM mainframe that invariably showed the US winning. At the Bank, he more than tripled the staff and turned an institution run by engineers into one run by economists; yearly lending grew from about $1 billion in 1968 to $13 billion in 1981.

In both eras, he seemed to focus more on what was easily measurable rather than strategically relevant. The North Vietnamese were willing to absorb losses that American planners had not anticipated, and land reform was not considered relevant until much later. The Bank did some good, virtually eradicating river blindness in West Africa, for example, but also drew criticism, with a culture of lending with little regard for how the money would be used.

Archives suggest that McNamara privately believed by November 1965 that the Vietnam War could not be won, but continued selling it publicly for years, boasting enemy body counts. Asked later why he stayed silent, he offered the excuse that his loyalty lay with the president. Oliver concludes, “if there is a thread we can trace through his career, from the Pentagon to the Bank, it is that no amount of technocratic skill can substitute for ethical judgment. McNamara was one of the 20th century’s great systems builders—a man who could tame vast bureaucracies, enlarge them, rationalize them. That doing what was right might require sometimes stepping outside the system simply did not compute.”

The Effects of U.S. Public R&D on Global Growth — Dylan Matthews

Last week, I mentioned a big Congressional Budget Office report estimating the economic impact of federal spending on non-defense research and development efforts (think basic and applied science at the National Science Foundation or National Institutes of Health, more engineering-like work at ARPA-H or ARPA-E, etc.). That report leaned heavily on the work of economists Andrew Fieldhouse and Karel Mertens, whose recent American Economic Review paper constitutes some of the best evidence we have on the economic returns to R&D spending in the US.

Now, Fieldhouse and Mertens have teamed up with three other researchers (Gustavo De Souza, Ishan Nath, and Valerie Ramey) to extend their work and estimate the global impact of US R&D spending. It stands to reason that science and engineering in the US should have effects abroad. When Bell Labs researchers invented the transistor, it didn’t take long for that to be powering radios in Japan. The team finds that the US captures only about half the returns to US-based R&D, with the other half benefiting the rest of the world. The US-only returns are already enormous, and well worth it even if you only care about US well-being. But if, like us at Coefficient, you try to look at well-being worldwide, this finding suggests US R&D is an even better deal than we thought.

One more thing

Finally, in case you missed it, Jordan published a new website last week, Ordinary Abundance, which commenter Zac Hill described as “almost the inverse of a memento mori - a reminder of the sheer infinitude of life, its almost banal fecundity of possibility.” Read Jordan’s blog post about it here.

In our work, we focus on the former, trying to improve the lives of people around the world by making housing more available, energy more abundant, and health more secure. To date, we have not funded the Institute for Lower Expectations (though they keep asking). That focus will continue. But to fully capture the value of improving reality, it is important to step off the hedonic treadmill every now and again, take a breath, and reflect on how far we’ve come.

There’s a bit of this thinking in the air these days. My colleague Matt Clancy wrote a piece last year exploring how new technology improves his own life; Charles Mann published a six-part series in The New Atlantis showcasing the complex, awe-inspiring societal infrastructure that many of us take for granted; he and Virginia Postrel also recently started a podcast, called Everyday Abundance, with a similar goal; and Jerusalem Demsas and Kelsey Piper wrote about all of the seemingly small joys of modern life that past generations couldn’t dream of.

In that piece, Kelsey writes, “I try — once a year or so, perhaps around Thanksgiving — to surround myself with the thoughts of all the people who have ever lived. Almost every single human who has ever lived was very poor. With them accompanying me, I walk through my house and try to see it through their eyes.” While she was drafting those words, I was working on a project to help people do exactly that.

The limit of human felicity

In their Everyday Abundance episode on music, Charles and Virginia highlight this quote from an 1888 novel by Edward Bellamy:

[I]f we could have devised an arrangement for providing everybody with music in their homes, perfect in quality, unlimited in quantity, suited to every mood, and beginning and ceasing at will, we should have considered the limit of human felicity already attained.

I find this sentiment breathtaking. Less than 150 years ago, “accessible music in the home” was a dream that some felt would represent the absolute pinnacle of human happiness; for most of my adult life, I have been able to listen to much of humanity’s discography whenever I like, and, without necessarily realizing it, I’ve used that privilege to make cherished memories, deepen relationships, and better understand myself.

Day to day, I find it difficult to conjure that sense of awe and appreciation. But Bellamy’s quote got to me, and I wondered whether similar historical expressions of hope, relief, and joy for the things we now find commonplace might help more people access those feelings themselves.

My attempt to capture those sentiments is ordinaryabundance.com.

The core of the site is a tour through past perspectives on our modern amenities. I loved hunting down these quotes, so if you know of any others that deserve to be highlighted, please do reach out (or comment here). I hope you take a look, and if you crack a smile next time you do a load of laundry I’ll call it a win.

For now, I’ll leave you with the closing paragraphs of the site, and what I personally took away from this exercise:

It is to humanity’s credit that we remain restless in the midst of all of this progress. We continue to look forward, pushing the frontier further with new treatments, new tools, and new institutions that will help future generations in ways we can’t even picture yet.

But our lives today are a gallery of past generations’ heroic efforts to do the same. It serves us, and honors them, to recapture whenever possible the old sense of awe at these wonders that have long since become commonplace.

Prediction markets are already ethically controversial because of concerns about insider trading and market manipulation. So Kalshi says that, for now, it will limit the markets to Phase 3 trials after they’ve been fully recruited, and will include some guardrails requiring employment verification to bar insiders like lead investigators, biostatisticians, safety-monitoring committee members, the drugmaker’s own staff, and trial participants from taking part.

But I wonder whether, or how, these markets would be useful. Anne Wojcicki, who runs the 23andMe Research Institute, says the idea is ‘very patient-empowering’ and that they could help people follow trials relevant to their disease, but I think the case is pretty weak: Kalshi says their markets would open only once a trial has finished enrolling, so they couldn’t influence a patient’s decision about whether to enrol, at least in that study; rather, the main change a patient could make based on the market would be dropping out. If that happened, it would make it even harder for trials to retain enough statistical power to reach their endpoints.

That kind of self-fulfilling outcome seems like it would have even more downside in an industry that’s already very conservative and finds recruitment challenging enough. (Maybe some doctors would ask for access to promising experimental drugs for their patients through ‘compassionate use’ based on prediction market signals, but in practice that depends on the drugmaker’s willingness to supply the drug before it’s approved, which is usually pretty limited.)

Still, prediction markets might be more useful to external researchers, investors, and policymakers, in shaping follow-on research or funding around them. They might also help surface suspected fraud earlier, the way some short sellers have done. But, because they would make the stakes more directly linked to each trial, I’d guess they’ll also raise the risk of it happening in the first place. (See, for example, a recent case of statistical manipulation in a high-stakes phase 3 trial that was only uncovered after the drug was approved!)

It’s probably going to become more important to maintain the integrity of clinical trials, with stronger guardrails in blinding researchers and patients in the trial, strengthening regulatory independence, and maybe also requiring researchers to pre-register the full statistical analysis before the trial begins.

1. Summer housing roundup and the end of parking minimums in NC — Alex Armlovich

2. Can you pay off the NIMBYs? — Nisha Austin

3. What’s a dollar of R&D worth? — Dylan Matthews

4. What's new in the economics of innovation — Jordan Dworkin

5. Advanced market clinical trial design commitments — Saloni Dattani

6. Sobering takes on Canadian wildfires — Willow Latham-Proenca

Stay tuned for a What we’re reading spotlight from Saloni tomorrow. And Matt is on holiday this week.

Summer housing roundup and the end of parking minimums in NC — Alex Armlovich

Halina Bennet has a summer housing roundup at Slow Boring: Beyond the already-covered ROAD Act, California's Governor Newsom signed AB 179, replacing the state's notoriously fragmented affordable-housing financing with a single application and penalizing localities that charge development impact fees on state-funded projects (saving $60,000–$70,000 per home). San Francisco slashed its unfunded inclusionary housing mandate from 15% to 5% after a court-ordered feasibility study found that even 5% could render projects infeasible under current costs and rates. Massachusetts folded a "YIGBY" (Yes in God’s Backyard) provision into its economic development bill, letting religious institutions build multifamily by right on land they've owned for three years. Finally, Zillow's latest census analysis, estimating a “national housing deficit” by comparing the number of doubled-up families to the available housing supply, held essentially flat in 2024 for the first time since the post-2008 underbuilding era began, driven by a 50-year high in multifamily completions. Even these crude accounting-based proxies for housing underproduction (which do not directly consider relative prices or inter-metropolitan mobility) nonetheless find the 4.7-million-unit deficit remains concentrated in the usual suspects: New York, LA, Boston, SF, DC.

Separately, North Carolina just became the first state to eliminate nearly all mandated parking minimums statewide. Governor Stein signed HB 162 (the Parking Lot Reform and Modernization Act) taking effect January 1, 2027. The coalition story is remarkable for most legislation but typical of the greatest YIMBY wins: the bill united environmentalists, farmers, developers, and business leaders. Reason notes the liberalization covers commercial uses too, not just housing, which again is unprecedented in statewide parking reform. The bill started as a stormwater management bill and the bipartisan framing (focused on stormwater regulatory relief first, and urbanism second) is what got it through a Republican-controlled legislature. Whatever the ideological context and framing, it remains fascinating to see parking relief laws begin to leap forward in car-heavy red states while trundling along incrementally in transit-rich blue states.

Can you pay off the NIMBYs? — Nisha Austin

A forthcoming paper by Michael Hankinson, Edward Glaeser, Joseph Gyourko, and Morris Davis surveyed roughly 1,700 residents of expensive areas and found that most people do have a price. Over 80 percent accepted at least one hypothetical housing proposal when offered cash compensation. The amount depended heavily on density: residents in dense urban areas needed only a few hundred dollars, moderate-density neighborhoods required low thousands, and low-density suburbs needed well above $5,000. Proximity and project type mattered too: proposals next door doubled the price, higher-density projects quadrupled it, and low-income-designated housing multiplied it fivefold. I was surprised to learn that when compensation was offered as a parks-and-streets fund rather than direct cash, residents demanded ten times more. As Rogé Karma writes in the Atlantic, this has prompted several think tanks to propose direct payments to residents in communities that hit housing production targets. What we'd love to see is similar research on infrastructure (in particular, transmission lines, wind farms, and industrial facilities). The evidence base there is almost nonexistent, even though siting opposition is just as central to those fights.

What’s a dollar of R&D worth? — Dylan Matthews

Here at AGF, we tend to think investment in research and development, whether by governments through agencies like the National Institutes of Health or in private businesses and universities, is worth an awful lot. When you get down to it, economic growth is about getting more benefit out of the same amount of physical stuff; you can only extract more benefit from the same stuff if you have better technology, and you get better technology through scientific research.

But just how beneficial is R&D? A big new report out of the Congressional Budget Office, coauthored by economists Sheila Campbell, Jaeger Nelson, Eli Schrag, Heidi Williams, and Caleb Wroblewski, has some of the most careful analysis of this question to date, using two radically different methodologies that come to quite different numbers. One approach results in an estimate that each federal dollar spent on R&D, financed by an increase in the budget deficit, grows GDP by $5.15. An alternative framework leads to an estimate of $1.98; the numbers in each case are higher if you pay for the funding through spending cuts, rather than borrowing it. Even the smaller number implies that R&D spending is an incredible deal, with each dollar invested doubling its value.

What’s new in the economics of innovation — Jordan Dworkin

This week I’m at the NBER Summer Institute, taking in lots of new work on the economics of science and innovation. Two of the papers being presented have already made appearances on this blog (one on idea diffusion, and one on the impact of AlphaFold). The rest of the agenda is also worth a look. Chen, Liu, & Ma take a new approach to estimating the economic value of science, building a text-based innovation network from 463 million papers and 81 million patents to trace how knowledge flows into technology; Songyuan Teng decomposes pharmaceutical innovation into novelty (new chemical building blocks) and recombination (new assemblies of existing blocks) and models their relative importance; and Mohnen & Gans have a new study on AlphaFold showing that it sped up experimental research but narrowed its scope (echoing other findings on contraction in AI-driven science).

Advanced market clinical trial design commitments — Saloni Dattani

I enjoyed reading a post by Witold Więcek on the Clinical Trials Abundance blog, titled ‘Market shaping and trial design belong together’. He explains why people interested in solving market shaping problems – such as economists designing advance market commitments and philanthropists funding multiple R&D projects for global health problems – should be interested in innovative clinical trials.

Most drugs today are tested in separate clinical trials: one drug vs one placebo, which is much more inefficient than running them together in a platform trial (with, for example, five drugs vs one placebo). That structure also makes it hard to make head-to-head comparisons between different drugs, since separate trials might enrol different kinds of patients, run at different times, and measure different outcomes.

That means the standard way to run clinical trials is less useful for decision-making. A market shaper, for example, might want to specify some criteria that different drugs have to achieve in order to receive a prize or payout from an advance market commitment. But differences in trial design and setting would leave open a lot of uncertainty about which ones qualify. So, he argues, why not have them play a role in designing the trial as well?

Sobering takes on Canadian wildfires — Willow Latham-Proenca

If you were in the Northeast US last week, you probably got at least a noseful of Canadian wildfire smoke. We’ve written before about the extraordinarily high health costs of wildfires, but Zeke Hausfather makes a convincing case that Canada’s fires – unlike many in the U.S. West – aren’t mainly driven by institutional mismanagement. Unlike in California, these Canadian fires are mostly occurring in boreal forests that have never been managed, so there’s no unnaturally accumulated detritus to burn. Devastating crown fires driven by weather (heat waves + lightning strikes) are the norm in these environments - so higher average temperatures directly drive more burning without additional human intervention.

Robinson Meyer makes a related case in Heatmap, arguing that the failure of climate research to integrate the potentially massive health effects of wildfire smoke is a potent illustration of how big the unknown-unknowns of climate change could be. While there was some pushback about the deep history of wildfire research, the fact that the resulting health effects weren’t integrated into flagship models of economic damages stands – as a result, policymakers had very little basis to understand and incorporate smoke risks before these types of fires became a regular occurrence. Part of this might be a map of science problem – wildfire risk, epidemiology (famously tricky for events like this), and mainstream climate modeling just didn’t overlap enough to stimulate research on their intersection until facts on the ground made the gap obvious. It’s worth wondering whether this is a problem that AI will help with, or one that the jagged frontier of scientific research will exacerbate.

Here are a couple highlights from our grantees:

• The Inclusive Abundance Institute published “The Biggest Housing Bill in 30 Years Is Now Law,” a look at what the ROAD Act means for the abundance movement.

• YIMBY Melbourne launched the Australian Zoning Atlas, mapping planning controls across all major Australian cities.