When my friend’s wife had a stroke, he used AI to check her new medications. Years before, she had gastric bypass surgery. He wanted to know if the surgery would change how the drugs worked.

It did. One of the main drugs is absorbed in the part of the intestine that her surgery had bypassed. He flagged it for her doctors. They adjusted the plan. AI may have saved her life.

My friend is a scientist. Given time, he could have found this himself. But in that emergency, there was no time. AI gave him a fast second opinion.

This kind of fast, informal help usually comes down to who you know. You get it if you can do the research yourself, or if you have someone to call.

Growing up working class, we felt lucky to know a doctor, a nurse, or a lawyer. They were the people you called when the system got confusing and you couldn’t afford to pay someone to explain it. Is my landlord required to give me hot water? Can I sue to get my deposit back? If you knew the right person, you got a straight answer.

In high school, I turned down AP courses because I couldn’t pay the test fees. No one told me waivers existed. Years later, I stumbled onto a list of colleges that cover full tuition for low-income students. It changed my life. A college counselor might have handed me that list if I’d had one.

AI gives that same kind of information to people who don’t know the right person. Even though AI makes mistakes and isn’t fully reliable, it still helps people do better research in their language while they wait weeks for an appointment. It can help someone put their symptoms in context instead of panicking. These benefits don’t show up in GDP or in headlines.

That is why a new bill in Albany worries me.

A well-meaning bill

New York Senate Bill 7263 tries to stop AI from posing as a licensed professional, which makes sense to me. If a chatbot pretends to be your doctor and hands you bad advice, you should be able to sue.

But the bill is written so broadly that it would punish almost any AI that gives useful information regarding a “licensed field”.

The bill makes anyone who "owns, operates, or deploys" a chatbot liable if the bot gives a "substantive" answer regarding medicine, law, nursing, veterinary care, pharmacy, social work, and mental health counseling, among others. A disclaimer like "I am not a doctor" does not protect the chatbot’s deployer.1

The biggest problem is that "substantive response" is not clearly defined. ChatGPT gave my friend information about how his wife's medications would interact. Is that substantive? Does substantive mean “could be helpful”?

Because the term is so vague, an AI company or deployer can only protect itself by refusing to give useful answers. The bill offers no other path. Lawyers call this strict liability. You pay when something goes wrong, no matter how careful you were.

It does not matter whether the answer was accurate, or whether the bot listed every risk and warned that it might be wrong. The liability is the same. A New York law firm warns that operators will simply shut off whole categories of answers, since no disclaimer removes the risk. The only safe move is to say nothing that looks substantive.

Who this hurts

We should want AI to give people context and useful information. Many people have no steady access to a lawyer or a doctor.

In New York in 2010, 98 percent of tenants in eviction cases and 95 percent of parents in child support cases went to court with no lawyer. For many of them, the choice was not AI versus a professional. It was AI versus a Google search, or nothing.

I want AI to help a tenant who can’t afford a lawyer understand a confusing eviction notice.

The bill also turns lawsuits into a business. If a violation is “willful,” the operator has to pay the plaintiff’s legal fees. Shifting the fee can be a good thing, but it also rewards lawyers for filing lots of suits and winning only a few. Even when the harm is small or unclear, most AI companies or deployers will settle, because settling is cheaper than going to court. A lawyer could build a bot whose only job is to quiz every chatbot in the state and flag any “substantive” reply.

Big companies can probably absorb this. I worry about the smaller groups. A small nonprofit that wants to deploy a chatbot trained on local domestic violence resources so a survivor can get answers at 2 a.m. while she waits for a caseworker. Or someone with long COVID who builds a custom GPT trained on the latest research and shares it for free. Or a nonprofit law firm that wants to test a bot trained on visa options for Spanish-speaking immigrants.

None of these makes money. These deployers would be taking on legal risk they can’t afford.

What a better bill would do

I want people to be free to build better tools for each other. A good bill would target only operators who claim to be licensed or fake a credential. If it goes further into “substantive” answers, it should at least protect information that is accurate or accompanied by context, rather than letting anyone sue any operator who tries to help.

As AI gets better, I expect some of the lawsuits to come from businesses that don't want the competition. A licensed professional could use this law to go after a tool that was starting to take their customers.

This can be fixed. California’s version, AB 489, is narrower. It also lets state agencies enforce the rule instead of opening the door to a lawsuit from anyone.

It’s unpopular to talk about AI’s upside, but this is part of the problem. The harms make the news, but the benefits are hard to count. Thus, we write laws from a lopsided picture. The costs won’t be clear either. We won’t see the careful, useful tools a nonprofit or a volunteer would have built.

We want more places giving careful, useful information, not fewer. For lower-income people, the status quo is already no information. A law that leaves them with even less is not consumer protection.

Note: I didn’t dig into whether this bill affects Anthropic or OpenAI directly. One law firm says it doesn’t; other sources say it does. The law may apply when you use their models directly, but not when someone else deploys their model inside another product. I stuck to the parts I’m sure about.

Fact checked with Tom Walczak’s Claude skill.

I was a U.S. diplomat for over ten years. The toughest day of my career wasn’t when a bomb exploded outside the embassy.

Walking back to my desk afterward, the embassy felt different, vulnerable. But I knew we were in good hands. Multiple levels of American and local government protected us. We were prepared. Months earlier, I practiced evacuating a mock burning building.

The toughest days were after the 2016 election. I was stationed in Muslim countries for much of Trump’s first presidency.

The first people I saw after the election result were my Senegalese embassy team, who mentored and trained me. Local embassy staff serve alongside Americans, sometimes at personal risk. Reading the headlines about calls for Muslim bans, tracking Muslims, and closing Mosques hurt.

Because they’d seen how multicultural the American diplomatic corps is, I didn’t get the hardest questions from them.

Outside the embassy, the questions made me rethink what I knew about democracy.

Locals in Senegal and Egypt often asked:

How does a democracy elect someone running on this kind of rhetoric?

Me: Democracy does this sometimes. We abide by the votes no matter what.

How can this speech be allowed?

Me: In America, the government cannot punish you for your words except in very few cases.

Should we trust democracy if democracy can lead to this?

…

I thought democracy and liberalism would be easy to defend.

I knew democracy was the best form of government. I could feel it. But standing there in front of people hurt by this rhetoric, the explanations weren’t satisfying.

I realized for the first time that democracy can be deeply unpredictable, worrying, and strained. It requires active maintenance. It actively reshapes itself; it lacks overarching control. As Dan Williams writes, handing over power to people with different political or religious views and tolerating dangerous and offensive speech doesn’t come naturally to humans; “they are insights that must be achieved.”

The hardest part was free expression. Defending it sounds like defending the speech's content. But you’re actually defending everyone’s right to say what people or the government doesn’t agree with.

In Egypt, journalists and activists are regularly investigated, jailed, and sometimes disappeared. The laws allowing this aren’t labeled “censorship”. They protect public order or religion. The same tools used to “protect the peace” are what governments use to silence dissidents and opposition leaders.

Free expression is how people stay safe as they push back against those in power.

What the local front pages didn’t convey

I often clarified that the majority of Americans are not anti-Muslim. Americans elect Muslim members of Congress. America’s multiculturalism doesn’t change because of who is dominating the news.

I feel this deeply.

Sometimes, people told me I didn’t look American. I never doubted for a second that I am America. They were wrong.

My brother is America. He came here at 12 when his mom died, after cleaning graveyards and bagging groceries in Mexico. He learned English and put himself through college. He became a decorated U.S. Marine Captain.

I learned English in public school and relied on subsidized school lunches. A few years after waiting tables and attending community college, I became a diplomat at 23.

My brother and I made careers in elite public service despite coming from an immigrant family with no political connections. Most systems in history kept power where it started. The messiness of democracy is what decentralized control looks like.

Even U.S. political families lose elections and power. The power to choose is the people’s. This also means elections can lead to any result.

The U.S. system needs work and is causing pain. It is far from ideal.

But the underlying structure is worth defending. Power can be contested, speech the government hates is still protected, and elections decide outcomes.

I’m honored I got to do that work. Even on the days when it hurt.

That’s when it’s most important to defend. When it’s hard. When you’re questioning whether democracy deserves it.

Psychiatrists are in high demand, so it can take months to see one! Psychologists are also trained to prescribe antidepressants, but often not allowed to.1

Idaho allowed pharmacists to renew insulin prescriptions. Idaho’s ER visits and medical bankruptcies dropped. Usually, renewing required a doctor’s appointment.

When Idaho allowed pharmacists to renew inhalers, fewer families went into medical bankruptcies, and fewer kids went into medical shock.

Montana and Colorado went further. Pharmacists can now diagnose and treat basic ailments like UTIs, pink eye, and cold sores.

These require no new spending or new training. Access to medical care improved by allowing medical professionals to do what they’d already been trained to do.

Occupational licensing is why healthcare is cheaper in some states

Primary care costs roughly a third in Pennsylvania as in many other states, partly because nurse practitioners can see patients for routine visits (e.g., a cold) without a physician present.

Whether nurse practitioners, physician assistants, and pharmacists can do what they’re trained to do depends on state regulations and the strength of medical lobby groups. This is a big deal because there are more nurse practitioners than general practice physicians in the US: 200,600 NPs vs. 153,980 GPs in 2019.

The medical lobby argues that care is worse without a doctor present. That’s technically true. But it ignores the actual alternative: people not seeing anyone if they can’t afford or access a doctor. When more mid-level providers are checking in, more breast and prostate cancers get caught at Stage 1 instead of Stage 3.2

The people who lose out first are rural and low-income. Whether a Minute Clinic can open in your area depends on whether nurse practitioners can pay a supervising doctor $500 to $1,200 a month. If doctors are scarce, this might make clinics impossible in areas with the most need.

This is a problem that democracies will always have to watch out for

These protectionist rules will naturally arise because organized interest groups (businesses or professional associations) lobby for them and set their requirements, often to block competition.

Dentist groups block dental hygienists the same way physician groups use “scope of practice” to limit nurses and pharmacists.

Strict licensing also creates shortages. Foreign-trained physicians must repeat their entire training to practice here. Telehealth providers can’t see patients across state lines, even in rural states with the greatest need.

Legal representation has the same problem. In New York in 2010, 98% of tenants in eviction cases and 95% of parents in child support cases appeared in court without an attorney. To provide legal services, you must have a law degree, pass the bar exam, hold a valid license, and work for a firm “owned, managed, and financed exclusively by lawyers.” This raises the quality of legal services, but means more people go without legal services.

These protections are often not set at a level that helps the public

• Professionals often design their own licensing standards, making requirements arbitrarily high.

  • Until 2024, Louisiana required a floral arrangement exam judged by existing florists, who evaluated their own would-be competitors. The pass rate was 36%, harder than the bar exam.

  • Nevada requires travel guides to undergo 733 days of training and pay $1,500.

• Interest groups use licenses to block competition in unrelated sectors.

  • In many states, hair braiding is illegal without a 400+ hour cosmetology license covering coloring and chemicals. Natural hair braiding, which has been passed down through generations, is pushed underground, with women of color or immigrant women at risk of being shut down.

• Licenses are easy to add and hard to remove.

  • In Washington State, massaging a horse required a human massage license (hundreds of hours of training for a different skill) plus 100 additional animal-specific hours. To fight the requirement, the Washington Animal Massage Association proposed a new standalone 300-hour animal massage license as an alternative. No one testified against it.

• The schools built around licensing requirements fight to keep them.

  • Cosmetology school tuitions now run $15,000-$20,000, even though median pay is $17 an hour. Nail techs often need 400-600 hours, barbers and cosmetologists often need 1500. Only 10% of those hours are related to safety. For context, an Emergency Medical Technician (EMT) only needs around 120-180 hours.

These dynamics explain why occupational licensing now covers 30% of the U.S. workforce, up from 5% in the 1950s.

Sometimes, licenses are good! After WWII, they helped women prove their qualifications in a workforce that was skeptical of letting them stay. They do raise wages, though often by raising prices or by outlawing competition (like banning nurse-owned clinics).

Well-structured licenses clearly improve safety and quality in two fields: medicine (particularly chiropractic) and the police field. Outside those fields, licensing has little to no effect on quality, or sometimes makes it worse by shielding providers from competition.

These also tend to block women, immigrants, and Black workers from licensed occupations. Sometimes intentionally.

Licensing that excludes people of color, southerners, foreigners, and convicted felons

• Massachusetts required a fortune teller license largely to keep Romani people and traveling carnivals away.

  • Similarly, Texas is considering offering construction tests only in English (no longer Spanish or Vietnamese) and requiring citizenship.

• “Moral character” clauses bar people with felony convictions from licenses in half of all states, for any conviction, regardless of relevance or how long ago it occurred.

  • Sometimes, people are trained and working in a field while incarcerated (e.g., forest firefighting) and then barred from that same work after release.

    • States with the heaviest burdens of occupational licensing saw a reoffending rate increase of over 9% within three years of release. The states with the lightest burdens saw a 2.5% decrease. Employment after prison is one of the strongest predictors of not reoffending.

  • Only 21 states allow people with criminal records to ask a licensing board whether their record will disqualify them before they spend time and money on required training. Proving rehabilitation to licensing boards might be impossible.

• Southern states have more local-level occupational licensing than the rest of the country. Much of it dates to the Jim Crow era, when it was used to stop formerly enslaved people from moving freely through the South.

• The American Medical Association funded a 1910 report that closed 75% of American medical schools, including five of the seven Black medical colleges and most schools that enrolled women and lower-income students. It eliminated the apprenticeship model that offered an affordable path into the nursing field via hospital training.

  • The new requirements called for two years of basic science followed by two years of clinical training. Schools whose students couldn’t afford four years without income closed.

  • Its author claimed to support Black medical education. His writings said Black students should treat only Black patients, fill roles subservient to white physicians, and be trained in “hygiene rather than surgery” to protect white Americans from disease.

  • The closures also hit the schools that had begun admitting women. The few that survived mostly didn’t. The result was the near-elimination of women from the physician workforce between 1910 and 1970.

The downsides are spread out and hard to see

• Licensing costs fall on workers who can’t afford it. Low-income women take on thousands of dollars in debt for for-profit cosmetology schools.

• Shortages in critical jobs, e.g., nursing. When training costs fall entirely on workers, fewer people can afford to enter the field, no matter how much the job is needed or how well it pays once you get there.

  • Less surge capacity after disasters. After Hurricane Ian, Florida needed roofers quickly. Workers came from neighboring states. Some were arrested because their licenses didn’t transfer.

• People can’t move when they need to. Licensing might have cut interstate mobility by up to 7%. When Arizona announced it would accept any other state’s professional license across all fields, more people moved there, and tax receipts increased.

  • My co-author Alicia notes that her mother, a P.E. teacher, would have had to redo her schooling after the family moved to South Carolina.

• If you move states, you need a new therapist, even if you meet online. When I worked in government and wanted evening therapy appointments, few DC-area therapists had evening slots. I wasn’t allowed to see a California therapist who could have seen me virtually during her afternoon.

• Prices increase by around 15% when a service becomes licensed. Daycare costs run 30-50% higher in states with stricter licensing requirements, with no measurable safety gains.

What can change

Almost all occupational licensing happens at the state level (see the full list). The only federally licensed professions are aircraft pilots and maintenance technicians.

At the state level:

• Require licenses to be periodically reauthorized with evidence that they protect safety, not just that the licensed industry prefers them. Utah does this.

• Reserve full licensing for fields where evidence shows safety gains. For most professions, certifications or apprenticeships work and cost far less.

• Encourage paid apprenticeships. In construction, employers can get liability coverage for workers they’re training, creating a paid path in without years of upfront debt or unpaid coursework.

• Stop barring people with criminal records from licenses for convictions unrelated to the job.

• Expand Universal Licenses. Half of states recognize credentials from other states in some form, but most versions cover few professions, default to the more stringent requirements, and/or require two years of in-state residency before kicking in.3

Federally:

• Use federal funding conditions or the interstate commerce clause to push for license portability and felon licensing reform. The federal government cannot directly regulate state licenses.

• Create tax incentives for apprenticeships.

This essay was inspired by our last DC Abundance gathering, which featured West Virginia University Knee Regulatory Research Center Director Alicia Plemmons. Sign up here for our next DC Abundance meetup. All errors my own.

People get stuck in low-wage work for a variety of reasons, often a lack of transportation. In much of the U.S., a bus is not reliable transportation. It often doesn’t show up, or it shows up late.

A car is only reliable if it runs. When my brother’s car broke down, he lost his job at Red Lobster. He couldn’t get there consistently, so they let him go. He needed the job to fix the car.

That car is both essential and precarious. The poorest fifth of American families spend 38 percent of their income on their car, if they even have a car. When that one car breaks down, the whole household scrambles.

The Waymo debate in America focuses on liability frameworks, rideshare driver jobs, and edge cases where the technology fails. It should also focus on how bad the status quo is for poorer people. Self-driving cars, if they get cheap enough, could be a safe alternative to car ownership for people who can’t afford one or can’t drive.

If self-driving cars get cheaper the way refrigerators and cell phones did, borrowing a car for 20 minutes becomes cheaper than owning.1 If self-driving cars are available to purchase, several family members, or even several households, could share one. One car could drop a parent off at work, return for another, and still take an elderly relative to a midday medical appointment. Waymos could become an on-demand shuttle service similar to our early-1900s privately run streetcar system. Not owning a car wouldn’t mean not having reliable transportation.

Transportation often keeps people stuck

When you earn $15 an hour, getting a working car is its own project. A used car costs $3,000 to $8,000 upfront. Before it’s useful, you need insurance, a license, and registration. My brother’s one and only night in jail was because he couldn’t afford insurance and registration.

You also need $500 to $800 on hand for whenever the check engine light comes on.

A lot of people I know never cleared those humps. They asked family for rides. One friend walked over an hour to work every day, even with a sprained ankle.

Even when you do get a car, it runs your life. So much of my mental bandwidth went to keeping mine running. I knew the status of every car in my circle: mine, my friends’, my parents’.

When my car broke down on the highway during a storm, I cried. Trucks blew past so fast the frame shook. I knew one wrong swerve would have killed me. But I was more worried about missing class. A missed class meant a B. A B could cost me my chance at winning a scholarship. My straight As and club presidencies would count for nothing.

I’d waitressed to buy that car. Other kids got theirs as gifts and half-tried in school. I worked extra hours just to compete with them. That wouldn’t show on a college application.

I was one of the lucky ones because I had a car at all. I saved up because my parents didn’t often need money from me and could drive me to my first job, a privilege my older brother didn’t get. Nothing was stopping me from driving.

A lot of people can’t drive.

Some friends of mine have seizures. Some have anxiety around driving, which is understandable. We give 16-year-olds two-ton vehicles for roads where families walk on sidewalks a few feet away.

Some have revoked licenses after DUIs and drive anyway, because being stuck at home without work is worse. It’s hard to keep people off the road when the road is the only way to get around.

Waymo could make a dent in this; even Uber did. When Uber came to Houston, alcohol-related crash fatalities dropped 24%.

Road deaths fall hardest on the people with the least margin

About 40,000 people die on American roads every year. More of those deaths happen in low-income communities. In D.C., where I live, lower-income neighborhoods have eight times more traffic fatalities than the city’s wealthiest area. Nationwide, high-speed urban roads in low-income areas have 300 percent more crashes than those in high-income areas. There’s surprisingly little research on why.

There is even less research on how road deaths make families poor and keep them that way.

A loved one of mine lost her partner in a crash when their son was a baby. She ended up in a tough place and got together with someone she knew was abusive. She didn’t feel like she had other options. She stayed for decades.

My friend who walked miles to work on a sprained ankle was covering his own rent at 18. His working parent had died in a car crash. His remaining parent couldn’t legally work in the U.S.

When you live in that world, you don’t think to blame the car. It all just looks like bad luck. It doesn’t have to. Waymo vehicles are involved in 92 percent fewer serious-injury crashes than human drivers on the same roads.

I’d rather have trains. But.

I want more public transit, walkable cities, and trains. I’ve advocated for them for years. The Netherlands cut road deaths sharply without self-driving cars, and we could too.

We don’t have the political will. If Waymo can move things in the right direction, we should let it.

If a quarter of cars on the road were Waymos, families would own fewer cars and cities would need fewer parking spots. This is a big deal. Concerns about parking are a major reason neighbors block dense development. In U.S. cities with more than a million people, 22 percent of downtown land is parking. That is some of the most valuable, walkable land in the city. Replace some of it with housing, and you get more homes and more walkable neighborhoods.

Much of the Waymo debate centers on rideshare drivers. They are a small slice of the people autonomous vehicles will affect. They are not the lowest-wage workers in the picture. Rideshare drivers at least have a working, clean, relatively new car. That already puts them ahead of the bottom quintile.

Ninety-four percent of Lyft drivers work fewer than 20 hours a week. Most stay less than two years and use rideshare to fill gaps between other shifts. Only about half of Uber drivers are still driving after a year.

Even if the job losses were larger, it would feel wrong to hold back a technology that could change how millions of people get around. Road deaths and car costs fall hardest on the people with the least margin to spare. They already have enough working against them.

Effective Altruism has a great track record. In 2023 alone, an EA-backed organization prevented 40,000 deaths and 20 million malaria cases. Another moved over $1 billion in cash directly to the world's poorest families and changed how USAID measures its own effectiveness.

But EA has a blind spot: systems change. It’s harder to measure, the benefits are diffuse, and it seems less tractable. But it’s often more impactful.

China’s decision to liberalize its economy lifted hundreds of millions out of extreme poverty. The economists who inspired that decision only needed a 10% chance of changing policy for their work to be competitive with EA’s most effective direct interventions.

EA is missing this.

We focused on bednets while a malaria vaccine was stuck for 23 years

Bed nets are remarkable. Saving a life for $5,500 is extraordinary. We should fund this.

But there’s been a malaria vaccine since the 1990s.

RTS,S—the vaccine the WHO finally recommended in 2021—spent decades in clinical trials and agency reviews with no urgency, even though malaria kills over half a million people a year, mostly children.

After Phase III trials confirmed it worked in 2015, the WHO required an additional multi-year pilot program due to meningitis cases in a few tested regions. Those cases turned out not to be caused by the vaccine, which most observers expected.

No institution in the approval chain was required to weigh that six-year delay against the vaccine’s 13% reduction in child mortality.1

Abundance, at its core, is responsive governance

Abundance asks: Is the government actually delivering on what it set out to do? What are the unintended consequences of current methods?

Democracies rarely ask this. Drug approvals take decades. In most U.S. cities, building housing is nearly impossible because rules that were each reasonable in isolation have accumulated into a system where neighbors can block almost anything. The national daycare shortage is made worse by the fact that daycares are illegal above the first floor in most states due to fire codes written before sprinklers existed.

Abundance is the project of finding and fixing these failures in the sectors that make the biggest difference, like science.

Why abundance is important

Science. The federal government almost didn't fund the research that led to the COVID-19 vaccine. Katalin Karikó spent decades trying to get grants for mRNA technology. Her university told her to abandon it or accept a demotion. She took the demotion. mRNA got funded only because DARPA, a small agency that bets on speculative science, took a chance. The federal grant system is now so averse to failure that the next Karikó might not find support.

Cost of Living. The U.S. housing shortage shows up in prices and in lower productivity growth. Austin built a lot of housing; rents dropped and are now cheaper than before the pandemic. The same rules blocking housing also block daycares, medical clinics, nursing homes, and clean energy facilities, and impact prices. Cost of living fuels public discontent, which is bad for democracy.

State Capacity / Democratic Trust. New York City spent six years and millions of dollars failing to repair the Wollman Rink in Central Park. The city kept running over budget and missing deadlines. Donald Trump offered to take it over, finished it in four months, and came in under budget. He became a local hero in 1980s NYC. When the government can’t do basic things, people willing to bypass the usual process appear more competent. That’s a bad norm.

The stakes get higher with forest fires. Controlled burns, where the government deliberately burns smaller areas before the dry season, are the most effective tool for preventing catastrophic wildfires. Californians have lost homes and lives to fires that were likely preventable. But the U.S. Forest Service needs three to ten years to complete the required environmental paperwork for a single controlled burn. Fires happen during that wait.

Toby Ord lists “accumulation of bad bureaucratic features” as a source of existential risk to civilization. This is how it happens: gradually, through a thousand small failures that erode trust in institutions. When government looks incompetent, polarization increases, norms weaken, and authoritarian alternatives look more appealing. Democracy is held together by collective agreement to democratic norms. These failures are what a democracy looks like as it becomes less resilient.

A functional democracy is not just good for Americans. A stable America and international respect for democratic norms are a global public good.

Global poverty. Economic growth is the single biggest driver of escaping poverty, more than foreign aid, direct transfers, or development programs. Almost all of the reduction in extreme poverty from 1990 to 2020 came from Asia. Sub-Saharan Africa received substantial aid over the same period.

What made Asia different? Countries like South Korea, Taiwan, Singapore, and Hong Kong started from roughly the same income levels as many African nations in the 1960s. They invested in infrastructure, reduced barriers to starting businesses, and opened to trade. They adopted abundance-oriented governance.

Abundance in the U.S. has knock-on effects that help global poverty. When the U.S. develops and commercializes a technology, it spreads globally. South Korea, Taiwan, and Singapore built their economies partly by adopting American semiconductor technology as the foundation for export-led growth. They went from poor to rich in a single generation. Solar is following a similar path. Panels are now cheap enough that developing countries can leapfrog fossil fuels.

Energy is a bigger deal than most Americans realize. There is no such thing as an energy-rich poor country. I lived in Nepal. Rolling blackouts hit for multiple hours every day with no warning. It’s hard to hold an internet-dependent job or build industries that can compete globally. The faster rich countries develop clean energy, the sooner the costs drop for everyone else.

When democratic governments visibly build things and work well, they also make that model more attractive globally. Competence is a form of soft power.

Why abundance is neglected

Abundance failures are structurally invisible.

When a housing project dies in permitting or a researcher can't get a grant for speculative science, no one hears about it. We don't know how much better things would be if the federal government didn't need three public notices to create a voluntary 5-minute meeting request form. The people who would have benefited don't know they missed out.

This makes abundance different from most policy issues. Gun violence produces victims with names. Pollution produces sick children. Abundance failures produce a much worse world than it should have been.

There's also no constituency for fixing it. The people who benefit from faster permitting and cheaper housing are everyone, a little. The people who benefit from the current system like incumbent businesses or existing homeowners are concentrated and organized.

Even abundance successes tend to stay invisible. When thousands of bridges get maintained without issue or when the FDA approves a drug faster, it doesn’t make a headline.

Building coalitions around invisible problems is hard.

Why abundance is tractable

Ezra Klein and Derek Thompson’s book Abundance became a bestseller. Abundance groups are popping up. Students for Abundance chapters are forming at universities. Something has shifted.

The work is also more concrete and measurable than it looks. Abundance is a list of specific questions with clear answers.

• Why did the infrastructure bill fail to produce more housing?

• What would the FDA need to change to approve treatments faster?

• What’s blocking nurse practitioners or doctors from opening their own clinics?

• Why did the mRNA technology almost not get funded?

The abundance agenda is still being written. The bottleneck is more people to do the sector-specific research and enact fixes.

What this means for AI risk

If you only care about AI risk, abundance is still worth your attention. Almost every AI scenario goes better with a government that works.

COVID is a good example. The U.S. invented mRNA but still lagged on clinical trials. Trials happened partly in the UK because its data infrastructure was better. China sent us face masks during our shortage. The failures e.g. slow diagnostics, slow manufacturing, slow regulatory approvals are what abundance is trying to fix. This boosts government capacity and biodefense.

Abundance also helps the government have surge capacity for shocks like a financial crisis or a jump in unemployment. Medicare, Medicaid, and debt interest are consuming a larger share of federal revenue. Cheaper healthcare, less regulatory drag, and more functional agencies create fiscal room to respond.

Not every way to prepare for AI is about AI itself. Better underlying systems help a lot.

The movement needs people

Abundance won’t solve any single EA cause. But it makes progress on almost all of them. It needs people.

I didn’t mean to start DC Abundance. A friend and I hosted a meetup to see who’d show up. We hoped for eight. Over fifty came. It felt like early EA: good ideas, not much infrastructure yet, and people figuring out what comes next.

You can be a part of that.

A few years into my Foreign Service career, a colleague leaked something to the press. The thing they leaked wasn’t true. I couldn’t figure out why they’d done it.

A more experienced colleague explained it to me: the leaker was trying to break into partisan Washington. Saying what powerful people want to hear opens doors.

That colleague was competing in a world with entirely different rules.

I’d never operated in that world. I was deep inside what some people call the “deep state,” the long-tenure career civil servants who outlast administrations.

From our hiring process to the way our analysis is tested, we build a deeply apolitical culture that probably seems just as foreign to others as the partisan world seemed to me.

This culture is a core structure of good policymaking and is constantly under attack. I want to start by explaining how this culture happens.

The exam nobody can game

The Foreign Service selects people through a long exam process that ends with an eight-hour simulation of embassy work. Most people fail. If you pass, you go on a ranked list by score and get called in order. You cannot get in by knowing the right people or donating to the right party.

My colleagues came from all fifty states and a variety of careers. Some were former school teachers, software engineers, lawyers, and journalists. The process doesn’t strongly reward growing up on the coasts, majoring in global affairs, or reading the right publications.

This system was designed to replace the “political spoils” system where jobs went to loyal partisans. Reforms seemed urgent because a sitting U.S. president was assassinated by a man who felt he’d been denied a spoils appointment.

The culture it produces

A near-random, deeply meritocratic hiring process, combined with constant feedback loops on whether your work is accurate, help make a culture that took me years to fully understand. I’m still learning what that culture is, mostly by leaving it and observing the contrast.

Here’s what it actually looked like in practice.

We rotate so we can’t get entrenched. Diplomatic assignments change every two to three years. You might spend two years covering migration issues in Southeast Asia, then move to human rights in Latin America. You can’t build a career by becoming the champion of one cause. You master understanding complex issues quickly, learning which experts or practitioners to seek out.

We treated headlines as a starting point, not ground truth. We had newspaper subscriptions from across the world. I noticed that mainstream headlines were a tool for understanding what those media circles believed, not what was actually happening. Almost every time I dug into a story, a lot of context was missing. You have to build trust with people on the ground to get the full context.

We were almost entirely disconnected from online political discourse. For ten years, I wasn’t following debates about what progressives or conservatives should believe. I struggled to find political Twitter useful or addicting. The things I needed to know I got by talking to people in the field, reading reports, and coordinating with analysts across the government. Online discourse told me what highly online populations were feeling. It wasn’t an efficient way to learn about reality.

Good analysts could represent every stakeholder in the room. If you were preparing an ambassador on a major IMF loan, your analysis had to ring true for labor groups and bankers since both might be in the room. If you were drafting trade policy recommendations, you had to understand the business community, labor groups, and the political constraints on every actor involved. Analysis that read like advocacy was looked down on. It meant you didn't understand the other side. The government interacts with every side.

An agency can lean left and still be nonpartisan

The State Department leans left, primarily because the professional class of degree-holders leans left. This creates blind spots, but it isn’t the same as being partisan as an institution.

Partisan means operating inside a coalition: you produce outputs that serve a partisan group’s goals and get rewarded. The career civil service doesn’t work that way. There is little reward1 for telling a new administration what it wants to hear.

When a new administration arrives, career officials brief them on every country, every negotiation, and every relationship, often with years of inconvenient history attached. That context sometimes conflicts directly with what the administration wants to do.

From the outside, an obstruction and someone accurately describing reality look identical.

Replacing career hires with loyalists doesn’t fix the ideological imbalance. It makes the next cycle worse. When a new administration arrives and finds the executive branch stacked with the previous party’s loyalists, they’ll purge more aggressively than they otherwise would have. Each administration leaves the next one with more justification to do the same. The career civil service shrinks a little more each time. Both parties have claimed the deep state was working against them, so both will find reasons to hollow it out gradually.

Like much of the national security world and the military, the Foreign Service comes from every state and every background. We are not a coordinated group. The only thing we had in common at the start was wanting to serve and passing the same exams. What developed over slowly, through rotations, peer review, and getting things wrong in front of stakeholders, was a culture of telling the truth, especially when no one wanted to hear it.

People feel guilty about using Uber. This confuses me.

Uber1 pays $15 to $25 an hour, usually $19, around $30 in big cities. This is good pay, even after substracting car expenses. Two of every five U.S. jobs pay less than $20.2 One in five U.S. jobs pay less than $15.

When I ask Uber drivers about their main work, the answers are usually restaurant management, security guard, or someone starting a food truck. These aren’t the sub-$15 jobs in America. Those are home health aides, daycare workers, and retail associates, jobs I’ve written about before.

Growing up around people making $5 to $15 an hour, I saw that one thing kept many of them stuck: transportation. Uber driving requires car that works, is new enough, is very clean. You can’t let your kids dirty it.

I could count on one hand the people in my community who had that.

The fact that anyone with a car and a clean background can get near median U.S. hourly wage of $23.80 is great.

Nonethelesss, gig work is hard. But feeling guilty and avoiding Uber means drivers have slower nights. When I waited tables at $2.13 an hour, more customers meant more income. More trips means more income for drivers.

What to do instead of feeling guilty:

Be ready when they pull up. Every minute they wait is unpaid.

Pin your exact location. Circling the block looking for you costs them unpaid gas and unpaid time.

Don’t slam the door. The loudness can be painful.

Keep the car clean - a spill will end their shift! Any messes, like food or sand, or even cigarette smell can force a driver off the road mid-shift to pay for a cleaning. If they only had a three-hour window between jobs and family, that ends their workday.

Don’t spill alcohol. This can take them out of commission for days. Any lingering smell means passengers assume the driver was drinking. Drivers might pay for multiple cleanings before they can get back on the road.

Use your real name. People can sign up as “Daddy” or other placeholder names. If a driver gets robbed, that’s the only name they can give to the police.

Drivers worry about getting attacked by passengers, though this shows up less in the online discourse. Passenger-victim stories spread further than driver-victim stories, likely because most readers are far more likely to be passengers.

Drivers also feel unseen by how their work is talked about. "I hate how the media portrays us as victims," one told me. Most drivers aren't trapped in gig work. 94 percent of Lyft drivers work fewer than 20 hours a week. It's a second income, something to pick up between shifts. The complaints drivers have are more specific: how little of the fares they receive, high platform fees, insurance requirements that apply to rideshare but not taxis.

The exploitation frame comes, I think, from a class gap in who’s doing the writing. From a certain vantage point, gig work looks like an awful deal. From inside the $20-and-under labor market, it looks different. It’s a job that doesn’t require English fluency, doesn’t require a resume, and pays more than waiting tables.

But even sympathetic coverage gets the policy diagnosis wrong. It focuses on what’s broken about one particular sector, even though low wage workers don’t stay in one sector. People move between food service and retail and rideshare depending on what’s available.

When I had three jobs, making any one of them better wouldn’t have helped much. What I wanted was health insurance that didn’t depend on me keeping any particular job. Big policy goals like cash transfers and cheaper housing help no matter what job you’re in. These goals are more difficult and have less clear villains to blame, but they deserve more attention.

Some of the rules we’ve built have made it harder for people to seek care and live independently. Here are some things we could change, most of which require no new spending.

I plan to keep updating this list.

1. Make it legal to rent out a lockable bedroom in a shared building

This is illegal in most cities.

Single Room Occupancy hotels, which are a private bedroom and shared bathrooms, housed millions of low-income Americans for most of the 20th century. In 1950, SRO units were 10% of all rental units in some major cities.

Elderly people were as much as half of SRO tenants. However, most cities outlawed them in the 1980s. This made the lives of the elderly and poor more precarious. Pew notes this increased homelessness:

As the nation’s least expensive source of housing disappeared, homelessness soared in major cities across the United States. In New York City, for example, the number of homeless residents increased from a barely visible population to almost 30,000 by 1987. About half of men entering homeless shelters in the city in 1980 reported they had previously lived in SROs.

Seniors are the fastest growing group experiencing homeless.

2. Let families build a room on their own property for a struggling loved one

In much of Japan, you can build things on your lot like a studio or sleeping quarters for a relative. Most of America is zoned single-family only. You usually can’t add a backyard room for an aging parent.

Backyard units used to be how people with less money lived near family. It also was a source of modest passive income for the families who built them.

Accessory Dwelling Units (ADUs) are an attempt to bring these back, but most jurisdictions make them difficult to build. For example, with “temporary use” permits that expire when the senior dies, “removability” requirements that conflict with building codes demanding permanent foundations, sunsetting clauses that make them bad investments.

3. Treat manufactured homes the same as any other home in zoning law

A manufactured home or a “mobile home” costs roughly half what a site-built home does. People over 60 are a third of adults living in them. It’s one of America’s largest sources of affordable senior housing.

Many jurisdictions ban them outright, require large minimum lot sizes, prohibit replacing a unit when someone moves out, or segregate communities far from doctors, transit, and groceries. Vermont and Maine effectively legalized these homes: no zoning regulation can exclude manufactured homes except on the same terms as conventional housing.

4. Let small groups of people share a home and a caregiver

Four to six seniors sharing a house and a caregiver is effectively illegal. Three barriers stop it. Many zoning codes cap unrelated persons at four. Medicaid allocates aide hours per individual with no way to pool them, even though one aide under one roof is more efficient than four aides visiting four homes. And the moment seniors share a caregiver, many states classify it as an assisted living facility requiring fire codes, staffing ratios, commercial inspections designed for a 100-bed building.

These regulations and licensing requirements work well for protecting people in institutions. However, requiring commercial kitchen inspections for a house with four seniors and a caregiver stops people from forming tiny community-centered solutions.

The fix: create a regulatory category between “person alone at home” and “licensed facility” e.g. a small shared care home with proportionate safety rules.

5. Stop requiring government permission to add nursing home beds

NIMBY blocks some nursing homes, but another barrier is that 34 states require a Certificate of Need just to add nursing home beds. You need state approval to build supply. Sixteen states go further with outright construction moratoria.

These laws were designed to prevent overbuilding and control Medicaid costs. In practice, they protect existing facilities from competition and keep supply artificially low. Repealing CON for nursing homes and classifying small care homes as a permitted residential use by right would expand options.

6. Remove the legal ambiguity around homesharing

Homesharing matches older homeowners — property-rich, cash-poor, often isolated — with people who need a room to live in and are willing to help in the home for some hours. More than half of participants over 50 feel safer, happier, and less lonely.

It would happen on its own if the legal framework were clear. But running a matching service at scale can trigger commercial-use zoning violations. Advertising a spare room raises questions about whether your home is now a rooming house. States should explicitly classify renting a room to a matched housemate as a residential use.

7. Make it clearly legal to open your home to someone else who needs meals and daily help

Right now, if you wanted to open a spare bedroom to an elderly neighbor who needs meals, company, and help with daily tasks (but not medical care) you probably couldn’t tell whether that’s legal due to liability, facility classification, residential home use, and Medicaid reimbursement.

Pennsylvania is helping get this service off the ground: the state runs a program where ordinary people host up to three adults who need daily assistance. The host gets paid. The guest lives in a home, not an institution.

This often doesn't require a new program. It requires states to clarify the arrangement is legal.

8. If states raise the minimum wage, they need to raise Medicaid home care rates too

Home health aides are projected to be the largest occupation in the U.S. economy by 2032. Fourteen million Americans depend on them to stay out of institutions. In every state, their median wages trail retail and customer service by $3.73 an hour, with 75–80% annual turnover.

The reason is structural: Medicaid pays for nearly two-thirds of all home care spending, which means most aides’ wages are set by state reimbursement rates, not the labor market. When a state raises its minimum wage, every private employer adjusts automatically. Medicaid-funded agencies can’t unless legislators separately vote to increase reimbursement rates.

The result: every minimum wage increase makes home aide work relatively worse compared to every other low-wage job. Hence, a shortage.

9. Let home aides work in any state without re-licensing

Some states require more than double the 75-hour federal minimum for training. These extra requirements fragment what should be a national labor market. A single federal standard would make workers’ lives easier, and allow them to move to where the need is.

10. Create a visa for caregivers

38 percent of home health aides are foreign-born and may be at risk from changes in their legal status. There is no temporary working visa for them. Temporary visas for nurses, authorized in 1989 and 1994, both expired and were never renewed.

11. To address the nursing shortage, let hospitals train nurses again

80,000 qualified applicants were turned away from nursing programs in 2024 because of lack of instructors. The wage gap between faculty roles and clinical roles is $18,000–$27,000. Letting hospitals run training programs again would help. West Virginia University Medicine has a hospital-based nursing diploma program for high school graduates.

In the 1980s, professional associations pushed nursing toward four-year university degrees, shifting training away from hospitals into universities and colleges with government-capped budgets.

Some of these rules persist through inertia. Others are actively defended by the operators, associations, and property owners who benefit from them. The people harmed are scattered and often don’t know what they’re missing.

None of this replaces stronger social supports: better cash transfers, expanded Medicaid, a long-term care benefit.

These aren’t in competition. You need both.

I spent six years in low-wage work. Waitressing, a call center, selling laptops at Circuit City, handing out T-shirts for a radio station.

Now I work in policy. When I read coverage of low-wage workers, it flattens something that's actually quite complex. Low-wage work isn't a job. It's a whole world with its own logic and norms. Most coverage misses that. I think that gap is costing the left politically.

Here’s what the media tends to miss.

Low-wage work is a system, not a job.

When I waited tables in Texas, I worked my way up to the coveted weekend morning shifts. Those shifts paid $90–110. Weekday evenings paid $30–40. So I filled the weekdays with other work.

A rideshare driver I recently interviewed picks up passengers during his commute to his main job managing a restaurant.

My mother was a daycare worker, a substitute teacher, and an Avon representative simultaneously.

This is how low-wage work actually functions. Workers stack income sources. They optimize each one (e.g. hours, tips, scheduling flexibility) and fill the gaps with others. Gig work isn’t a separate economy.

Media coverage usually treats fast food workers, gig workers, and home health aides as three separate populations with three separate problems.

Workers switch jobs constantly—and across sectors.

The turnover rate in low-wage work is roughly one job per year to eighteen months. Workers don’t just change employers. They change sectors.

The women around me moved between daycare, home health aide, cashier, and retail depending on who was hiring, which manager was tolerable, and which shift fit around their kids. Those latter three jobs (plus fast food or counter work) are the four occupations that make up 29 percent of all employment paying under $15 an hour.

Gender shapes the options. Women are 67 percent of workers earning minimum wage or less, concentrated in home care or childcare work. Men in my community more often went to the oil fields, manual labor, construction, mechanic shops. These jobs typically pay at the upper end of minimum wage or above it.

These aren’t separate labor markets. They’re one ecosystem with a lot of internal movement.

Large employers aren’t obviously the villains.

The dominant media frame is that large corporations exploit low-wage workers while small businesses are clearly preferable. Workers don’t really see it that way.

A friend worked at a small food stall. When it closed, the owners didn’t pay his last two weeks. Recovery took months.

When Circuit City shut down, I was paid for my last shift without any effort on my part. The corporate structure meant there was someone to complain to. It also meant the media would care if they didn’t. Journalists pay attention when the employer is a recognizable name.

If anything, the common complaint about large employers is that they’re selective. Small businesses are more willing to hire someone with no experience. I didn’t get a Starbucks offer until I had years of waitressing behind me.

Large employers aren’t the bottom of the labor market. They’re something workers must work up to.

The media's framing doesn't match what workers actually experience.

The reasons people stay are more complicated than the narrative allows.

Some people are stuck in low-wage work for concrete reasons: no reliable car, no childcare, limited English, a past record.

But others stay because of preference. Some people genuinely like hands-on work. Some prefer not to manage others even when offered the chance. Some have anxiety about school or uncertainty about careers.

Pretending these personal factors don’t exist doesn’t help the people dealing with them. It just makes the coverage feel inaccurate.

The political consequences.

Almost everyone I knew in that ecosystem was a Democrat. That’s not true anymore.

During Obama’s first term, Democrats represented 81 of the poorest quarter of congressional districts. Republicans held 30. By 2024, those numbers had flipped: Republicans held 65, Democrats 46.

A lot of factors drove that shift. But one of them, I think, is that working-class don’t feel seen in media or policy circles.

Consider the push for a $15 minimum wage for fast food workers. From where I sit now, having earned more than $15 for a decade, this seems like good and fair policy.

But if I were still waitressing, it would feel almost random. Why fast food? What about the daycare worker? The warehouse people? In the logic of the low-wage system, it looks like people only get help if they happen to be in the right sector at the right moment of political attention.

Or take the push to require college degrees for daycare workers. Media coverage treats this as progress. But it only makes sense from one direction: college-educated parents looking at daycares, not the women working in the daycare. For many workers, it's a pink slip.

I wrote about how daycare regulations drive up costs last week. Someone told me I was distracting from the fight against corporate power.

I know this pushback is well-meaning. But a $100 reduction in monthly childcare costs is enormous when you’re earning under $15 an hour. Dismissing it as a distraction is exactly the problem.

The left hasn’t stopped caring about working-class people. But the way it expresses that care has drifted away from how working class people talk and toward what college-educated advocates think they need.

Working-class voters noticed.

I keep hearing that abundance is a nice idea but it doesn’t help families who are struggling today. So I made a list.

These are state and local rules that make life more expensive than it needs to be. I started with childcare, healthcare, and housing because that’s where much of a family’s money goes.

Nobody wrote these rules to hurt families. Most were enacted for real reasons. But many haven't been updated in decades. The costs add up.

Every item on this list can be changed without new subsidies or government spending programs.

CHILDCARE

Nearly 70% of parents spend a fifth of their income on childcare. Federal subsidies are unlikely under this administration. State and city budgets usually can’t cover the gap.

But costs can come down.

Zoning bans formal daycare centers in most residential neighborhoods. This limits supply and raises prices. Even where zoning allows them, approval takes years and costs tens of thousands. One California daycare center tried to expand from 14 to 48 spots. Neighbors fought it for four years.

In-home daycares are often the most affordable option for families, but they’re underground and hard to find. About 90% of paid childcare settings are in-home daycares — someone caring for kids in the house or apartment where they already live. Roughly 85% of these are part of the underground economy, partly because requirements to run one legally are expensive and confusing. Licensing fees, occupancy permits, inspection rules add up fast. Reducing these costs would bring the most common form of childcare in America out of the shadows.

In-home daycares are also often banned because of zoning and HOA rules. Most states allow in-home daycares. But cities and HOAs ban them in practice, because zoning treats watching kids for pay as a commercial use of the home you live in. Some require a conditional use permit, which neighbors can block. California recently classified in-home daycares of up to 12 kids as a normal residential use. HOAs and zoning boards can’t block them.

Staff-to-child ratios vary wildly across states for older kids; the strictest ones raise costs without a clear evidence behind them. For older preschoolers, North Dakota caps it at seven kids per caregiver. North Carolina and Florida allow twenty. The most permissive states allow nearly three times the children per staffer, which means more revenue per employee and lower tuition for parents. Many states set ratios stricter than what the National Association for the Education of Young Children recommends. Researchers disagree on whether the strictest ratios meaningfully improve outcomes, but the impact on cost and supply is large.

Supervision and ratios often don’t account for flow during the day. While it makes sense to require a staff member for 6 toddlers when they are awake, some states (like Texas and Iowa) allow one staff member to supervise 12 napping kids so the other staff member can have lunch. In some states, daycares end up having to hire a floater to cover breaks so that two staff sit in a darkened nap room.

Strict age-group rules force providers to hold spots empty. Most states ban mixing age groups e.g. no three-year-olds with eighteen-month-olds, even briefly. When ages do mix, the ratio for the youngest child applies to everyone. Providers may need to keep spots open and not generating income to make this work. Even when parents feel their younger kids are ready to move up early, it is illegal. This also increases labor costs. To break even, small providers likely end up breaking this rule despite the legal risk.

Daycares can’t operate above the ground floor in most states. Moving upstairs could cut occupancy costs by up to 50% in high cost cities. Occupancy is the second biggest daycare expense after labor, at 10-20% of operating costs. Ground-floor commercial space is the priciest real estate. The bans stems from evacuation concerns, but NICUs operate on upper floors of hospitals. Modern fire safety, like sprinklers, and staff training have solved this concern. Most countries allow second-floor daycares.

Degree requirements price out good providers. Some states require college degrees for non-managerial daycare workers. This raises costs, on both workers and families, without clear benefits.

None of these reforms will make childcare cheap. Only government support for parents can do that. But they can make it cheaper, and they can do it now.

HEALTHCARE

Pharmacists can’t prescribe for common conditions in most states. Thirty states allow pharmacists to prescribe birth control. Only California, Idaho, and New Mexico allow pharmacists to prescribe medications for minor health problems like pink eye, cold sores, urinary tract infections, yeast infections, and minor skin conditions. Pharmacies are everywhere. Doctors’ offices often aren’t.

Dental therapists, i.e. mid-level providers who do fillings and extractions, are illegal in most states. Minnesota legalized them. Rural and low-income communities got meaningfully more care and cheaper care. The ADA fights these laws everywhere. They argue everyone deserves a full dentist. But cost keeps millions of Americans from seeing any dental provider at all. Also, dental therapists work under the supervision of a dentist.

Nurse practitioners and physician assistants can’t open their own clinics in many states. They can do most of what primary care doctors do. Twenty-nine states allow experienced nurse practioners to open their own practice. Only six states allow physician assistants to practice to the full extent of their training.

Certificate of Need laws block new clinics and hospitals from opening. About 30 states require permission from private local healthcare providers to open or expand medical facilities. The federal government stopped using these laws because they were found to reduce neither costs nor improve quality. The main beneficiaries are incumbent medical providers that face less competition.

HOUSING

YIMBY reforms work. Austin rents dropped roughly 15% from their pandemic peak after a construction boom. An Austin or Minneapolis-style YIMBY victory in California would cut child poverty by as much as the Child Tax Credit. Housing costs contribute to poverty rates.

The basic policy ask (legalizing duplexes, triplexes, apartments, and backyard cottages in residential neighborhoods) is well-organized and well-known.

But some less-discussed rules also raise costs significantly:

Parking minimums add roughly $88-500 to rent or mortgage costs. The range varies by location. Many cities require 1–3 per single family home or apartment. The cost gets passed to renters and buyers whether they own a car or not.

Three story buildings with only one staircase are illegal in most U.S. jurisdictions. Single-staircase bans make small apartment buildings expensive to build and rent. A second staircase in a four-story building costs $260,000 and takes up to 7% of floor area. Allowing a single staircase makes family-friendly apartments with three or four bedrooms feasible. Most countries allow single-stair designs for ten-plus stories.

Permitting add up to a third to the gap between what a home costs to build and what it sells for. In Los Angeles, the most populous county in America, it takes 4.2 years to get approval for a mid-size apartment building. Studies of land before and after approval show the permit process alone makes land 50% more valuable — a $1 million plot becomes worth $1.5 million with the permit. LA and San Francisco are extreme cases. But even across the rest of the country, a study of the 50 largest metro areas found none with both high rents and a high number of permits.

This is a living post. I’ll keep adding to it over time.

I’ll also add elder care, food costs, occupational licensing, and more in future posts.

If you know of a rule I should include, let me know.

My mom was a daycare worker for most of my life. My best friend’s mom was one too. My childhood memories consist of many daycare workers, mostly women.

Daycare work usually pays minimum wage. The women I knew who did it for years stayed in the work because they loved it.

DC and a few other states now require daycare workers to get degrees. The research doesn’t show this improves outcomes for children below grade school level. But the requirement is there, citing it professionalizes the sector and improves quality of childcare.

The Washington Post staff wrote two pieces about DC’s decision.

One argued it’s probably a good idea. The title framed DC as a trailblazer: “D.C. among first in nation to require child-care workers to get college degrees.” It cited research that teachers need “literacy skills to introduce young learners to an expansive vocabulary.” The article focused on a daycare worker who was glad to get a degree.

The other Washington Post piece was a feel-good feature when the first class graduated.

Nobody talked to the workers who left.

The media almost never runs good-news stories about low-wage work. The one time they did was celebrating a policy that pushed people out of jobs they loved.

I think the Washington Post staff mean well but it comes off as tone-deaf to workers. Government grants cover the costs of the two-year degree. But the hurdle is often that you’re asking them to do coursework in a language they’re still learning. Many daycare workers, like my mom and my best friend’s mom, struggle with English skills. This makes it hard to navigate paperwork and grant applications. A degree takes a lot of time away from other paid or unpaid work they already do.

To be told you can’t do a job you’ve been doing well for years, sometimes decades, is hurtful.

Daycare workers end up switching out of childcare work into other roles. In my circles, daycare workers pivot in and from cashier, retail, or home health. Options tend to be limited by where you can get to with limited transportation options, language skills, etc.

Formal daycare is already out of reach for many American families. Informal childcare is the most common non-parental childcare. One third to one half of employed parents of kids under five rely on friends, family, and neighbors.

Requiring daycare workers to have degrees makes what looks like a luxury good, formal daycare, even more of a luxury good. It effectively outlaws cheaper versions of daycare.

Daycare workers see this. They also see how regulations put them in impossible positions daily. If a child falls and has a bad nosebleed, rules require washing hands and putting on gloves before applying pressure. That’s minutes of a child bleeding while you grab gloves that must be stored out of reach of kids.

Research shows that U.S. upper- and middle-class voices get their preferences represented in government at nearly equal rates. It’s working-class voices that consistently don’t (see Vox’s deep dive on the research).

Research found that:

Bills supported just by the rich but not the poor or middle class passed 38.5 percent of the time, and those supported by just the middle class passed 37.5 percent. But policies supported by the poor and no one else passed a mere 18.6 percent of the time. “These results suggest that the rich and middle are effective at blocking policies that the poor want,” the authors conclude.

It's hard to say how much media coverage shapes which bills pass. But I notice the Washington Post staff coverage centered on the perspective of credentialed professionals and someone becoming credentialed. The daycare workers who quietly moved on to other jobs just disappeared from the story.

Coverage of this issue makes me see how working class perspectives can be left out.

About 75 American universities promise to cover costs for students whose families can’t afford to pay. These guarantees have existed for years. But the information doesn’t reach the students who need it most.

I stumbled upon this list at 18, and it changed my life.

I had no college plan. State schools cost $10,000 a year, more debt than I could imagine repaying. I wasn’t surrounded by people with careers. The math didn’t feel real.

Then I found the list. Funding didn’t depend on becoming exceptional, just on getting in.

Suddenly trying in school felt rational. I got serious about academics, joined debate, student government. I applied to sixteen schools with no counselor, sometimes wondering if I was delusional for trusting the internet this much.

I got in. Three semesters after stumbling across that list, I flew to an out-of-state university.

Despite how much this changed my life, information about these promises doesn’t spread widely.

Major news outlets rarely cover these programs. Universities actively want to publicize them. They want strong applicant pools and higher percentages of Pell Grant recipients. So why doesn’t the information reach the students who need it?

The audience mismatch problem

Financial aid content tends to geared for higher-income families, people already planning for college who don’t qualify for need-based aid. Low-income students who've ruled out college as financially impossible aren't searching. They don't know to look.

Counseling offices often miss these students too. My high school counselor never mentioned fee waivers when I declined AP classes because of the test fees. Later, she pushed me toward early graduation. I would have had a Texas diploma, but not the coursework required for college.

These aid programs are still not common knowledge yet. AI models will tell you about them, though not right off the bat. When I tested three AI models asking how a low-income student should pay for college, two suggested community college, scholarships, and military service. Only ChatGPT immediately mentioned need-based guarantees. When I asked Claude why it didn’t, it said it didn’t want to instill false hope.

The problem isn’t the AI models. The internet itself doesn’t make this information easy to find.

Many non-competitive public universities offer these guarantees, though only to certain incomes and in-state residents.

The University of Texas at El Paso, for example, accepts nearly all applicants and covers full need for Texas families earning under $100,000. About 40 percent of Americans live in states with similar programs.

There’s no comprehensive public database of these smaller programs1.

The most visible lists focus on highly selective schools that offer aid to anyone at any income level. This version spreads more widely because it’s useful to a broader audience of college-bound students, but it’s not the list that would help the most students.

Why the information doesn’t spread

I wrote a book focused on less selective schools with these guarantees. Organizations serving low-income students bought it, but the information still spread primarily through intermediaries rather than directly to students or families.

What spreads widely online tends to be either outrage-inducing or politically useful. Financial aid guarantees are neither. They don’t fit neatly into partisan narratives about college affordability or student debt.

Small Wins in Information Access

While AI models might not flag these programs in its first response, AI models flag them during follow up questions. This is better than the status quo! It took me over six months of scholarship searching to stumble upon the same information.

For teenagers today, this is changing lives in ways headlines won’t capture.

AI isn’t a grand solution, but it’s something while we work on better systems.

Half the battle is noticing the disconnect exists. Universities with aid programs could advertise together, online and in public spaces. High schools could put income-based program posters on their walls. Counselors could mention them to more students, not just the “college track” kids.

Because that information itself can put students on the college track.

I don’t know where I’d be if I hadn’t stumbled across that list. University changed how I saw myself and what seemed possible. Going abroad, living in DC, working on policy, these stopped sounding like outlier dreams. It’s why I now still reach for crazy-sounding goals. Like writing on Substack.

I just hit 1,000 subscribers. Thanks for being a part of my writing dream!

But when asked how AI will affect them personally in 2026, they’re mostly optimistic. People expect AI to have a neutral or positive impact on their own lives.

The divide gets sharper when you separate users from non-users.1

People who use AI expect improvements across the board: in their own mental health and wellbeing, productivity, and ability to express themselves. Non-users expect the opposite: negative impacts on every metric.

This tracks with what people feel when they actually use AI.

People are more likely to report feeling curious, inspired, and empowered than to feel concerned or stressed. Some negative emotions show up, but the overall picture skews positive.

I personally find AI very empowering, especially for quick back-of-the-envelope estimates, literature reviews, and brainstorming.

But if individual AI users feel better off, why does the broader discourse feel so dire?

This happens all the time.

People consistently feel good about their own situation while believing everyone else is struggling.

When asked about their personal finances, 73-75% of Americans say they’re doing well, pretty consistently year after year. But ask about the local or national economy, and pessimism spikes a lot.

To be clear, questions about the state of national economy attract skewed responses2 because people tend to answer differently depending on if their party is in power.

But this pattern shows up in less partisan questions too.

In the UK, researchers asked people about local vs. national problems: teenage loitering, crime, littering, unemployment, graffiti, drug dealing. Every single time, people said the problem was milder in their neighborhood but severe nationally.

This pattern extends to happiness itself. Across 30+ countries, people massively underestimate how happy their fellow citizens are.

In South Korea, people guessed 25% of their fellow citizens were happy. The actual number was 90%.

In the United States, Americans guessed fewer than half their neighbors were happy. The reality is also 90%.

Why does this happen?

The most likely culprit is that negative news travels farther.

We experience stability in our daily lives. We see our neighbors, our coworkers, our routines. But our sense of the nation comes mostly through media, which systematically surfaces problems, conflicts, and fears.

No one designed this setup. But it leads to anxiety and a sense that things are falling apart even when they’re not.

And no clear way to fix the problem.

For AI specifically, this matters.

Part of the gap is probably the same media dynamic. It’s also true that people can reasonably believe AI will help them personally while worrying about larger effects! On jobs, concentration of power, or other risks they won’t experience directly.

Still, I think the fear-first framing carries costs, even when it’s well-intended.

Policymakers, with the their team of analysts, will see this type of less-alarming survey results. But the public won’t. The public will be frustrated by inaction on AI, without seeing that policymakers struggled to get traction, in part because of mixed data on AI impacts now.

More broadly, when public discourse doesn’t match people’s more positive experiences, it stresses people in a way that can’t be solved by policy alone. Also, subtler concerns that deserve attention get drowned out.

In sum, building support for smart policy gets a lot harder.

Last week I published a piece about how working-class people talk about Uber differently than the media does. It resonated with a lot of people. A few readers pushed back. They still thought Uber deserved special criticism.

This surprised me. When I think about the jobs I and my community have had, Uber doesn't stand out as particularly bad. It stands out as particularly visible.

Other low-wage jobs that get less attention:

Call center work. I called strangers to ask them to take surveys. If I didn’t get enough surveys completed, my shift would be cut early.

Manual day labor. A close friend used to line up at 6 am at a day labor office, hoping for 8 hours of work. He didn’t know whether or where he’d work until he was assigned.

Seasonal work. Another friend worked at an amusement park classified as seasonal, so the park legally paid him below federal minimum wage.

Direct sales. In high school, several friends sold kitchen knives door-to-door. A family member sold Avon cosmetics. These jobs have no hourly wage and no easy way to calculate hours spent.

These jobs look pretty undesirable. But each job had pros and cons that made it worth it to each person.

I loved that the call center hired me. I was underage, needed parental signature, and had no job experience. Places that pay at or below minimum wage have some kind of invisible perk like this.

Uber driving perks are flexible hours and not requiring English. DoorDash’s big perk is that you can have your children in the car.

Why does app-based gig work get so much hate?

Visibility. Everyone’s heard of Uber. News about Uber travels nationwide, unlike a local amusement park.

Optics. Small tech teams earn Bay Area salaries while managing workers earning far less.

Novelty bias. New things, like apartments or data centers, often have to justify themselves.

Media filters. Journalists rarely publish stories from gig workers with positive stories. News audiences prefer negative news solved by partisan solutions.

The outrage is well-meaning. But it comes from taking a narrow sample of low-wage work, being surprised at its nature, then blaming the most visible version.

What also gets lost in that framing is what workers themselves are asking for.

Proposals to mandate employer benefits sound great. But they’d increase the already-growing cut platforms take. Most gig workers wouldn’t qualify for benefits. 90% of DoorDash drivers and 48% of Uber drivers work fewer than 10 hours a week.

What many drivers actually want is to keep more of the fare.

That’s what Empower offered. It’s a rideshare app where drivers set their own rates, get 100% of the fare, and pay a flat monthly fee for the software. When DC tried to shut it down (over commercial liability insurance), drivers protested in favor of it.

71% of Uber drivers prefer independent contractor status. An app that lets them keep 100% of the fare is a genuinely different model from a worker’s perspective, even if it doesn’t look different to media or politicians.

Most coverage framed this as another tech company skirting regulations. It glossed over the part where drivers explain why they prefer this leaner structure and their push for a law that would let them operate independently, with their own commercial insurance, without working for a company like Uber. Current DC law doesn’t allow that.

The missing context

Low-wage workers talk to each other. They know which jobs are worth it. They have information that rarely makes it into news coverage.

Sometimes, their preferences don’t match the prevailing narratives. Media coverage rewards clean partisan framing. Worker demands that don't fit either side tend to disappear. Reason Magazine was the only outlet that covered the drivers' perspective on Empower. That's not because only libertarians care about workers. I think the story just didn't quite fit anywhere else.

Listening to workers won’t always lead to the right answer. But it gives you really important context that is not visible.

I waited tables for six years at $2.13 an hour. Now I'm a policy researcher in DC. These two worlds talk about technology completely differently. Let me use Uber driving as an example.

The critique: “Uber doesn’t offer health insurance”

Media coverage often frames Uber as exploitative because it doesn’t offer health insurance. But most jobs available to my family members (daycare worker, retail cashier, coat store clerk, construction) don’t offer health insurance either.

I didn’t get private health insurance until I became a diplomat at 23.

This isn’t ideal! But nobody in my world is angry at Uber for something that’s standard across hourly work, day labor, and shift work.

The bigger problem is that America ties health coverage to employment, and Medicaid is bare-bones.

The critique: “no guaranteed wage”

Another common criticism: gig work doesn’t guarantee minimum wage. But some of the best jobs in my community are ones without guaranteed wages because if you’re “fast” or “get good shifts”, you can earn far more.

When I was being advised by family and friends on jobs in high school and college, restaurant server was top of the list. In Texas, the tipped minimum wage is $2.13 an hour. It was still the most sought-after job because on good shifts, you could take home over a hundred dollars. I kept that job for six years.

The frustrations I do hear from my community

Uber requires a relatively new car (10-15 years old). This locks out members of my family unless they’re willing to take on a lease. This adds risk, cost, and rules out using Uber only occasionally to supplement your existing job.

The apps take large cuts of the fare. One driver told me about a ride that cost the customer $79, but he only got $19. That’s an outlier example, but cuts are frustratingly high. In California, roughly a third of each fare goes to state-mandated insurance requirements. These requirements were originally supported by Uber and Lyft as a way to gain legitimacy when rideshare was fighting to exist. Now those same mandates make it hard for more affordable, driver-friendly alternatives to enter the market.

What people appreciate

Uber and Lyft don’t require you to speak English. For immigrant communities like mine, this matters a lot!

Rideshare apps hire people with certain felony convictions, as long as seven years have passed. I have family members and friends with records. Getting any job afterward is hard. The seven-year policy is comparatively forgiving.

Shift workers rarely have regular schedules. As a restaurant worker, your supervisor might schedule you to work Monday to Wednesday then Saturday. The next week, you might work Wednesday to Saturday. This makes juggling the schedule of a second job hard. Apps that let you pick up work on short notice make life easier.

Why the disconnect exists

I don’t think mainstream media or journalists have bad intentions. I suspect it’s partly a supply-and-demand problem.

Most Americans spend less than a minute a day reading news online. The people who subscribe and read online articles skews toward groups with more high-paying employment options. Journalists, their editors, and their main readers don’t overlap much with the communities where gig work is seen as a good option.

In America, richer and poorer people are in separate social groups, mostly befriending within their social class. These worlds are disconnected.

Why this matters

For context, people earning minimum wage or below represent only about 1 million Americans out of 330 million. Uber has roughly 1 million U.S. drivers. These aren’t large chunks of the population.

But the pattern matters: when we evaluate new technologies, we often apply frameworks from people who have options that the actual users don’t have. This biases how we assess progress: the people most helped by an innovation are often least represented in debates about whether it’s good.

In 2018, a tuberculosis vaccine showed breakthrough results: over fifty percent efficacy. Seven years later, it’s still not deployed. Tuberculosis kills 1.5 million people every year.

In that same period, U.S. commercial aviation carried billions of passengers nearly sixteen years without a single fatal crash.1

Both systems involve complex technology, heavy regulation, and life-or-death stakes. One has become remarkably good at learning from failure. The other hasn’t.

What a culture of safety actually looks like

In 1991, air traffic controller Robin Wascher cleared a Boeing 737 to land on an occupied runway. Thirty-five people died.

She wasn’t fired. She was offered her job back.

The reasoning: punish mistakes and people hide them. Hidden mistakes can’t be fixed. And if one error caused a catastrophe, the system had already failed. Wascher was only the last link in a chain of broken safeguards.

So they fixed the safeguards. They changed how LAX sequenced runways, built a taller control tower, upgraded ground radar. Her error became data. The data made flying safer.

The FAA has its failures. But its safety culture still leads to decisions most regulators would never make.

The FAA lets infants fly on parents’ laps, even though turbulence kills roughly one baby per decade. Requiring a separate seat would push some families to drive instead, killing an estimated sixty babies per decade in highway crashes.

One baby death versus sixty. The FAA chose one.

What drug development gets wrong

Most regulators would never make that call. They optimize for deaths they’ll be blamed for, not total deaths.

As someone who’s been in government ten years, I know this is is not from malice. It’s often a structural problem. When an approved drug causes harm, there are congressional hearings. But lives saved from a faster pipeline are invisible, diffused across hospitals and countries where no one’s counting. Congress might set standards, but the court of public opinion sets risk tolerance, primarily based only on the clear harms.

The result is a system built for caution, not speed. In 2022, a four-person startup wrote their clinical trial application in 48 hours. The industry average is four months. The startup’s main innovation was running every step in parallel, rather than sequentially. Large pharmaceutical conglomerates dominate, managing portfolios where no single drug matters enough to rush. Things could move much faster from diverse entrants.

Why the problem compounds

Clinical trials have become so expensive that, for the most part, only large pharmaceutical companies can run them. That prices out nonprofits, universities, and governments, players who might prioritize global health over returns.

The Tuberculosis vaccine was developed by pharmaceutical company GSK using government and nonprofit funding. GSK also makes a shingles vaccine with the same core technology.

TB kills 1.5 million a year. Shingles kills fewer than a hundred.

GSK prioritized shingles because the U.S. Medicare system provides millions of insured patients. A tuberculosis vaccine would help far more people, but most of them live in low- and middle-income countries, which are not a lucrative market.

This decision to roll out the shingles vaccine over a tuberculosis vaccine is rational for GSK. GSK is a for-profit company accountable to shareholders, not global health.

This tells me our system desperately needs to get better.

Why aviation escaped this trap

When a plane crashes, the NTSB investigates independently. Reports are published and widely read. Over decades, this built a flywheel: transparent failure analysis, private sector collaboration, safety, public trust, and a culture that treats every death as a system design problem.

Drug development has no equivalent because, partly because success is not as obvious as millions of passengers landing safely. Additionally, the pharmaceutical companies that interface most with the FDA aren’t pushing for speed; expensive trials are a competitive moat.

How our system could be

A drug approval system that treats invisible deaths as urgent would, along with other things, prioritize experimentation at every level: adaptive clinical trials that update based on incoming data, conditional approvals that let drugs reach patients while evidence accumulates, flexible manufacturing rules that let academics test ideas without million-dollar facilities.

As I write this on December 31st, another year closes on that tuberculosis vaccine. Seven years. Ten million deaths. Aviation’s safety culture proves we know how to build systems that learn, but it worked partly because rule-makers fly on the planes they regulate. Drug development’s costs fall on people far from the conversation. Getting a better drug development system is possible, but will take much more targeted, concerted effort. I’m optimistic we can get there.

The most common fare I’ve paid on Empower over the last six months is $7.65.

For a recent trip from downtown to the airport, Uber wanted $32. Empower wanted $17.25.

I use it constantly, and so do a lot of car-less people I know. That price difference is a pretty big deal!

For many, it can be the difference between getting to the clinic or skipping an appointment. Between getting a ride after a night shift or walking home alone after buses stop running.

DC is trying to shut Empower down, primarily over liability insurance1. DC law requires $1 million in coverage per ride.

The $1 million requirement isn’t sized to typical accidents. When $100,000 is the limit available for an insurance claim, 96% of personal auto claims settle below $100,0002.

The high ceiling shifts incentives: plaintiffs' attorneys have reason to pursue cases they'd otherwise drop and push for larger settlements. Fraud rings have emerged to exploit these policies. The American Transit Insurance Company, which focuses on NY rideshare insurance, estimates 60-70 percent of its claims are fraudulent. Uber recently filed racketeering lawsuits against networks of law firms and clinics allegedly staging fake accidents in New York, Florida, and California.

That $1 million requirement traces back to Uber’s early days. When the company was fighting for legality across America, taxi commissions called ridesharing dangerous. To win over skeptical politicians, Uber proposed $1 million in coverage, matching limousine services and interstate charter bus companies, not taxis. It became the national template. Had Uber aimed to match taxi limits, the mandates would be $100,000 to $300,000.

Now Uber is advocating to lower the $1 million mandates. The company (and its drivers) complain that insurance is around 30% of fares, particularly in states like California, New Jersey, and New York which also require additional $1 million uninsured motorist coverage and/or no-fault insurance. Even in DC, with very strong anti-fraud protections3, the base $1 million requirement makes up about 5% of every fare—roughly a quarter of Empower’s advertised price advantage.

The missing market

High mandates also act as a moat. In DC, becoming a licensed rideshare company requires a $5,000 application fee, a $250,000 security fee, and infrastructure for that $1 million coverage. You have to be well-capitalized before you serve your first rider. This is likely why we see few bare-bones apps or local competitors to turn to when Lyft and Uber are surging.

Empower shows people want options. The app doesn’t let you schedule rides in advance, store multiple cards, or earn airline miles. Drivers don’t always turn off their music. Empower’s not trying to target the same audience as Uber. But the New York Times estimates Empower handles 10% of DC’s ride share market. People are comfortable with the rideshare industry’s scrappy options.

I think the core question is: now that society has accepted rideshare, should we revisit the rules that helped us get there?

Coverage of the potential shutdown rarely focuses on who stands to lose most: price-sensitive riders. Most coverage focuses on Empower’s lack of commercial insurance without explaining that the mandate is three to ten times higher than what taxis carry. Few explore whether or how Empower’s model actually differs: drivers can set their own prices. Drivers fund the platform through monthly fees rather than a cut of each fare. Drivers who get commercial insurance can also use it for private clients.4

I take the downsides of more lax regulations seriously. I write and gave a TEDx talk on traffic fatalities. More cars on the road isn’t inherently good. There are real costs from congestion, crashes, and pollution.

But I also care about prices for things people depend on.

People now trust and rely on this mode of transportation. Ridesharing has become pseudo-infrastructure for car-less Americans and a tool against drunk driving. In areas of Houston where rideshare first rolled out, drunk driving incidents appear to have dropped 38%.

We should want rideshare to remain affordable, especially as we build the excellent public transit we need.

When protection mandates price someone out of a ride entirely, they might drive drunk, leaving themselves and their victims without any $1 million policy. For those without cars, they might walk home at night despite feeling unsafe. There’s no million-dollar coverage for getting mugged on a dark street.

The people who suffer most are the ones who can least afford it.

I zoom in on the insurance question here, partly because few others have. For further sources on DC's Empower debate, this New York Times piece is the only coverage I've seen that also considers what low-income passengers and drivers stand to lose. Reason Magazine covers DC regulations’ impact on prices (1,2).

In 2022, two Nepali entrepreneurs wanted to open a clinic in their community. They’d serve patients in their native language. But the surrounding hospitals blocked them from opening.

Under “Certificate of Need” laws still active in thirty states, existing healthcare providers can veto new competitors.

The clinic never opened.

These laws work much like housing NIMBYism: incumbents blocking newcomers to protect the community.

The Original Logic

Certificate of Need laws emerged in the 1970s from a specific economic worry. If too many facilities opened, doctors with empty beds or appointment slots might recommend unnecessary treatments to stay busy.

This wasn’t unfounded. The fee-for-service payment model does create incentives for unnecessary procedures.

But the evidence didn’t support the broader concern. The federal government repealed its version of Certificate of Need decades ago after realizing the worry about oversupply was overblown. Sixteen states followed.

Thirty kept the laws.

Advocates for these laws argue they serve a purpose. In very rural areas, two clinics might not both survive. Allowing one provider to reach higher volumes might mean higher-quality specialized procedures.

But the outcomes in those thirty states show patients have access to fewer facilities, pay more for care, and wait longer for treatment. Meta-analyses show higher mortality rates for heart attacks, heart failure, diabetes, influenza, and Alzheimer’s disease in Certificate of Need states. Every presidential administration since Reagan has called for repeal.

But CON laws are just one layer.

The System That Ate Primary Care

When your parents needed a doctor, they probably went to an independent practice. That world has nearly disappeared. Independent doctor-owned clinics have been replaced by large hospital systems on one end and urgent care centers on the other. The middle vanished.

A web of regulations and payment rules made independent practice unviable.

Insurance reimbursement favors size. Large hospital systems negotiate higher rates than independent doctors can. A solo primary care physician gets paid less for the same appointment than a physician employed by a hospital network. Small practices are harder to run profitably.

Medicare’s payment rules accelerate consolidation. When a hospital system acquires an independent practice, it can immediately charge more for identical services under “hospital outpatient” billing rules. Same doctor, same exam room, higher bill. This creates incentives: hospitals buy practices, revenue increases, the network expands.

Specialists control the pricing mechanisms. The American Medical Association, dominated by specialists rather than primary care physicians, controls the medical coding system that determines Medicare payments. A cardiologist performing a fifteen-minute procedure earns significantly more than a primary care doctor spending thirty minutes managing a diabetic patient’s complex medication regimen. While specialists require longer training and bear greater liability, the gap exceeds these differences. This also causes an incentive loop: fewer doctors choose primary care. AMA’s core membership continues to be majority specialists.

The result: Americans lack accessible primary care clinics. People skip preventive visits and routine care, then turn to urgent care centers and emergency rooms, the most expensive and least effective entry points into the healthcare system.

Doctors Are Scarce By Design

The shortage of independent clinics gets worse because there aren’t enough doctors. When physicians are scarce, their salaries rise. Hospitals can afford these higher salaries. Small independent practices often cannot. So scarcity pushes doctors toward hospital employment in two ways: better pay plus administrative support, versus lower income and uncertain finances in independent practice.

Some main drivers of this artificial scarcity:

1. Medicare caps residency positions. The federal government funds residency training, the required years of supervised practice after medical school. This funding has been capped near 1996 levels. The US population has grown and aged since then. The number of training spots has not.

The original justification was reasonable: federal funding shouldn’t subsidize unlimited training positions. But the cap was never indexed to population growth or healthcare needs.

In 2022, over 3,300 medical school graduates couldn’t match with a residency because there weren’t enough slots.

Hospitals can pay for extra residency slots themselves, but it’s expensive. Only wealthy institutions can afford it. These hospitals cluster in already well-served areas, especially the Northeast. New doctors train where there are already plenty of physicians, then often stay where they trained. The geographic imbalance gets worse.

2. About 100,000 physicians trained abroad live in the US but can’t practice. They cannot practice medicine here because they haven’t completed a US or Canadian residency program.

The stated justification is quality control. Medical training standards vary globally.

But the requirements are so rigid that even excellent physicians trained at top international medical schools can’t practice in the US. Some states are beginning to create alternative pathways. Progress is slow. Part of the reason is that foreign-trained physicians increase supply, which threatens incomes for doctors already practicing here.

What We’ve Lost

I think about my own Mexican American relatives who would have gotten care sooner if they’d had access to doctors who spoke their language and shared their culture. The distrust many immigrant communities feel toward healthcare is real, but there are solutions.

These structural problems also make it hard for technology to fix things. AI that helps doctors process billing paperwork will probably get adopted. Doctors spend roughly half their time on paperwork. But AI tools that improve diagnostic accuracy or prevent emergencies eat into revenue under fee-for-service payment. The system rewards volume, not outcomes. But the bigger problem is the artificial scarcity itself: we need more doctors and more clinics, and the structures above prevent both from existing.

Artificial scarcity in healthcare makes the whole system seem stuck, hopeless and unfixable. But a lot of these are fixable. Policy choices like these can be reversed.

This essay was inspired by our last DC Abundance gathering, which featured Niskanen Center Fellow . See Lawson’s full report on Health Abundance. Sign up here for our next DC Abundance meetup. All errors my own.

If you know ambitious high schoolers interested in tackling healthcare challenges, check out Progress in Medicine, a summer program that shows teens how past medical heroes solved seemingly impossible problems and helps them figure out how they could contribute to solving today’s problems.

Noah had always wanted something like AI, a tool that could help us think, create, and solve problems. Yet, surveys show Americans are far more pessimistic about AI than people in China, Korea, and most other countries.

I think I know why.

When Beliefs Become Signals

Over the past decade, tech skepticism has become strongly associated with the American left, especially among educated progressives. And when a belief becomes a marker of being informed and progressive, people anchor toward that belief to signal group membership. This effect can be so strong that it trumps even their own experience.

For example, when a Republican is president, Democrats report that economic conditions in the country are worse. Once a Democrat comes to power, economic conditions get better very quickly, according to surveys. When a Democrat is president, Republicans do the same. People’s real-life spending and employment experiences don’t match what they report on surveys.

On climate change, an Avaaz survey that polled 10,000 people between the ages of 16 and 25 found that over half of respondents expect climate change to “doom” humanity. Yet Gen Z are still enrolling in universities and planning careers. And despite being framed as humanity’s greatest threat, climate change consistently fails to crack the top ten voting issues, particularly among lower-income Democrats and non-Democrats.

Researchers call this “expressive responding” or “partisan cheerleading” where people signal loyalty to their political team on surveys rather than reporting their actual beliefs. It’s most common among highly engaged partisans who know which answers mark them as informed members of their tribe.

And I suspect something similar is influencing how people respond about AI.

Beliefs as Identity Markers

Americans express deep concerns about AI in surveys. But like climate change, AI is also not a top-ten issue for voters. Expressive responding could contribute to this if the left has an anti-AI or anti-tech stance. Since roughly half of Americans lean left, a shift in how the left talks about tech can reshape national sentiment.

The Opinions Editor of The Guardian notes that the Left is increasingly defined as “fearful, agnostic, or hostile to technology.” This mindset fuels, for example, the “degrowth” movement, which views technology as the cause of the climate crisis rather than a source of green energy solutions.

This shift has been particularly visible in media coverage. Matt Yglesias, reflecting on the New York Times' shift, said on X that the paper made a deliberate editorial decision to cover tech with “a very tough investigative lens” that was “highly oppositional at all times and occasionally unfair.” Journalist Kelsey Piper confirmed she’d heard directly from Times reporters that there was a top-down directive that tech could not be covered positively, even when there was a true, newsworthy, and positive story.

You can see this dynamic in how some concerns about AI spread. Progressive circles have embraced the idea that individuals using ChatGPT is environmentally irresponsible because of water consumption. When researcher Andy Masley pointed out that 2,500 ChatGPT prompts use roughly the same water as making a single sheet of paper, he faced some backlash. Some people weren’t relieved the problem was smaller than reported. They were defensive.

To me, this was a sign that this had become an identity belief, meant to be protected rather than examined.

What Gets Lost

When skepticism becomes identity, it’s harder to see the full picture. AI is delivering benefits its millions of users, although these are hard to track in GDP.

I’ve experienced AI’s intangible welfare improvements. AI makes me a more confident writer. I red-team my drafts from different perspectives before posting. When I was confused why my sore throat turned into a cough last week, ChatGPT explained that coughs often follow as inflammation clears. Google’s SEO-optimized medical results weren’t useful.

Researchers have tried to measure these values. One study found the average US user would need to be paid $98 to give up ChatGPT for a month. This would point to up to $97 billion in consumer surplus, based on 2024 models. That’s likely an overestimate because people demand more to prevent losing something than they’d pay to get it (loss aversion bias). But even accounting for that bias, there’s clear value above the $0-$20 users actually pay.

To be clear, I’m not saying AI’s downsides don’t matter. AI can pose challenges around labor displacement, concentration of power, cyber and bioweapons risks, etc. I care about them enough to spend my time working on those risks.

But I want our discourse to be nuanced and high-quality, so we don’t get sidetracked prioritizing the wrong risks. Or thinking that you have to be all in on AI or against it. We can be excited about new tools while pushing for them to be developed responsibly.

And we do that best when we can talk about AI openly, honestly, and in good faith.

EDIT: To be clear, I think this dynamic explains 15%-50% of the reason why Americans are responding to surveys with such pessimism. It explains why we’re outliers but not the entirety of the pessimism.

Some people expect AI to eliminate writers. After all, if AI can generate great text, why would we need humans putting words on pages?

This misunderstands what writing actually is and how people consume it.

How People Actually Read

To see why AI won’t simply replace writers, start with how people actually discover what to read.

Yes, AI will change how we browse the internet. AI might become many people’s interface with the web. There will be less Googling for articles, less stumbling onto whatever writer the algorithm serves up.

But that’s not how most reading happens anyway. Most reading is curation by people you trust.

I didn’t Google to find the Abundance book, progress studies, nor my favorite writing book. These were flagged to me by writers I trust. I followed their judgment.

This pattern has been stable throughout internet history: Patreon, Substack, traditional subscriptions. People follow and pay for writing because they want to support specific institutions, angles, or voices they trust and want to amplify. Today’s information landscape depends on these structures.

And they’re proving more resilient than expected. Many predicted AI-generated deepfakes would destroy our ability to trust media, that we wouldn’t be able to tell real from fake, that information ecosystems would collapse. That didn’t happen. Rather than drowning in billions of deepfakes and AI slop articles, people still get their news from institutions and voices they’ve followed over time. Trust structures held.

Even Substack writers, who lack an institutional home, build audiences the same way: one piece at a time, earning trust through consistent judgment about what matters.

This trust is what AI needs to displace, not just text generation capability.

What Makes Writing Stick

The writing that builds trust is, at its core, about connection.

Writers write about what lights them up. Writers build audiences made up of people who light up from the same things. And, often, the writing resonates because readers share some of the writer’s “implicit knowledge” about what matters.

Implicit knowledge is information you possess but can’t easily explain, often learned from experience or social interactions. A developer knows which architectural decisions will cause problems six months later. A longtime transit advocate will know what energizes her local allies the most. It’s accumulated pattern recognition from thousands of experiences.

AI struggles with implicit knowledge. Anthropic points to this as a primary bottleneck to AI utility and adoption. This knowledge lives “in the minds of account executives, marketers, and external contacts” and can’t be easily transferred into databases to train or inform AI.

But implicit knowledge isn’t just a business problem. It’s how audiences form. Consider Matt Yglesias, the center-left political writer. His readers don’t just agree with his positions. They sense something beneath the arguments, a worldview or set of instincts they share but couldn’t quite name until they found his work. I recognize my own unspoken thinking in his words.

This is what makes writing powerful: it transforms the implicit into the explicit.

“The Housing Theory of Everything,” published in Works in Progress, did this for me too. The article synthesized ideas about zoning into one argument that reorganized how I understood the world. I needed a day just to process it. The authors took knowledge that existed nowhere else in that form—part explicit research, part implicit intuition about how systems connect—and made it legible.

This is also how trust gets built. Ezra Klein and Derek Thompson spent years conducting interviews, writing essays, following their curiosity across topics. By the time they published Abundance, readers had watched them think in public long enough to know: these writers share my implicit intuitions about how the world works.

That accumulated trust is what lets them challenge prevailing opinions and actually be heard.

Why AI Doesn’t Clearly Replace the Writer’s Role

Even in a world where AI solves many problems, we’ll still have genuine disagreements. People will likely still disagree about whether abortion is murder, which drugs should be legal, and whether San Francisco should have skyscrapers. They’re contests between worldviews.

Someone still has to decide which worldview to advance. AI doesn’t wake up thinking “here’s why everything bagel liberalism is the problem.” It has no skin in the game, no convictions about what message matters.

But choosing the message isn’t enough. For writing to shift how people think, rather than just collect engagement, you need the kind of work that uncovers neglected topics, explains complex issues clearly enough to change minds, earns trust through balance and rigor. Alarmism gets clicks. Persuasion requires something harder.

This difficulty explains why great writers increasingly resemble senior software engineers, who became more valuable as AI learned to write code.

AI handles the routine, repetitive code. Senior engineers provide what AI can’t: knowing which approach actually fits the problem, understanding unwritten context from years of experience, making judgment calls.

The parallel holds for writing. AI generates clean text on any topic. But the writing that shifts worldviews depends on judgment, context, implicit knowledge, the skills AI struggles with most.

Writing will still be a tough market. But I expect writers will still get cited, still get subscribed to, still change minds. The skills that make writing powerful are precisely the ones AI isn’t replicating.

What I’m Very Uncertain About

Maybe AI solves implicit knowledge transfer soon. Maybe it learns to recognize which new ideas deserve synthesis. Maybe smart glasses capturing everything we see and say provide enough context to close the gap.

I don’t think it’s impossible. But these capabilities might prove much harder than language generation did.

AI excels at tasks with clear evaluation metrics: solve this equation, debug this code, summarize this document. Clear metrics are critical for training AI.

The skills writers rely on don’t have clear metrics. Knowing which implicit knowledge to surface and which topics will shift how people think are harder to grade quickly or objectively. Engagement metrics exist, but optimizing mostly for likes and comments rewards outrage over the harder work of informing and persuading.

The writing that changes worldviews requires something harder to measure. And thus harder to train AI on.

I don’t know what being a writer will mean in ten or twenty years. But the evidence isn’t there yet to assume the worst, that writers will be obsolete1.