Tolerance is built up over time. It’s also lost.

I hated spicy food until my early 20s. It tasted bad, so why eat it? Same bucket as horror movies. No appeal, and probably bad for you. Even worse, spicy food kills your taste buds, so the more you eat, the more spice you need to taste anything. A slippery slope.

Then I read about why spicy food feels good. It triggers an endorphin release. There are other claimed benefits (cardiovascular, metabolic, lower all-cause mortality), though the studies are observational, so hard to say. I don’t seek it out, but I eat it now.

I used to drink caffeine every day. Now I don’t. Caffeine is a hardcore, addictive psychoactive drug that our entire society is cool with. For most people, it just means more energy and alertness at certain hours of the day. I crash hard and metabolize it slowly, which wrecks my sleep. Caffeine’s a wonder drug for many friends and family.

If I have a cup of white tea in the evening, I’ll pull an all-nighter. When I drank caffeine regularly, I needed at least 10x that to stay up all night.

As a millennial, it’s wild to look back at what we ate as kids. Breakfast was dessert. Lucky Charms, Cocoa Puffs, Pop-Tarts, Cookie Crisp. Our Minute Maid juices had as much sugar as soda. Insane amounts of sugar were normal in the 90s and 00s. In high school, I thought of Jamba Juice as the healthy option. A few years ago I bought a 28oz Green Machine at JFK. My whole mouth stung. I looked it up: 78 grams of sugar. I threw it away. My sugar tolerance has thankfully plummeted.

Life Tolerance

Things that start out hard and awkward become normal. I remember my first business calls with strangers. I was nervous. Beforehand, I mapped out what they might say and how I’d respond to every branch. Now I take those calls every day and read up on who I’m talking to while dialing in.

This goes for all work. Intros, interviews, pitches, follow-ups, firing people. Think about how much of what you do every day would have terrified a past version of you. For me, it’s most of my day. I remember when the thought of driving was terrifying.

It goes the other way too. There are things I did as a kid that I’ve lost all tolerance for. I couldn’t sit through a class again. School always felt like child prison, but I tolerated it because I had to. I sat through classes reading a book slowly, out loud with classmates that I could have read quickly at home. I studied for tests that entailed rote memorization of inane trivia.

Buildup

People don’t tolerate sudden change. They tolerate slow change. Resistance builds and the Overton window moves. It’s why city life shocks outsiders. Someone visits a big city and wonders how residents tolerate homeless drug addicts on the streets who are a danger to themselves and everyone else. It happened slowly. Nobody woke up one morning to a flood of homeless people. It built up over time, and so did the tolerance.

We become okay with bad things. We tolerate them. That’s part of life. We’ll always have things we have to tolerate, and we don’t even notice most of them. That’s where technology comes in.

Something stops looking like a problem once you’re used to it. Few people complained that their phone was bad at email and games. Then the iPhone came out.

Most of what you tolerate today will look like a flip phone in ten years. You just can’t see which parts yet.

Subscribe now

Most of what I write is to help me better understand and process why we do what we do. To understand others and ultimately myself. And yet, some things are too hard to fully process. Unfathomably unexpected. They’re just sad.

In The Whole Story, John Mackey talks about his journey with his father’s dementia and eventual passing. There were plenty of moments of joy along the way but painful moments as well.

When remembering someone, remember who they were to you. Don’t have the last months or years define them but don’t completely erase those either.

The memories, the lessons learned. It’s ok to be sad. To sit with that. Unbearably sad things happen. Sans terminal illness, the most terrible things that happen are the most unexpected.

When they do, being state oriented is the worst. How do I feel? Awful. I’d have to be sick to not feel awful. But that doesn’t make it better.

If you’re like me, your mind will push you to analyze your current state and continue asking why. As with any advice, it depends on your personality. But if you’re reading this, you’re likely like me and ask why. Much can be gleaned from asking why.

We used to perform 5 Whys when something critical went wrong. Each answer of why something went wrong is an assumption by the investigator. And when a system is hardened, failures tend to be multi-variable. Because of this, teams have moved toward contributing-factor analysis and blameless postmortems. Blameless postmortems.

I recently finished The Inner Game of Tennis, and this rhymes. Negative self-talk after messing up doesn’t improve performance. Observe, learn, iterate. You’re not your padel game. The best players “play out of their mind” and miss when they become self-aware.

At the core of it, if something goes wrong, it can be anything. Maybe the system failed. Maybe the failure is a feature, not a bug of the system.

It used to be that it didn’t make sense to design a system to handle so many extreme edge cases. However, now with AI, edge cases can be written into policy and soon, systems can handle everything.

And yet, we’re still human. Humans will always make mistakes and have “bugs.” Deterministic software may one day solve most of our problems. Humans will remain human. Doing human things.

Sometimes there’s nothing to say. Nothing to analyze or do. A terrible thing happened. It happened. You think to message them. But they’re gone. Forever.

Life freezes. It feels that way. But without you pulling it, the tide comes in. Without you pushing it, the world still spins. Entropy continues.

Wallow. Feel the range of emotions life gives us. You’ve mourned. You’ll remember them, their life, and how they impacted you. It’ll hit you in spurts. They’ll always be with you… now what?

Onward.

Subscribe now

AI keeps evolving. This has happened before with knowledge workers’ most widely used tool.

Google

The most useful thing I did in my early 20s was take an online course Google offered on how to use Google. I learned how to find relevant websites, search email, and even find products and images that others couldn’t. I did this by using the dozens of little features Google quietly put into search that almost no one utilized. The basics include typing has:attachment in Gmail to show only emails with attachments. Searching within a single site with site:, or putting - before a word to exclude results that contain it. If you’re searching for a wedding band (the kind that plays music, not the kind you wear), you can put -ring to filter out jewelry.

These skills let me quickly find answers to any question.

Eventually, Google kept improving to the point where you could just write exactly what you’re looking for instead of having to be clever with searches. Searches are so fast that you can write exactly what you want first, then add modifiers only if you need to narrow it down.

AI

AI is now undergoing the same evolution. Before, you had to be a bit clever around prompts with what you ask and how you ask it. That’s becoming less relevant. I wrote how I use AI last year. How I use it now has drastically changed, and it’s only been a year.

For context, it took Google over a decade for me to notice a significant change. It took AI less than a year, and the pace shows no signs of slowing.

In screenwriting, when you have writer’s block, a common technique is to write the subtext. AI is getting to the point where you can purely write the subtext. Writing the subtext is going to become more and more powerful.

Simply say what you want and why you want it. AI will do the task. If you don’t tell it why, it’ll make “stupid” mistakes. AI is getting better at understanding you with the growing context window, but stating why for a task or search is still important to keep it grounded.

There are examples of letting AI run completely unbounded: you ask it to fix all the bugs, and it deletes the codebase because that “fixed all the bugs.” That’s a rare, extreme example but shows how “The Why” matters. How I structure prompts:

1. What I’m trying to accomplish

2. Why I’m doing it

3. (Optional) Provide the framework (how to think about an area)

Sometimes, I’m asking about an industry where I don’t know the frameworks, so I rely on the AI to supply them. Soon enough, the AI will know most frameworks, so step 3 will become unnecessary for most.

A quick example: “Shorten this email” forces the AI to guess what matters, and it’ll guess wrong half the time. “Shorten this email. The recipient is busy and only needs to approve the budget” tells it exactly what to protect. Same with “review this contract” versus “Review this contract. I’m the buyer, and I care most about liability after closing.” The first gets you a laundry list of things you mostly don’t care about. The second reads like it came from your lawyer. The why does the work of a hundred instructions.

Underestimating AI

It’s fun to talk to people outside the tech sector and get their thoughts on AI. I was chatting with a young, top lawyer who thinks AI can never do what lawyers do. He said it’s because AI lacks the frameworks and high-level understanding of law.

His exposure to AI is dealing with AI slop from overzealous prosecutors. From his vantage point, AI is garbage and gets things wrong all the time.

From my perspective, law seems like one of the easiest things for AI to solve, since the frameworks and case law are known. You just need the model to ingest all the case law and have the proper frameworks to accomplish a legal task. The large model companies should all be able to solve law. Unbounded, it will hallucinate, but with the right context window, full case-law history, and proper frameworks, it’s solvable.

Future

Maybe AI gets so good that it thinks for us and solves all intellectual problems. If so, this skill set of using AI will be useless in five or ten years. I have my doubts. Either way, it’s worth the deliberate practice in the meantime.

Subscribe now

Ask someone to name the most reliably Democratic group in America. Most people guess Black voters. Some guess Jews.

The answer is atheists. Atheists are the most Democratic group in America. Per Pew, 84% identify with or lean Democratic. Historically, Black Protestants come in at 72%, Jews at 66%.

The mirror image on the right is White Evangelicals, 80% Republican. The two most partisan blocs in the country are the born-again and the godless. In 2024, 64% of monthly-plus attenders voted Trump. One in five Democrats are now atheist or agnostic. One in twenty Republicans are.

Black voters look like the counterexample: the most churched bloc in the Democratic coalition, voting 83% left. But the rule works within groups, and history works across them. The Black church was the organizing infrastructure of the civil rights movement and stays loyal to the Democratic Party. Black partisanship functions on ethnicity. Poll Black Democrats on abortion or gender, and their views are to the right of White Democrats. They vote left anyway.

Now Jews, since this is where the Internet gets stupid.

The conspiracy version says the left is run by Jews. The data says Jews are 2% of the country and sit at both poles. The New Left was disproportionately Jewish. So was neoconservatism. The Ashkenazi mean IQ is around 112, and the culture prizes questioning, so you get a population smarter and more disagreeable than average: more successful, more abrasive, and overrepresented among the loud on every side. Jews don’t run the left or the right, but they staff the debate team of both.

The Jewish Vote

So why does the average Jew vote left? The average Jew sits in the 100 to 120 IQ band, and that band votes left everywhere, Jewish or not. It’s the midwit band: smart enough to absorb the university’s politics, not independent enough to question them, and convinced that being left is what makes you a good person.

You can tell it’s a class effect because the more Jewish someone is, the more conservative they vote. Reform Jews went 84% for Harris. Conservative Jews 75%. Unaffiliated 70%. Orthodox Jews went 74% for Trump, and 75% had already identified as Republican in 2020, three years before October 7. Harris got 11% in Lakewood, the big Orthodox town in New Jersey.

The gradient is that clean because “the Jewish vote” is two different votes. We call people Jews who are atheists. Pew says 27% of American Jews are Jews of no religion: ethnically Jewish, religiously nothing. They vote like atheists because they are atheists. Orthodox Jews vote like evangelicals because they live like evangelicals: big families, religious schools, traditional roles, communal authority.

The famous 70-30 Jewish split is just the weighted average of a secular professional class and a religious counterculture, and it hasn’t moved in a decade. 70% Democratic in 2012, 70% in 2016, 71% in 2024, Gaza and the campus encampments notwithstanding.

None of this is because atheism implies leftism. Ayn Rand was an atheist. Half the tech right is. In other countries, secularists scatter across the spectrum. Parties change over time.

The old left had a working class that wanted wages. The new left expanded their tent to the third world, and grievance studies majors who got indoctrinated into socialism and cultural relativism at universities. Some atheists have noticed. Trump’s atheist share went from 11% in 2020 to 15% in 2024. If that continues, atheists will eventually split by class like everyone else.

Looking Forward

The last thing the numbers say is who wins. Orthodox Jews are 9% of American Jewry but 17% of Jews under 30. Two-thirds of them are Haredi, and the average Haredi family has around six kids. Reform families have 1.4. The Jewish vote will move right for decades without a single mind changing, and the same arithmetic runs through every religious group in the country. The believers are having children, and the seculars win converts.

Which is the point of the quotation marks. There’s no Jewish vote. There’s a secular professional vote that happens to contain most Jews, and a religious vote that contains the rest.

When “the Jewish vote” does flip red around mid-century, remember that nobody flipped it. The pundits will announce a great realignment. There won’t have been one. The polls are lagging indicators. The immigration numbers and maternity ward are leading indicators.

Subscribe now

Fremont, CA is ranked as America’s happiest city, for the seventh year in a row. I’ve lived in a dozen cities. I grew up in Fremont. From a happiness standpoint, it’s in the bottom half of cities I’ve lived in. And that’s being generous.

Kind, brilliant people come out of Fremont and go on to do great things. But “happy” isn’t a word that comes up when I think of the place. I sooner think of tiger moms and their children attempting to live up to expectations.

Like most social things, happiness is contagious. Happy people have happy friends. As I wrote before:

Happy friends make you happier. If your friend is happy, you’re ~15% more likely to be happy, and if your friend of a friend is happy, you’re ~10% more likely to be happy. To put that in perspective, for the average person, a $10,000 raise increases happiness by ~2%.

I’ve spent time with experts who write about which countries or cities are happiest. They’re typically lacking happiness.

Methodology

The ranking is WalletHub’s. It grades the 182 largest US cities on 29 metrics in three buckets: emotional and physical wellbeing gets 50 points, income and employment gets 25, community and environment gets 25.

No resident gets asked anything. The study stacks statistics: depression and suicide rates, adequate sleep, sports participation, marijuana use, opioid prescriptions, life expectancy, food insecurity, income growth, the share of households earning above $75,000, poverty, job satisfaction, Glassdoor ratings on local job openings, unemployment, bankruptcies, work hours, commute times, divorce, hate crimes, weather, parkland acreage, leisure time, a community pride index, and WalletHub’s own ranking of the most caring cities. A Life-Satisfaction Index worth 3 of the 100 points, which is by WalletHub’s own description a composite of physical, financial, and social health risks.

Arbitrary Weights

Depression, sleep, and food security all matter. The problem is with the measurements. Sleep (6.35 points) counts twice as much as suicide (3.17 points). A well-rested city with a high suicide rate outscores a sleep-deprived city where nobody kills themselves.

Past-month marijuana use dings a city at 1.6 points, while opioid prescriptions are at 0.8 points. Leisure time counts double but only exists at the state level, so Fremont gets credit for how much Californians relax.

Demographic Misunderstanding

Each metric only correlates with happiness when everything else is held fixed. Take the depression rate, which counts double and where Fremont ranks seventh-lowest in the country. The underlying CDC number comes from one survey question: “has a doctor ever told you that you have a depressive disorder?” It measures diagnosis. Getting diagnosed requires going to a doctor and saying something out loud. Fremont is 64% Asian, and Asian American adults are the least likely group in the country to receive mental health treatment: 25.4% of Asians with a mental illness got care in 2021, versus 52.4% of white adults. If you don’t get tested, you can’t get diagnosed.

And take divorce. Fremont has the lowest separation and divorce rate in America at 9.3%, and WalletHub scores that as happiness. Asian Americans have the lowest divorce rates of any racial group. A low divorce rate can mean strong marriages. It can also mean a culture where leaving is failure and failure is unacceptable. I had friends growing up whose parents didn’t live together, and they both were seeing other people. And yet, they weren’t divorced. They’d never get divorced.

What’s Not Measured

Beyond the flawed metrics, the bigger issue is what the survey didn’t measure. Not one of the 29 metrics measured whether anyone in Fremont would describe their friends as happy. Which is typically the best way to tell reality– it’s how the Polymarket gambler who bet Trump would win the 2024 election and the popular vote won. It was based on a survey asking people who they think others will vote for.

And nothing adjusted for cost of living. The income bar is a flat $75,000 nationwide, borrowed from a 2010 Kahneman and Deaton paper claiming happiness stops rising past that number. Kahneman himself said that’s wrong in 2023: the combined data showed happiness keeps climbing with income for most people, and the plateau shows up only for the unhappiest 20% (people dealing with things money can’t fix) at around $100,000, which is just the old threshold adjusted for inflation.

WalletHub still runs the disproven $75K number. Nearly 80% of Fremont households clear $75K, the highest share in the country. With Bay Area real estate skyrocketing, the average Fremont home now costs $1.5 million. A household at WalletHub’s happiness-maximizing income is 20 years of gross pay from the average house.

Measuring Happiness

So, is there a way to measure happiness? Should we even be optimizing for it as a society?

Happiness is fleeting. Fulfillment lasts. One thing I appreciate about Fremont is how hard everyone worked. Kids grind through spelling bees, piano, violin, whatever their parents pointed them at, and most will never touch the instrument again. Doesn’t matter. Going deep on one thing and working incredibly hard at it is a horizontal skill set. It transfers to everything. Many never develop it. Those who can’t grind can’t do anything lastingly meaningful in life.

That grinding isn’t happy. It’s miserable. You learn to tolerate and appreciate it. Then turn it on when you need to. That’s what it takes to make an impact.

Bhutan coined Gross National Happiness in 1972 and runs government policy through it. The last time the World Happiness Report ranked Bhutan, it came in 95th out of 156 countries. GDP isn’t a great metric. It’s still better than some made-up happiness metric.

Subscribe now

Robin Dunbar recorded people talking in shopping malls and found that roughly two-thirds of conversation time goes to social topics. Other people, mostly. Who’s sleeping with whom, who got promoted, who stiffed the group on the check. A later study using body-worn recorders clocked it at 52 minutes a day.

Almost an hour of gossip a day doesn’t survive a few hundred thousand years by accident.

I try to remove all parasocial relationships. The few I had have passed away. So this isn’t a defense of celebrity gossip. That’s a low-risk, generally boring way to build rapport with strangers, and it costs nothing because the target will never hear it.

Gossip about people you’ll see again is a different thing. It’s negative, it’s draining, and the warning you got about it as a kid was mostly earned. It’s also the only enforcement mechanism most groups have.

Karens and bullies serve a purpose. They enforce norms we have for a reason. Gossip does the same job. If someone is a liar, others should know. If someone stiffs his contractors, or sleeps with his direct reports, or borrows money and never repays, others should know. If nobody ever talked about anyone behind their back, the cheaters would run the world. It’s the Invention of Lying problem (movie where only one person in the world could lie). Defection is the dominant strategy when it’s undetectable.

This was tested at Berkeley. Participants watched an economic trust game, saw one player cheating, and were then given the chance to pass a warning note to the next person about to play against the cheater. The note cost real money out of their own winnings. They paid. The higher someone scored on altruism beforehand, the more likely they were to pay. Watching the cheating made people agitated, and sending the warning made the agitation go away.

The urge to tell someone is a warning system.

Gossip Suppression

The cleanest argument for gossip is watching someone successfully stamp it out.

Elizabeth Holmes ran Theranos on information control. Engineers weren’t allowed to talk across teams. Departures came with NDAs. Tyler Shultz complained to New York’s health department under a fake name, watched it disappear into an inbox, and then spent two years being followed by private investigators and pressured by lawyers to name the other whistleblowers. His family spent more than $400,000 in legal fees. His grandfather sat on the Theranos board and sided with Holmes.

No formal system stopped her. The board was Kissinger, Mattis, and Shultz, three men with no diagnostics expertise between them and enormous amounts of prestige. Walgreens ran diligence and signed.

What finally ended it was a reporter convincing former employees to gossip. It took twelve years because the informal channel had been lawyered into silence.

Absolutes

Pacifism is the most privileged position available. It’s the Hobbits in the Shire, safe because Rangers they’ve never met are patrolling a border they don’t know exists. Refusing to ever gossip is the same posture. Your hands stay clean because someone else is doing the warning.

Always forgiving is the same trade. Forgive everything, and you’ve taught people bad behavior is free.

Consequences

I was on a red-eye last week where a guy left his reading light on with the whole cabin trying to sleep. People murmured. A few glared. The woman behind him stood up and angled the light off her face. So he turned on more lights.

Nothing happened to him. He’ll never see any of us again, and not one of us can tell a single person whose opinion he cares about.

I used to think a social credit score was dystopian. We’ve all seen the Black Mirror episode. I mostly still think that. But sitting nearby, the case for a number stapled to that man’s name felt just.

The reason it doesn’t work is that a single legible metric gets gamed the moment it carries weight. We already run a small version. Uber deactivates drivers who slip into the mid-4s, so riders hand out five stars reflexively and only 2-3% of drivers ever fall below the line. Amazon and Yelp get astroturfed. Any score people can see and act on becomes a score people optimize instead of a score that measures.

Gossip resists this because it’s thousands of overlapping private assessments, each weighted by how much the listener trusts the speaker. Reputation laundering works on institutions, as the Theranos board demonstrated, and it works terribly on the twenty people who have actually worked with you.

Golden Mean

A society drowning in gossip is miserable. A society with none is worse. Min gossip and the social climbing sociopaths run everything. Max gossip and everyone performs for an invisible audience.

The reference check is the institutional version. You call a former colleague and ask them to talk behind someone’s back, and we all agree this is fine because there’s a formal process. It’s the same thing.

Judaism has autistically analyzed this more than I have. Lashon hara is the prohibition on saying true negative things about people. The Chafetz Chaim codified it in 1873, then carved out the exception. If someone is about to enter a business deal or a marriage with a person who will hurt them, you are required to warn them. He grounds it in the commandment against standing idly by while your neighbor’s blood is spilled, and extends it from bodily harm to money. Idle gossip is forbidden. Protective gossip is mandatory.

His conditions: tell the person directly first. Say whether you know it firsthand or whether it’s hearsay. Don’t exaggerate. If there’s a way to protect the listener without the disclosure, use that instead. And the listener may act on a warning but may not treat it as proven.

The Talmud judges the listener more harshly than the speaker, on the logic that gossip does nothing if no one receives it. The exception runs on that same premise. Accomplice or beneficiary, depending on whether it protects him.

We fail one condition. The Chafetz Chaim requires the motive be purely protective, and the Berkeley study says otherwise. People pass on the warning because watching the cheating agitates them and telling someone makes the agitation go away. It works anyway. Most enforcement does.

52 minutes a day. Thirteen days a year, every year, spent discussing people who aren’t in the room. It’s the most expensive enforcement mechanism we run, and nobody owns it. Spend it on the people who earned it.

Subscribe now

No empire ever loved its victims like the West did. The map is the proof.

The Eulogy on the Map

Drive the length of California, and you’re reading a eulogy. Malibu is Chumash. Tahoe is Washo for “the lake.” Ojai, Napa, Shasta: three more languages, none of them spoken by anyone in the towns that wear the words. Yosemite is Miwok, though not even the word the valley’s own people used. Yosemite was what outsiders called the band who lived there, and it means “they are killers.”

Zoom out, and the pattern holds. Roughly half the American states carry indigenous names: Massachusetts, Connecticut, Michigan, Mississippi, Kansas, both Dakotas. Chicago is a Miami-Illinois word for wild onion. Manhattan is Lenape. Seattle is named for Chief Si’ahl, who watched the city take his name in the 1850s and then, in 1865, pass an ordinance barring his people from living in it.

The Standard Move

Conquerors aren’t supposed to do this. The standard move is to write your own name over the old one, the way Constantine wrote his over Byzantium. And plenty of Western conquerors did exactly that: New Amsterdam became New York, and half of Africa spent a century as Leopoldville, Stanleyville, and Rhodesia. Nor is keeping names uniquely Western.

The Ottomans kept “Constantinople” on their coins and documents, as Kostantiniyye, for nearly five hundred years. The Arabs kept Damascus as Dimashq, Cordoba as Qurtuba, Alexandria became al-Iskandariyya, the conquerors seemingly politely Arabizing the name of the Greek who founded it.

But look closer, and those are a different phenomenon. The Ottomans kept Kostantiniyye because the city was alive: live tax rolls, live patriarch, live customs house. The Arabs kept Dimashq because they ruled Damascenes and needed to bill them. It was in the interest of administrative continuity.

The American case is something else. We kept the names after the people. The Dakota Territory was organized in 1861; the Dakota were driven from Minnesota in 1862, with bounties on their scalps. The name stayed, the people went.

The Sciences of Appreciation

And here is where the West becomes genuinely strange, because it didn’t stop at names. It built entire sciences of appreciation for the people it was replacing.

In 1786, William Jones, a British judge in Calcutta, stood up and announced that Sanskrit was “more perfect than the Greek,” and in doing so, discovered Sanskrit, Greek, Latin, Gothic, Celtic, and Old Persian had all sprung from a common ancient language. This became known as the Indo-European language family, the deepest fact we know about half the world’s tongues. In 1822, Champollion cracked the Rosetta Stone and handed Egypt back three thousand years of its history. In 1837, James Prinsep deciphered the dead Brahmi script and in the process resurrected Ashoka, the greatest emperor India ever produced and one India itself had entirely forgotten. India’s own past was returned to it by an employee of the people occupying it.

The Spanish version is darker. In 1562, Diego de Landa burned the Maya codices at Mani, one of the great crimes against human memory. Then he went home and wrote a meticulous account of Maya writing, calendar, and customs. Four centuries later, that account became the key that let scholars decipher the glyphs. The arsonist kept notes, and the notes saved what the fire destroyed. His contemporary Sahagun spent decades interviewing Aztec elders in Nahuatl and produced the Florentine Codex, the finest ethnography of a conquered people the world had yet seen, compiled inside the empire doing the conquering.

America industrialized the impulse. The frontier was declared closed in 1890, and the Bureau of American Ethnology was transcribing languages as their last speakers died. Edward Curtis hauled a camera across the continent to make forty thousand photographs of what he titled The Vanishing Race. Today, the United States Army flies the Apache, the Black Hawk, the Chinook, the Kiowa, and the Lakota, and fires the Tomahawk.

Love for the Dead

No one else ever did this. Persia appreciated its subjects magnificently, but as living clients: Cyrus rebuilt Babylon’s temples and Darius ruled Egypt as pharaoh. Baghdad translated Aristotle; Akbar translated the Mahabharata. Theirs was an appreciation of the living, extended for reasons of state. The Mongols tolerated every god and excavated nothing. Rome built no museum of Carthage.

The West alone invented archaeology, philology, ethnography, the museum itself: love for the dead, at scale. The British Museum may be the strangest building on earth, a conqueror’s shrine to everyone it conquered. It makes sense when you assume the high intelligence of the conquerors, as they’re naturally curious.

Safe to Love

So the appreciation is real. Cooper published The Last of the Mohicans in 1826; the Indian Removal Act was passed in 1830. The romantic Indian entered American literature at the exact moment the actual Indian was being marched out of Georgia. Sentiment tracked elimination like a shadow.

A name, a language, a culture become safe to love at precisely the moment they can no longer make a claim on you. When hunting lions became an easy sport, lions became majestic. The Massachusett could be a state seal once they weren’t a threat. We have a history of romanticizing rivals who can no longer win.

Which is why the map of America reads the way it does.

The Spanish, who ruled living native populations and married into them, gave their towns saints’ names and kept the native words mostly for the land. The English-speaking settlers, who replaced populations rather than ruling them, scattered indigenous names everywhere because the names had been orphaned.

Tribute or Taxidermy

The love is real and still active: we changed the name of the highest mountain in North America to Denali in 2015, renamed it back to McKinley again in 2025, and will apparently keep fighting over it, which is not how anyone treats a matter of indifference. Land acknowledgments, tribal museums: the sentiment stays strong, a century downstream of the conquest.

The West often overappreciates the cultures it replaced. It’s easy to forget the savagery when you never experience it firsthand. Half of the United States is named in memoriam by the only civilization that ever fell in love with its victims and cataloged them properly. You can frame it as tribute or taxidermy. The names stay on the map.

Subscribe now

In 1984, GM’s plant in my hometown of Fremont, California was the worst car factory in America. Absenteeism ran at twenty percent. Workers drank on the line and sabotaged cars out of spite. GM had quality goals stacked to the ceiling. Then Toyota took over the plant, rehired most of the same workers, and within two years, the quality matched that of Toyota’s plants in Japan. Same people. Same union. Same goals.

Everyone quotes the line about falling to the level of your systems. Almost nobody who quotes it can define a system. So they nod at the aphorism, buy OKR software, set goals, and blame people when the goals fail.

Goals Are Overrated

Goal-setting theory is one of the most replicated findings in organizational psychology. Locke and Latham spent four decades showing that specific, difficult goals beat “do your best” in 90% of studies. Tell a team to cut renewal processing time in half, and they’ll outperform a team told to improve renewals. This is as close to settled science as management gets. Anyone telling you goals don’t matter is arguing with a mountain of evidence. But the research comes with fine print.

Goals work when people get fast feedback, are genuinely committed, and can actually influence the outcome. Remove any one of these conditions, and the effect dies or inverts. Those conditions are the system.

What is a System

A system within an organization is mostly two things.

1. Decision rights: who gets to decide what, without asking.

2. Incentives: what gets rewarded, i.e., what gets you paid, promoted, and praised.

Goals tell people what to achieve. Decision rights determine whether they can act. Incentives determine whether they want to act in ways that help. Three different jobs. Early-stage startups are easier systems because when people have significant upside in the business, they have an incentive to make it succeed.

Most leaders set goals and assume decision rights and incentives will sort themselves out. They never do.

Decision Rights

In 2003, a Johns Hopkins physician named Peter Pronovost gave Michigan ICUs a five-step checklist. Wash hands, clean the skin, drape the patient, wear sterile gear, dress the site. Nothing on it was new. Medicine had known these steps for fifty years, and every hospital on earth already had the goal “don’t infect patients.” The program worked because of one rule attached to the card: nurses were authorized to stop any doctor who skipped a step. A nurse overruling a surgeon mid-procedure. Median infection rates fell to zero within months and stayed there, an estimated 1,500 lives saved in eighteen months. The checklist got famous. The decision right did the work.

When the UK later copied the checklist nationally without the same authority behind it, the results were far weaker. That is what missing decision rights look like from the outside: nothing. People freeze, waiting for approval on decisions nobody told them were theirs. Or two people, both believing the decision is theirs, collide. Freeze or friction.

Incentives Fail Louder

Wells Fargo’s retail bank had a goal: eight products per customer household. “Eight is great.” Specific, measurable, tied to compensation, exactly what the textbooks prescribe. Employees delivered roughly 3.5 million fake accounts. 5,300 people fired, a $3B settlement, a CEO banned from banking for life. The employees understood the goal perfectly. The incentive system paid for accounts opened, so they opened accounts. A goal, achieved by a misaligned incentive structure, hits the wrong target with precision.

Academics cataloged this in a 2009 paper titled “Goals Gone Wild,” and it reads like a rap sheet: tunnel vision, gamed metrics, ethical collapse, cooperation destroyed. A goal amplifies whatever the incentives already reward.

W. Edwards Deming used to run a demonstration called the red bead experiment. Workers scoop beads from a bin, and management grades them on how few red beads they pull. The share of red beads in the bin is fixed. Effort changes nothing.

Ranking the workers changes nothing except morale. Deming estimated that 94% of performance variation comes from the system, not the individual. Setting personal targets on outcomes that the system controls is grading noise. I’d argue smart creative work is incredibly different from the work Deming observed. Even so, systems have massive impacts.

Why Do Smart People Keep Getting Systems Wrong?

Three confusions.

First, they mistake the artifact for the system. The checklist, the dashboard, the OKR tool. They buy the performance software, and nothing changes, because the artifact is the visible residue of a system, not the system itself. Pronovost’s checklist was five lines a med student could memorize. Its power was the nurse’s authority.

Second, they hear “systems” and think process, and process means bureaucracy, so systems must be the enemy of speed. Exactly backwards. Decision rights exist so people can act without asking. The alternative to a system is escalation. Every decision without an owner climbs the org chart until it finds one, and the top of an org chart is a bottleneck. Micromanagement is what happens when you refuse to build systems.

Third, they think systems versus goals is a choice. It’s a sequence. Goals run on top of systems the way software runs on hardware. The legendary Andy Grove, who invented OKRs at Intel, reviewed them monthly and insisted they stay away from compensation because a goal tied to a bonus becomes a negotiation. People sandbag. John Doerr repeats the warning in every OKR book, and companies ignore it anyway because linking goals to pay feels like accountability.

How to Do It

Amazon is the rare large company that runs the sequence in order. Its Weekly Business Review grades teams on input metrics (price, selection, delivery speed, things a team can actually move this week) rather than outputs like revenue, which nobody controls directly. Look at the two pillars inside that design. Inputs a team controls: decision rights. Grading on those inputs: incentive alignment. And it happens weekly. A goal set at the annual review gets attention twice a year. A weekly review touches it fifty times.

So, the order of operations. Map decision rights first: every recurring decision gets exactly one owner, written down. Audit incentives second: list what actually gets people paid and promoted, put it next to what creates value, and close the gap. Set goals third. In that order, goals land on people who can act and want to.

Conclusion

Systems thinking is how you get operations and really anything to work well at small and large scales. It’s applying proven economic theory to the real world, as Charles Koch says. It’s a simple, hard thing. Choose your principles and stick to them.

For my hometown’s car plant, what changed was a cord. Toyota gave every lineworker the authority to pull the andon cord and stop the entire production line, no approval needed, the moment a lineworker saw a defect. One worker would halt millions of dollars in throughput at their own discretion. GM told workers quality mattered. Toyota made quality a decision owned by lineworkers. The same lineworkers who ran the worst plant in the country later ran the best plant.

Systems are why the US government is so terribly run: why so much is spent, and so little gets done. Most government employees are good people trying to fix problems. It’s the systems they’re working in that fail.

Goals tell people what to achieve. Decision rights determine whether they can act. Incentives determine whether the action helps. Deming said 94% of the variation lives in the system. For most orgs, it’s at least 50%. Budget your attention accordingly.

Subscribe now

Last week, I shared the little things I do at home. Italy earned its own edition. You learn a country faster through its light switches than through its museums, and the differences are the fun part.

I don’t snack when eating in Europe, so I feel I eat less. Even though I’m carb-loading with every kind of pasta, I’m also walking around more, exploring, so the carbs don’t feel as bad as they do at home.

Beauty

It’s stunningly beautiful. Going through Tuscany, I see what inspired Napa and some of the most beautiful American designs. The rolling hills, flowers, and vineyards.

AC

Italians treat air conditioning as a garnish. A taxi at 100 degrees: AC on, both windows down, door cracked at the light. I rode in another car with the AC blasting and recirculation off, so we were cooling highway air one carload at a time. I showed the driver the recirculation button. He didn’t know what it did, but I convinced him to press it. America engineered summer out of existence. Italy accepts it, and it kills the old and infirm every summer.

The room units are their own adventure. Thermostats have buttons for everything except temperature: a snowflake, a moon, a water drop. You negotiate with the machine, which might be the most Italian thing about it.

Windows

Because they don’t understand AC, there are no fans in the restrooms. There’s a window, and it opens. You can hear the people outside, and maybe they’ll hear you too. America sealed its bathrooms and bolted a motor to the ceiling. Italy solved ventilation thousands of years ago and saw no reason to change.

Switches

No dimmers anywhere. A light is on, or it’s off. Every room has a bank of identical switches, and which one runs which light is a nightly puzzle. By day 3, you’ve memorized the map and feel like a local. At least the switch usually lives in the room it controls (Korea puts the bathroom switch outside the bathroom, which turns every shower into a trust exercise).

Bidets

Italy has been right about bidets for a century. Say you fall in the mud and get it on your arm. You don’t get mud off your arm solely by rubbing it with a paper towel.

Since 1975, Italian law has required one in every home. The idea is correct. However, the execution fails: the stream starts low and points down, so it washes everything except the target. You sit next to a mini-sink. It’s helpful for wetting toilet paper, which still beats the American system.

Next to the bidet hangs a linen towel. Beautiful instinct. Linen dries fast and lasts decades. The same towel also greets every guest until laundry day. Right material, wrong refresh rate. American paper towels and toilet paper are just built better.

Flies

Even the highest-end places skip the little spinning fans that keep bugs off food (a $15 gadget that works like magic). So flies join every outdoor meal. Italy seems to have signed a separate peace with its insects. They seem to get in everywhere. The burrata is worth defending, but dinner outside at golden hour is worth a fly or two.

Safety

The ruins are gloriously dangerous. Loose stones, open ledges, staircases where a fall kills you. No railings, no signs, no waiver. Everyone climbs anyway. I love it.

There are blind turns and sketchy main roads where it’s just dirt and rocks instead of a paved road because it’s a UNESCO site.

Then I walked into a gym with a massive sign: Pilates machines strictly forbidden without a trainer. The amphitheater has gone 1,900 years without a handrail. The reformer needed a warning label within a decade of arriving. Trust is grandfathered in at any historic site, and the nanny state puts kid gloves on anything new. It’s such a silly dichotomy.

There are a hundred more of these. Every country runs on 10,000 little defaults, and you only see your own from a distance. Each one is a small reminder that home is a set of choices defined by culture and history. I prefer America. Happy 250 to the best and my favorite country in the world.

Subscribe now

Don’t sweat the little things. It’s all little things.

We do small things because we prefer them. Maybe it’s a quirk, maybe it buys a little comfort. I always appreciate when someone thinks of the small stuff.

Shade

When it’s hot, I walk on the shaded side of the street. Walking one block in the sun is the equivalent of walking four blocks in the shade. Not all shade is equal. The side that’s been shaded the longest is coolest because the cement beneath it has had time to cool. Heat comes at you from above and from the ground. Walk in the shade that has been shaded longest. As someone who runs hot, this one matters.

Buttons

How do you button a shirt? Top-down, bottom-up, or middle-then-out? I used to go top down, but it sometimes came out misaligned, and I’d have to redo every button. Bottom up works best. The lowest buttonhole is easy to pair with the lowest button, and you work up from there.

Sitting

When you’re lounging, what’s the most comfortable way to sit? I prefer shoes off inside, feet up, or Indian style on the couch. I don’t get how someone can be truly relaxed with their shoes on.

Dark

When I drink water before bed, I may wake up in the night to use the bathroom. I used to flip on the light. Now I leave it off, find my way in the dark, and sit down. Falling back asleep is easier when my eyes never have to readjust. This is harder in hotel rooms, but it’s a game of slowly moving if it’s too dark.

Same idea with screens. In the dark, I keep my phone on the lowest brightness. For my eyes and everyone else’s.

Breathing

I breathe through my nose, down to the belly. At least once a day, I do box breathing, cyclic sighing, or slow breathing– different methods for the same thing. Slow the breath down. It’s shown to raise HRV, lower anxiety, and lift mood.

Calls

I get chores done while on calls. Watering my plants, tidying up, opening mail, folding laundry, or even workouts like planks, ab wheels, or stretches. I’ve had a mini trampoline everywhere I’ve lived for over two decades. I get the one with bungee cords because it doesn’t make noise when you jump. Great workout, but be careful with your jumps, as they can cause injuries. And you need enough cardio to keep your breath steady on calls.

I’m more engaged when physically active on a call. It’s like walking while talking, which is shown to wake up the brain and make it more creative.

When I get to know anyone smart and hardworking, they all have stories about having to flush a toilet during an important call. No one wants to, but sometimes you’re back-to-back and nature calls. Muting at the right time is make-or-break in those situations.

Pull-ups

I used to get tennis elbow all the time after pull-ups. A Cirque du Soleil performer taught me how to hang and do pull-ups every other day without tennis elbow.

As you pull, press into the bar through your pinky and ring fingers. Most people press through the index and middle fingers, which is what brings on tennis elbow. This was a game-changer for me.

Bananas

How to open a banana: upside down. Pinch the end opposite the stem, and it splits cleanly. Peel down, stick the little nub on the peel, and you’re holding the stem as a handle. The whole banana is good, and you skip the smushing you get fighting the stem from the top.

There are probably a thousand other little things I do or have learned that I enjoy. It’s fun to share, and I always love to learn these little life tips from others. Each is a tiny bump in how good a day feels. They add up.

Subscribe now

Kindness is getting rare. Politeness you can fake, and many do. Consideration is disappearing. So is the shame that once enforced it.

Blame the self-esteem movement, bad therapy, and worse parenting. We spent three decades teaching kids they were special. Participation trophies, no losers. The intent was fine. Confidence matters. But we only taught the first half. We boosted self-regard and never taught regard for anyone else. Empathy got redefined as a tool for validating how someone already feels, not the harder thing it actually is: stepping into a world that isn’t yours. A generation raised by therapists came out fluent in rationalizing their own feelings and illiterate in everyone else’s.

Then the phone and now AI arrived and rewrote the rules on top of all that.

Etiquette was never about which fork. Emily Post built the entire canon on three principles: consideration, respect, honesty. Those never change. The form does. The Romans grasped forearms; we shake hands. Same principle, different manner. George Washington copied out 110 rules of civility by hand at 14, and most still hold, because the reason underneath them is permanent. We didn’t lose the principles. We lost the manners that carry them, and we stopped writing new ones for the world we actually live in.

None of this is new. Etiquette has always been an instrument. The word comes from the French court, where an étiquette was the ticket that spelled out how to behave in the king’s presence. Louis XIV turned it into a weapon. After the nobles revolted against his father, he moved them into Versailles and buried them in protocol, who got to hand him his shirt in the morning, who was allowed to sit. A class that used to raise armies spent its days jockeying over place settings. Manners are how he tamed them. That’s civilization.

We need this again.

The Old Etiquette

Markers of character we used to read on sight and now wave past.

• Personal hygiene.

• Hold the door. Give up your seat for someone who needs it more.

• Be on time. Your lateness is a bill you hand to everyone waiting.

• RSVP, then show up.

• Tip well.

• Know when to say thank you, and write the note.

• Know when to offer to pay.

• Give gifts. Remember the occasion before the calendar reminds you.

• Chew with your mouth closed. Don’t reach across someone’s plate.

• Dress for the occasion.

• Receive well. Accept graciously.

• Introduce people. Learn names and use them.

• Listen more than you talk. Let people finish.

• Apologize cleanly when you’re wrong.

• Clean up after yourself.

• Show consideration for how other people feel.

• Real empathy. Step into someone else’s world.

• Don’t talk about your exes on a date.

As Edmond Burke said, “Manners are of more importance than laws. Upon them, in a great measure, the laws depend.” As our world adapts, so must our etiquette.

The New Etiquette

The smartphone and AI have invented entirely new ways to be rude. We need new manners for the world we’re in now.

• Know what to share and what to keep private. In person and online.

• Don’t send AI slop. It’s okay to use AI to write, but edit it to your own voice. This is especially true for condolence notes or wedding toasts.

• It used to be just ask Google. Now it’s ask AI, not your friends. Don’t ask the group chat, “Is this true?” Copy or screenshot it, then paste it into AI and ask. It’s a solid fact-check. For politics, it’ll usually get to the truth if you ask the correct questions. Don’t put the work on your friends.

• Learn to text. Match the medium to the message.

• Turn your screen brightness down and your flash off

• Don’t read other people’s screens. It’s reading their mail and their personal journal all in one.

• Take your AirPods out when someone’s helping you.

• Don’t screenshot private messages and pass them around.

• Keep your noise to yourself. No speakerphone, no music without headphones, no videos on full blast. No one asked to hear your playlist.

• Keep your smells to yourself, especially in confined spaces. No spray sunscreen indoors. No perfume in an elevator.

• The phone should be face up only if something urgent might come in. Besides that, the phone should be face down on the table or kept off the table. A phone sitting face-up detracts from the conversation, as if something more important might come up any second.

• Don’t film strangers without consent.

• No calls or FaceTime in a restaurant or confined space.

• Vape outside. Don’t blow it at people.

• Don’t spit nicotine pouches mid-conversation. If you must, put it into the container, not onto the table.

• Keep voice memos short, or just use Wispr Flow. Nobody opted into your monologue. Europeans take note.

• Settle up. Pay people back without being chased. And don’t Venmo-request a friend for $4.

• If you invite someone to an event, don’t Venmo them afterward for the cost of the event.

• If you’re leaving a job, say so first. Tell them you’re considering a move and why. Give them the chance to respond before you’ve already decided. Don’t blindside the people who bet on you.

What to Enforce

Enforcing etiquette can also go too far.

When a Byzantine princess ate with a golden fork at her wedding feast in 11th-century Venice, a monk denounced it as decadence and an affront to God, who gave us fingers to eat with. When she died of plague soon after, he called it divine punishment. The fork didn’t take hold in Europe for another 400 years, and we now hand it to toddlers. The monk got the target wrong.

Shame the fork, and you’re the monk. Enforce how people treat each other, and the forks sort themselves out.

Reset

Etiquette is resetting cultural standards. It’s how we reclaim what we dropped and shape what comes next. The fix is kindness, thoughtfulness, and intentionality. And the thing the self-esteem era worked hard to delete: Shame.

Shame is the enforcement mechanism. It’s the quiet social cost that made any of this stick, and we spent a generation insisting nothing was ever anyone’s fault. No shame, no standards.

Make shame great again.

Subscribe now

3,255,323 veterans are paid disability every month for tinnitus. Ringing in their ears.

There’s no test for it: no scan, no blood draw, no machine that can hear what they say they hear. They report it, and the government pays. It’s the single most compensated disability in the VA system, more than a million claims ahead of anything else.

Hearing loss, tinnitus’s audible cousin, is the injury we’d expect from the same gunfire and the same engines. Hearing loss has an audiogram. You sit in the booth, you raise your hand. 1.5 million veterans are paid disability for it.

So the semi-testable injury is claimed half as often as the one you can’t test. Same wars. Same ears. Twice the ringing.

No Country For Young Men

America used to have high trust. So did most of the West. We built our institutions on the assumption that people describe their lives accurately, and you can pay them on their word. Disability, food aid, tax credits, medical claims. Nearly all of it runs on self-report. Tell us your income, your household problems, your symptoms, and the check goes out.

That assumption was once close enough to be true. It’s not anymore, and the systems can’t tell.

An Experiment

I ran an experiment. I asked an LLM to score medical conditions from one to ten on a single axis: how easy is this to fake? Ten means the doctor can’t catch you. One means the lie is caught the instant a test is run.

Then, I lined the scores up against the conditions people are actually paid for. They matched. The stackrank of fakeability was the stackrank of prevalence.

Back and joint pain, a nine on fakeability, is the largest category of disability awards by a wide margin, at 34% of all cases. Mental disorders, an eight, come next at 25% of all cases. Then it falls, in order. Neurological, a five. Heart disease, a four. Diabetes, a three. Injuries, a two. Cancer, a one, at just over 3%.

The disability conditions we can prove are rare. The conditions we can’t confirm are everywhere.

Tinnitus is real, and it’s miserable. A close friend got it from being injured in the line of duty. Similarly, back pain and depression are real, and many need support from the suffering they cause.

But a system is the incentives it creates. When the lie pays, and the truth doesn’t, you incentivize lying.

Pay First, Verify Never

This repeats throughout government programs.

Medicare pays before it checks, by law. A clean electronic claim has a payment floor of fourteen days and a ceiling near thirty. The claims are billing codes that practitioners typed, asserting they did the work and that it was needed. Nobody confirms it at the window. So the grift lives in the codes.

Bill a ten-minute visit as a complex hour. Bill for the wheelchair that never shipped on a stranger’s number harvested off a robocall. The government pays now and chases later. But the chases rarely happen, and when they do, they recover pennies on the dollars sent out. The payment speed that protects honest doctors makes the theft frictionless.

The Earned Income Tax Credit does this with cash. The concept of an EITC is great as it can function as a negative income tax, incentivizing people to work. A smarter version of UBI. Unfortunately, how it’s administered is full of fraud.

The credit is refundable, so the IRS mails you a refund even if you owe nothing. It peaks at a precise income and a precise number of children. If you’re self-employed, no W-2 contradicts the number you write down. Report the income that maximizes the credit. Claim a child lived with you, on your word alone. The result is 32.7% of EITCs being improper payments ($21 billion) in 2025. That’s just the treasury’s own estimate of fraud for one year.

Who Pays

The agency only moves the money. The rest of us fund it. Every system that pays the liar taxes the honest by exactly that much. The bill has to be paid eventually.

Another way to frame it– the value of a statistical life is ~$10M. So, we can say 2,100 people’s lives are lost from EITC fraud each year alone. They’re getting away with murder.

Gov Systems

These systems were designed for a different country: a high-trust one, where your word was collateral and close enough to true. We kept the systems and lost the trust. Massive reform and restructuring of incentives are needed.

Subscribe now

I watched Good Fortune (starring Aziz, Seth Rogan, and Keanu Reeves) on a flight this week. Mediocre movie, but fascinating in the way it epitomizes how many people view the world. The movie gives insight into the type of people who vote for Karen Bass in LA.

Increasingly more people don’t understand how progress and wealth are made. I won’t name characters or plot details, but the below may spoil parts of Good Fortune if you plan to watch.

Gig Economy

The screenwriters assume the average person is a jerk and dings their DoorDash/Uber Eats delivery person as part of their collateral rage. One client was upset because of the lack of ranch sauce, and despite the gig worker’s protest, “I only pick up the food.” “One star!”

A gig worker talks on the phone to a client after waiting in line for two hours, “Please don’t cancel the task, or I don’t get paid.” The client verbally confirms they won’t, and immediately after the call, they cancel the task.

The gig economy and marketplaces like Uber and Airbnb work because most people are decent humans. They don’t trash places for no reason. The marketplaces put in place safeguards and incentive structures that reward good behavior and punish bad behavior on all sides. Without those safeguards, things go haywire.

Grift and Fraud

There’s a power law of good and bad actors. The number of bad actors in society increases as the incentives to do so increase.

Almost all government jobs and contracts assume a super-high-integrity, high-trust society. That made sense when they were originally set up. Unfortunately, America has enough people who defect that these no longer work.

Who would lie about running a hospice or a daycare for children? Think of how horrible a person you’d need to be. We’d universally want to condemn that person. This used to be true, but if most of the people taking advantage of the system are in a protected class, some will not want to acknowledge it. So now, fraud is rampant, and what was once a bipartisan issue has become partisan.

Not How It Works

In Good Fortune, the VC guy inherited all his money and then invested it. VC work is seen as just Zoom meetings and meals, with no discussion of deals or anything substantive. They completely misunderstand VC. As a former VC, I’m all for making fun of VCs and classifying investors, but at least have it be true.

Multiple characters say no one can make a ton of wealth without hurting others. Screwing other people is a prerequisite to success. Good Fortune treats this view as gospel. The movie ends with a speech by the VC guy telling the tech company that they only make money by abusing their workers. Somehow, his VC fund completely controls the company. Then he triumphantly says, “Fuck AI!” It’s a callback to an earlier scene where an AI delivery bot was going to replace him… at the gig job he hated.

Good Fortune’s regressive worldview fundamentally misunderstands how technology impacts us. In 1900, 40% of Americans farmed. Today it’s 2%. Tell someone in 1900 that machines would take 95% of farm jobs, and they’d brace for famine. Instead, we built the modern world. We designed cars, wrote software, flew planes, made movies, erected skyscrapers, and launched satellites. The 2% who stayed now feed the country better than any society in history. Losing the farmers brought us modern life.

Even so, investing in technology has never been a sure thing. Add up a century of profits from airlines and cars. The total comes in under what investors put in. Those industries remade the world, but building the future and getting paid for it are two different things. Railroads and telecom eventually paid themselves back. We’ll see how it turns out this time.

Perceived Good

One of the characters cast as a good person wanted to unionize their work. They’re seen as a hopeless romantic trying to do the right thing. Persistent in attempting to get better compensation.

Many of the very first unions made sense as working conditions were hazardous. That’s no longer the case. We now have federal, state, county, and municipal laws to protect workers, along with generations of case law on their side.

Unions evolved into protectionist cabals. They stop innovation (e.g., Longshoremen unions in America have stopped port automation that has been possible since 1993, as shown by the Port of Rotterdam). Unions are, with few exceptions, extractive entities that distribute employee wages to cronies. They work against the broader labor pool and raise the cost of goods.

Modern unions only make sense if you’re an incumbent who sees the world as zero-sum. They’re net negative for society but positive for the incumbent, as they create a barrier to entry.

The System

Good Fortune frames the system as rigged and terrible for poor people. Someone can do everything perfectly and will still be in the bottom decile. This isn’t the case. There are hardworking, smart people I know who have gotten unlucky many times over, but they’re not in the bottom decile. They’re not in the top decile, but even the most unlucky people I know who are smart and persistent are in the top quartile of success.

The wealthy have enough is another core message of the movie. But the counterfactual isn’t run. Let’s say all the wealthiest people stop working tomorrow. They’d stop starting, building, and fixing companies and societal problems. The world would be worse for it.

Capitalism isn’t perfect. It’s far from it. Even in the tech world, where we push meritocracy, it’s rare to look at a startup’s cap table and think, “That’s fair,” given who truly contributed to the company’s success. Similar to how democracy is the worst form of government, except for all the others, capitalism is the worst economic system except for all the others.

Subscribe now

A jury convicted Timothy Wright of strangling Joyce Vanderhoff and leaving her in the brush in Mays Landing, New Jersey. The proof was on his phone: turn-by-turn directions running from a spot beside her body to his front door in Pennsylvania, time-stamped the night before she was found. Sentenced to fifty-five years. Then, an appeals court let him out.

A detective had held the phone past the consent Wright gave. The facts didn’t change, but the court ordered a jury to pretend it had never existed.

When police gather evidence in violation of the Constitution, we punish the case. The gun is real, the blood is real, the timestamp is real, and the jury sees none of it. We blind the jury to discipline the police. This is insane.

If someone breaks the law to get evidence, they should answer for it (fired, fined, charged). But the fact doesn’t turn false because of how it was obtained. Suppressing it only creates more victims– the innocent and affected, and whoever the freed man reaches next.

How We Got Here

The Supreme Court made rulings. Weeks bound the federal courts in 1914.

Cardozo refused to bring the exclusionary rule to New York in 1926.

Later, Mapp bound every state in 1961. The idea was that a right with no remedy is no right, so the remedy for an illegal search is to pretend the search came up empty. Then it grew teeth. Under “fruit of the poisonous tree,” anything that the violation leads to is poisoned too. Find the body because of a coerced word, and the body disappears along with the word.

Luigi

Luigi Mangione, accused of shooting UnitedHealthcare’s chief executive on a Manhattan sidewalk, has pleaded not guilty. He’s clearly guilty. Police took him at an Altoona McDonald’s and opened his backpack without a warrant. This May a judge called that search improper and threw out what it produced: a loaded ammunition magazine, his phone, his passport, his wallet, and a computer chip.

The magazine is a magazine. The passport is his. They were ruled invisible to a jury on the question of where the officer’s hands were when the zipper opened.

Luckily, the judge allowed the most important items in (the gun and the notebook, found later during an inventory search). However, the other items were excluded from the murder trial because of the manner in which they were obtained. A judge could have just as easily excluded the gun and notebook from the evidence.

The Steelman

Let’s take the other side at full strength.

If we count illegal evidence, we incentivize police to seize it illegally. Kick the door, grab the phone, win the trial. Do it a thousand times, and we break the Fourth Amendment (protection from illegal search and seizure). Exclusion was never about the guilty man. It’s about the next thousand searches it scares off. Pull it, and nothing stops the door from coming down.

I get where this is coming from, but it misses the consequences.

Exclusion punishes the police by freeing their suspects. Then the victim’s family pays the fine for the cop’s mistake.

Put the cost of an illegal search on the man who ran it. Make it personal, automatic, and steep. We could have it pierce qualified immunity with a simple structure. A quick administrative track where it’s a three-strikes rule; a police career ends on the third illegal search.

A cop who knows a bad search ends him gets the warrant first. The truth still reaches the jury regardless. This aligns incentives. The deterrent would only bite the hand that did the thing.

Include the Innocent

We tell ourselves the exclusionary rule protects the innocent. It does the opposite. It frees the guilty. Freeing the guilty harms the innocent.

Subscribe now

I almost lost a finger to a house.

I was swimming at a friend’s place. The backyard had an outdoor movie theater, a large metal box that opened up and raised a screen on a hydraulic lift. The projector was already acting up. Then I heard a loud mechanical noise, and the lid started to close on its own.

I jumped out of the pool and ran over and pushed up against the closing lid with everything I had. It kept driving down through my full body weight. My finger got pinched in the seam and I yanked it out half a second before the lid won. The screen and the housing were destroyed. My finger wasn’t, barely.

As far as smart homes go, this isn’t even that crazy.

The Trap

Every newly wealthy person buys an estate, fills it with smart devices, and watches those devices slowly turn on them. The systems work the day they’re installed. Then they rot.

The pitch is good. You open one app, and every blind in the house glides up at once. You kill all the exterior lights from your phone while you’re at a restaurant. You change the music in the living room from your bed. The first week, it feels like you bought the future.

There are early warnings– three remotes for one room, a literal server heating up a closet on your property. But the whole system costs tens or hundreds of thousands of dollars, so surely you’re paying for something that lasts.

You’re not.

The macro switches are the first to go. The button that turns six lights on, then off, starts leaving one on. The dimming stops agreeing with itself: half the room is at 80%, the other half at 10%, and one fixture has decided to glow a different color. The Crestron system that ran everything locks up.

Crestron, Savant, Control4, Lutron, Vantage, there are dozens of these companies. They run flawlessly at install. It breaks once, they drive out and fix it, and you relax. Then it breaks again. Then, breaking becomes the new normal. Eventually, your bar for a good day is that the lights actually turn off when you tell them to.

What gets me is the universality. This has happened to nearly every wealthy person I know. It’s a running joke. People on their second or third house rip the whole thing out and go back to switches on the wall.

How Is This Not Solved?

It doesn’t look like a hard problem. Lights. Blinds. Music. Humanity has landed rockets on barges. Surely we can keep eight bulbs agreeing on whether they’re on. But the difficulty isn’t in any one device. It’s in the spaces between them where software can’t reach.

The deepest issue: there’s no single source of truth. Your phone app thinks the kitchen lights are off. The wall keypad thinks they’re on. The hub in the closet has a third opinion, and the bulb itself has a fourth.

In a well-built distributed system, you have one authority that every component checks against, like a bank ledger that all the branches defer to. A luxury smart home is the opposite. It’s four branches, each keeping its own ledger. The moment those ledgers disagree even slightly, the house starts acting haunted. You press “all off” and one light stays on, because as far as that light knows, it already got the message and ignored it.

Then there’s clock drift, which is the quiet killer. Every device in your walls has its own internal clock, and those clocks are not synchronized to a common time. Each one runs a little fast or a little slow, drifting by seconds a day. A command isn’t really “turn these on together.” It’s a dozen separate commands fired at a dozen devices that are each operating on slightly the wrong time.

On day one, the drift is milliseconds and invisible. Six months in, the gaps have compounded. Daylight savings time alone breaks thousands of these setups every spring, because a device with the wrong timezone fires the morning routine an hour early.

Stacked on top is protocol fragmentation. Your bulbs talk Zigbee. Your locks talk Bluetooth. Your cameras talk Wi-Fi. Your blinds talk Z-Wave. These are different languages, and the hub in your closet is a translator running four conversations at once. Every translation layer is a place where a message gets dropped or arrives late. The more devices you add, the more conversations the one overworked translator has to juggle. Translators that are slightly overloaded start losing a word here and there. (Matter, the new industry standard that’s supposed to make everything speak one language, has been arriving since 2022 and is still, in 2026, a partial fix at best.)

Lastly, the part nobody markets, every smart home setup is hand-coded. There’s no “smart home” software you install like an operating system. Your $200K system is a custom program your integrator wrote by hand for the exact wiring, exact devices, and exact layout of your specific house. The light groupings, the macros, the timing delays between commands, all of it is bespoke.

Your house is, in the most literal and least romantic sense, a unique snowflake. Which means there’s no general fix that ships to everyone overnight. When something breaks, the only entity on earth who understands your particular tangle is the guy who built it, and now you’re paying him by the hour to fix a system whose complexity guarantees it’ll break again.

Why Smart Homes Work Then Fail

At install, everything is in a clean, known state. Every clock is freshly set. Every ledger agrees. The integrator has just hand-tuned every group and every delay and tested each one in front of you. The system is brand new, and entropy hasn’t had time to accumulate.

Over time, clocks drift apart. A firmware update silently changes how one device behaves. A brownout resets four bulbs to a default that the program doesn’t expect. Each tiny desync is small enough to ignore, and there’s no mechanism that pulls everything back to a single source of truth, because that source of truth was never built. A bank ledger reconciles itself constantly. Your house never reconciles because there’s no way for it to know everything is in sync.

That’s why the lid kept closing while I pushed against it. The box had its instruction and no working sense of the world contradicting it. It wasn’t malfunctioning, exactly. It was doing precisely what a system with no source of truth does. It executed its last command.

Conclusion

AI is going to eat a staggering number of desk jobs. Anything that lives entirely inside a screen, where the rules are clean, and the state is knowable, is in play.

The physical world is a different animal. It’s messy in a way that doesn’t reduce to a clean problem, with infinite permutations and no single source of truth. Even my friends with maximal AI faith, the ones who think most white-collar work is “cooked,” land on decades when you ask them about real-world electrical, plumbing, or HVAC. The smart home is the preview. In 2026, we can’t keep eight bulbs agreeing on whether they’re on.

Although I’ll dent my own point here, because our intuition is bad at this. We feel progress as a straight line, and it tends to arrive in power laws. Something looks impossible for years, and then it isn’t, all at once, and everyone who said “decades” looks slow. Maybe that’s coming for the physical world, too. I hope it is.

Subscribe now

The creators of AI are terrible at branding the category.

The leaders say it’s dangerous, so it should be regulated. It’s in their self-interest, from both competitive (ban all new competitors) and self-important (we’re so important that the gov needs to be involved) sides.

For Gen Z, AI is a thing you use to cheat. Usage spikes during the school year. Some students who don’t want to cheat find it scary. Others use it as a life coach, outsourcing all their hard decisions, which is also a terrible idea.

None of that is what AI is. It’s just what it’s branded as and how it’s sometimes used. This has happened before.

Not a Social Network

The media decided Facebook was a “social networking site.” In the same category as Friendster and MySpace. The framing emphasized community, self-expression, hipness, and youth culture.

Shocker. The media was wrong.

Zuck didn’t want Facebook to be known as a social network. He was adamant about it. He’d talk about his actual vision for decades. The media didn’t care.

Facebook was… a digital facebook. A directory. A CRM for your social life. From the start, Zuck described it as a social utility tool. Real names, structured profiles, and the original .edu requirement all flow naturally from the directory mindset. A utility is for everyone.

Choose Your Own Adventure

Social utility tools are choose-your-own-adventure.

You can obsess over how others see you, post every day, like and comment, and refresh your notifications. Or you can see the whole thing as inherently playful. Send memes. Keep in touch. Ignore it when you’re busy.

AI is a Utility Tool

AI is a utility tool. It’s a utility tool unlike any that has ever existed.

The normal version of “tool” looks like this. Tennis racket. You learn grip and strokes. Forehand, backhand, volley, serve, lob, and eventually drop shots. Now, say an expert and a beginner are told to hit the ball as far as they can. The person more experienced with the racket wins.

That’s how tools have always worked. Skill compounds. Experienced users learn all the ways to use the tool. Experts win.

But if the tool is AI, you can modify the tool itself. Turning the tennis racket into a cannon to shoot the tennis ball beats the guy hitting it with a racket.

That’s the weird part. The thing you’re using rewrites itself based on how you talk to it. Same chat, different framing, totally different tool. People who haven’t sat down with AI in a few months come back and say, “You can do that?”

Yes. And next month, three more things.

AI is the Best

Some examples from the last few months.

I had a massive spreadsheet last week with a dozen tabs, and the numbers weren’t boxing. AI walked through it and found the single cell breaking the formulas. It took a few minutes and saved my entire morning.

I met someone years ago and forgot their name and company. I had high-level descriptions of them and not much else. AI instantly found them.

I had odd symptoms and asked AI to diagnose me like a doctor, then fed it more information as I noticed more. It flagged shingles. The doctor confirmed it the next day. Early treatment meant a mild case.

Created half a dozen acronym names for a project where the teams loved multiple and chose one. When working on larger projects, I can do work in a weekend that would take a small team a month.

That’s just AI today.

AI is a Craft Like No Other

There’s no magic bullet.

It’s a tool that turns into other tools. Turning it into other tools is itself a skill. There are patterns you learn (the way there are patterns in escape rooms, or in investing) that tend to work. Some patterns help some people and hurt others.

Hackers are similar. Hacking can be positive or negative, depending on how it’s done. The question is whether you’re using AI with deliberate practice or you’re using it like a slot machine.

For coding, AI can get 90% correct on the first try; if you identify the 10% that is wrong, run it again to reach 99%. Find the 1% wrong to get to 99.9% and so forth.

The tactics that work: stateful thinking, iteration, discernment. Discern the gap between what AI gives you and what you know.

Giving advice on AI is like giving advice on writing. You can hand someone hundreds of tips and tactics. At the end of the day, it’s still their voice. Depends on the person. Depends where you are. Depends what you’re trying to do.

But Here’s Some Advice Anyway

Keep the frame playful.

Directionally correct is fun. Part of the fun is catching when things break, like a game.

Check the reasoning. When you see it reason wrong, stop it midway, tell it what it did wrong, and redirect.

Iterate, then iterate again.

If you don’t need AI for anything in your life, and you don’t enjoy using it, you won’t use it. Enjoy your hobbies, or you’ll do them badly.

Have empathy for AI and the way it thinks. It’s not you. It’s stateful thinking.

If AI Were Perfect, It Wouldn’t Be Fun

Imagine the easiest game ever made. Press any key, win. Not fun.

Now imagine the hardest game ever made. It’s effectively impossible; you always lose. Also not fun.

AI is directionally correct. It’s supposed to break sometimes. That’s a feature of AI in its current stochastic form, not a bug. The breakage is where the craft lives.

Not Like Us

AI doesn’t think about or consider the consequences of its actions.

This is why, left unchecked, it can do seemingly insane things like delete an entire codebase to fix a bug. Mission accomplished. The bug is gone, along with everything else.

When humans are at our best, we’re expected-value consequentialists. We weigh outcomes and hold uncertainty in our heads. When AI is at its best, it predicts what smart experts would say or do.

The experts aren’t always correct, especially if they don’t have full context on your situation and exactly what you’re trying to accomplish.

Utilize the Tool

I interned at Palantir in ‘08 and ‘09 on the finance side. They had a strategy tool where you could create a custom strategy and write code to run any kind of stock research or trading strategy. Totally innovative at the time. Then I watched end users at Bridgewater take this infinitely customizable tool and use it to generate a dozen standard reports, which they fed into a Palantir dashboard that displayed them.

It seemed like such a waste.

I think about that all the time now. Most people are about to use AI the same way Bridgewater used Palantir finance in 2009. A handful of standard prompts and workflows. Given how powerful and customizable the AI is, that’s such a waste.

Don’t do that.

The tool changes as you use it. You change as you use it. The fun is in the play.

Subscribe now

Something was off.

Two days of itching on my upper left back and left thigh. A sore foot, no fever, and red dots. Nothing dramatic.

I opened Claude and typed:

“I’m itchy on my upper left back and left thigh. They’ve been itchy for two days. Neither are red at all. They both hurt a bit and feel itchy. I haven’t been scratching at all. There are now red dots on my leg, and my left foot feels sore too. What could it be?”

It flagged five possibilities: contact dermatitis, shingles, folliculitis, insect bites, or an allergic reaction. I went through each possibility, learning the symptoms and causes. Through the process of elimination, only shingles remained.

This was weird because I’m in my late 30’s. Shingles is supposed to be a disease of grandparents. The shingles vaccine is only recommended for people over 50. Current guidance says you should only get the vaccine before 50 if you have a weakened immune system, which I don’t.

But the model was clear: one-sided, painful, spreading along a dermatome. As soon as blisters appear, there’s a 72-hour window to take antivirals to slow the spread.

So I kept asking.

“How is onset defined for shingles?”

“It’s only red now, no blisters. So is that before onset?”

“What are the symptoms of someone with shingles?”

By the end of the conversation, I had a working hypothesis, a sense of what to watch for, and the language I needed to walk into a doctor’s office with conviction. I couldn’t see my doctor in person, so I booked a virtual visit the next day.

The doctor was unsure. There were no blisters yet, just the prodromal pain and a faint redness. Diagnosing shingles without a rash is a coin flip; there’s no fast blood test for it. But valacyclovir is cheap, the side effects are benign, and the cost of being wrong is a week of unnecessary pills. The cost of being right and waiting is permanent nerve damage. He wrote the prescription.

Later that day, the blisters came in exactly where the model said they would. By then, I had already taken my first antiviral. Shingles still wasn’t fun, but I had a milder version than I would have had.

What this is, and what it isn’t

The LLM didn’t write the prescription or take responsibility for the decision. It generated a hypothesis fast enough that I could carry it into a real appointment before the treatment window closed. The doctor still made the call.

Understanding that it was likely shingles led me to meet with my doctor. A past version of me would have Googled “itchy red dots leg” and gotten ten web pages of contact dermatitis.

The standard pipeline (notice symptoms, make an appointment, wait 3 days, describe symptoms in 90 seconds, get a guess) is too slow for illnesses like shingles.

How to actually do this

Describe what you actually feel, not what you think it is. My first prompt was a mess of contradictory observations (”red spots... neither are red at all”). The model worked with it anyway. Be specific. Side, timing, character of pain, what has changed today and over time.

Ask follow-ups until the differential narrows. The first answer is rarely the answer. Iterate, ask again in different ways in new threads. Then ask about onset, treatment windows, and what symptoms would look like next. Each question prunes the tree.

Bring the conversation to your doctor as your own framework. Not “the AI said,” but “I’m worried about X because of Y and Z, can we rule it out?” Doctors respond to specific hypotheses much better than they respond to “I feel weird.”

Shingles

I know more than half a dozen friends who have had shingles in their twenties or thirties. All healthy men who work similarly intense jobs as me. For many of them, it was the worst pain they’ve ever had. I only learned they had it before when I told them I had it. None of them fit the textbook profile.

The textbook says shingles is for the elderly and the immunocompromised. The textbook is being outpaced by what’s actually showing up in clinics.

There’s also a pattern of these cases in men after the COVID vaccination. The mechanism is plausible (transient changes in cell-mediated immunity can theoretically reactivate latent varicella zoster).

The window closes fast

Antivirals don’t cure shingles. They shorten it and, more importantly, reduce the risk of PHN– nerve pain that lasts months or years after the rash is gone. I have mild PHN. It acts up when my workload gets particularly high. It’s like a spidey-sense, except it comes from my left leg.

If something feels off, have the first conversation with an LLM, and the second with a doctor.

Subscribe now

I went to a K-pop concert the other week.

I lived in Korea from 2009 to 2010. Back then, it was funny watching them copy and paste the boy and girl bands of the 90s and have it working over a decade later, after the US had moved on. They never stopped. Now in 2026, it’s caught on big here.

K-pop and you don’t stop

I worked with some K-pop groups for a while, helping bring them to America and bringing American artists to Korea. Got some ridiculous stories. But that’s not what fascinates me.

The concert in Austin did.

There are nine of them in the group, but two were missing, so the dance routines had a gap of two people. Then they introduced themselves.

First one speaks Korean. Translator says, “Today, I wore pigtails.” Crowd erupts in cheers.

Next one speaks Korean. Translator says, “Today I wore a ponytail.” I thought this was less impressive. It’s half the work that goes into pigtails. And with pigtails, you have to make sure they’re even. Unless they’re purposely asymmetrical. Nonetheless, the crowd goes wild. Then she speaks Korean again, and the translator goes, “Do you like it?”

Crowd goes apeshit. They like it.

There’s a reel of the same K-pop star throwing a keychain into a massive open backpack and doing a dance when it goes in. Bottom-quartile accomplishment.

It’s the infantilization of society. Yes, it’s impressive if they’re an infant. But they’re in their 20s.

In Korea, you have to study English. So they speak it. They’re just too shy to. Their grammar is better than most Americans’.

Not all K-pop or celebrities are like this. But a decent number are. We applaud them for doing the bare minimum. I noticed, even growing up, how many people were so impressed by stories of a celebrity being courteous to a service person. I give praise when my six-year-old niece shows gratitude. For an adult, this is expected.

Parasocial Relationships

They’ve always been odd to me. Why have a relationship with someone you don’t know? You can appreciate or criticize their ideas. But to actively love or hate them personally is a whole other level. You don’t know them.

The politician super against gay marriage is gay. And as many learned over the weekend, the stoicism influencer is emotionally reactive. If someone makes one thing their entire personality, there’s a reason for it.

Make Greatness Great Again

As a society, we pushed that fatness is brave and beautiful. GLP-1s ended that trend. But the narrative that nothing is better than anything else persists.

It’s healthy to celebrate greatness. It’s lame to celebrate mediocrity.

There are a million things to be great at. Even combinations of being good at things can make you great. Be great at something. Appreciate greatness.

Subscribe now

People are bad at abstract thinking. In 1968, Peter Wason ran an experiment that’s now one of the most replicated findings in psychology. Four cards sit on a table. Each has a letter on one side and a number on the other. You can see:

A, G, 7, 8

If a card has an A on one side, it has a 7 on the other. Which card(s) do you flip to test it? Stop and answer before reading on.

Fewer than 10% of people got this right. A 2017 meta-analysis pooling decades of replications put the rate at 19%. Still bad.

The answer is A and 8. Flip A to check there’s a 7 on the back. Flip 8 to check there isn’t an A on the back, because that would break the rule. The 7 doesn’t matter as the rule says A → 7, not 7 → A. G doesn’t matter for the same reason.

Most people pick A and 7. Confirmation bias in action.

Now the same puzzle, dressed differently.

You’re a cop. You walk into a bar. The rule is: if you’re drinking alcohol, you must be of legal age. You see four people:

A 12-year-old. A 35-year-old. Someone drinking soda. Someone drinking beer.

Who do you check? Stop and answer before reading on.

Nearly everyone gets this right. Ask the 12-year-old what they’re drinking. Ask the beer drinker their age. The other two don’t matter.

Logically, it’s the same puzzle. The finding has been replicated dozens of times. When this version was run on Harvard students in 1989, ~75% nailed it. Same logic.

When we wrap a logical structure in a familiar social context, people become competent reasoners. Strip the context out, and they’re guessing.

Rawson et al. (2015) found that students learn abstract psychology concepts better when paired with concrete examples. Fyfe et al. (2015) found that kids learn math equivalence faster, starting with a literal balance scale before moving to symbols. Physics teachers explain current with water in pipes. Programmers explain recursion with Russian dolls. Every good teacher already knows this.

It’s why the sciences are so hard for many. And it’s why the Richard Feynmans of the world are such amazing teachers. They relate everything to the real world.

Concrete examples are the best way to explain things. Even if first principles are most true, it’s fastest to learn in metaphors.

Understand the examples to understand the pattern, then extrapolate. As long as your primitives are correct, this is the best way to learn and to teach.

Subscribe now

AI switches between 200 IQ and 2 IQ. It’s easy to make fun of AI when it’s 2 IQ. And it’s easy to show 200 IQ moments, then say, “we’re cooked.” Both are true.

There are videos making fun of ChatGPT for confidently claiming it can tell time, even though it can’t. These will continue to go viral.

I was recently searching for a group I talked to over a year ago in my email. I couldn’t remember anyone’s name or the company name. Only high-level info on the group itself. I tried Gemini, which has access to my calendar and email. I tried multiple prompts describing what I remember about the group, and it couldn’t find them. I tried Claude Cowork, using the same initial prompt with access only to my email. It one-shotted it… crazy how much better it is.

Then, in my next Claude prompt, it confidently stated that a well-known company is private when it’s been public for a year. And yet, we still trust AI to do tons of tasks. So much so that it nudges us.

Nudge

Most people do the default. The most famous example is that if the default option is to be an organ donor, 99% will be organ donors (as seen in Austria, France, Hungary, Poland, and Portugal). If the default option is not to be an organ donor, 12% will sign up to be organ donors (as seen in Denmark and Germany). That’s an 87-point gap, a 725% change in behavior.

The default matters in life, as it matters in software and games. My favorite example is the game, Mario Maker. The Mario Maker series is a game where players can create custom Mario levels and play either their own or others’. In an interview, one of the creators said it was intentional that when you start the game, there are two options. Create or Play. You start off automatically highlighting Create. You have to move to the right to select Play. Similarly, in Mario Maker 2, the options are Make or Play, and you start off highlighting Make.

It prompts the player to consider making a game. This leads players to create more games than they otherwise would.

AI Nudge

With AI, querying and making are in the same place. You can build in Claude Code and do work in Claude Cowork, but you can also build in Claude Chat. AI asks how it can help us today. It’s much more of a choose-your-adventure than nudging.

When I made Robot Party, it took about the same time to make that game as it does to write most my essays. It was different in how the game development involved context switching to short tasks in between versus writing, which requires blocks of undisturbed time. Beating the game on all the difficulties was the only time-consuming part, but it was fun. I designed the game because I wanted to play.

I’ve always wanted to make Robot Party. I searched the App Store years back for a game like it and couldn’t find it. Now, I can reuse that code to build similar games like Paint It Red. And reuse the lessons I learned to create entirely new games, like this tank game.

Viral

In Mario Maker, some level creators became famous for their ingenuity. They built a brand. Other random levels are equally amazing, but the creators don’t have a brand, so the virality was a one-off.

In the age of building with AI, it’ll be the same. Random things will go viral, then fade away. Some will build brands around their creations. Most will just consume. Nudge yourself to build.

Subscribe now