The $150B dollar business hiding in plain sight

Why does it matter? Because efficiency doesn't shrink demand — it detonates it

The conventional AI anxiety — that cheap code means fewer coders — has been empirically wrong every single time it's played out. The cotton gin, the steam engine, Nvidia's own CEO: they all point to the same counterintuitive truth. Make something dramatically cheaper and you don't need less of it. You need a mind-boggling amount more.

• AI will create more demand for code than it destroys — Jevons Paradox has proven this claim with every major technology wave in history
• Cargill does more annual revenue than Goldman Sachs, Nike, and Starbucks combined, and almost nobody knows what they actually do
• The middlemen who can't be cut out are the ones who built the physical infrastructure everyone else depends on — not just the ones who facilitate transactions
• Genuinely new product categories require manufactured demand before they can be sold: the Ice King gave free ice to bartenders before anyone in South America had tasted a cold drink

AI will create more coding jobs than it destroys — history has already proved this exact argument three times

Most AI job-loss panic is built on a logical fallacy that every major disruption has already disproved. Sam introduces the Jevons Paradox — "an economic principle stating that as technology makes use of a resource more efficiently, the total consumption of that resource actually increases rather than decreases" — and argues it's the only frame that matters right now.

His prediction is blunt: "I believe that code will get more cheap, therefore demand will increase a significant amount to the point that we can't even understand." Gutenberg's printing press didn't produce five times more books. It eventually produced Twitter, the Kindle, and a billion text messages a day. Nobody could have modeled that from 1450. The jump from input efficiency to total demand isn't linear — it breaks the scale you're trying to measure with.

Shaan adds the live version: people assumed each algorithmic improvement would reduce the need for Nvidia chips. Jensen Huang went on TV and corrected them. Training will get more efficient — "but the inference is going to go up by 1 million%." One million percent. Not a typo. The training cost drops. The total compute explodes. Jevons, again.

Sam's broader point is that the question itself is wrong. Stop asking whether AI will eliminate your field. Start asking where the new demand lands — that's where the opportunity is, and it's always bigger than anyone predicted.

The cotton gin made one worker 50x more productive — then caused slavery to expand 10x. That's the math AI is about to run on your industry.

Here's the darkest proof of the principle. Eli Whitney builds a handheld device that separates seeds from cotton mechanically. Before it, one person produces about a pound of usable cotton a day. After: 50 lbs. Fifty times more efficient. Some people at the time thought — hopefully — that maybe this would reduce the need for enslaved labor.

The exact opposite happened. Cotton got so cheap that it became America's dominant export — they started calling it King Cotton, displacing King George in the South's imagination. And to keep up with that exploding demand, the country imported "8 to 10 times more slaves than we currently had at that point pre Eli Whitney." Sam's read: "one could argue that the reason why slavery in America lasted as long as it did was because of cotton and the demand."

This is the mechanical version of the lesson. You don't model the efficiency savings. You model the demand explosion. Those are the numbers that determine the actual business outcome — and they're almost always larger, stranger, and more consequential than anyone forecast.

When a new technology is compressing costs in your market, that's the right question to run: not how much cheaper does this make my existing product, but what does total demand look like when the price floor collapses?

Cargill makes more than Goldman, Nike, and Starbucks combined — and your parents have definitely never heard of them

When Shaan Googled the Cargill family, the first words that came back were "silent dominance, middlemen at planetary scale." That's the ballpark.

$150 billion in annual revenue. Largest private company in America for the last 40 years. 88% family-owned. More billionaires in a single family than any other company has ever produced. Around $3 billion in profit in a normal year — roughly $600 million flowing to the family in dividends under the 80/20 rule: 80% reinvested, 20% out, no exceptions.

The reach is almost comic when you trace a single hamburger. Cargill probably sold the farmer the seed and fertilizer to grow the grain, bought the grain, stored it at their own grain elevator, shipped it on barges they built in their own shipyard to processing plants where it became animal feed, fed it to cattle, slaughtered them, stored the meat at their own packing plants, and sold the beef to the restaurant. The restaurant's salt? Cargill. The corn syrup in the ketchup, the soybean oil in the fries, the starch in the milkshake? All Cargill.

And it was strategic: "It was part of their strategy for a long time to be quiet about it just to not attract competition and to build this monopoly position in the logistics industry." The company went from a single grain shed to controlling 25% of the entire US export market for grains — and did it while staying off CNBC, off podcasts, and operating out of a château on a lake in the middle of nowhere. Money talks, wealth whispers.

You can always cut out the middleman — unless the middleman literally built the road everyone has to use

The conventional wisdom on middlemen: they're fragile, they get disintermediated, don't build there. Cargill controls 25% of US grain exports and nobody has touched them in 160 years. The reason is structural, not clever.

They were physically in the middle. They didn't just connect farmers to railroads — they built the grain elevators right next to the railroad tracks. You needed to get your grain to a train. Who else had built the storage? Nobody was going to do that. "The middleman was physically in the middle" — not informationally useful, but structurally irreplaceable.

From there it compounded. Barges were inefficient? They built their own ships, eventually producing vessels for the US Navy. They bought the biggest poultry company. They own mines to extract their own chemical inputs. They run a $10 billion hedge fund to manage commodity exposure at scale. Each extension made the next exit harder for any farmer or processor trying to route around them.

Sam's father runs the same model with onions — broker between farmers and trucks, 24-hour storage window instead of months. Same logic: own what everyone has to pass through.

The question worth asking about any middleman position: am I informationally useful, or am I structurally embedded? The first is always one API call from obsolete. The second is a multigenerational empire.

There's a formula for how long AI disruption lasts — and the answer might be shorter than any previous wave

Every major technology has a turmoil window — the gap between when old jobs disappear and when new demand creates new ones. The Luddites smashing weaving machines in England weren't just angry, they were risking execution (capital offense). 800,000 telephone operators evaporated when switchboards automated. Bank tellers thought ATMs would end them — while the total number of bank branches actually surged.

Sam maps a formula: turmoil duration equals the breadth of the technology (how many people it touches) multiplied by intensity (how deeply it changes their work) multiplied by co-invention time — how long it takes to build the adjacent infrastructure the new technology requires to deliver its full value.

Railroads took 50 years to reshape the economy because you had to physically lay thousands of miles of track and invent a new steel industry to do it. That co-invention lag stretched the turmoil window across generations.

AI is different. ChatGPT diffused faster than any prior technology. The co-invention requirements are mostly software — no steel mills, no laying track. Shaan's conclusion: "there is a world where the tumultuous period is relatively short compared to past breakthrough technologies." If you can estimate the co-invention drag, you can start positioning for the recovery before the turmoil closes.

Twenty billionaires, one family, 160 years — the secret is running a dynasty like a company with board meetings and an 80/20 rule

A company founded before the Civil War, still 88% family-owned, still generating billions. What most families get wrong: they don't talk about money. Parents deflect. Culture never gets named. "If you asked them how much do you make, they would say that's none of your business." What stays implicit doesn't get inherited — it decays.

Sam's alternative: run a family meeting the way you run a board meeting. State values explicitly. Report against goals. Allocate resources transparently. Cargill's 80/20 rule makes this structural — reinvest 80%, distribute 20%, no arguing, no exceptions. The Hearst trust went further: dispute the terms and you're out of the will.

Sam is already running a version at home. Each kid gets a personal operating principle — one is called "the Sam way: try, try, figure it out" — repeated until it becomes identity. He ignores the inconsistencies, keeps reinforcing the moments that fit the frame, and lets the reputation do the work over time.

The Cargills have professional CEOs now. The heirs are marine biologists. The entity outlasted every individual. That doesn't happen by accident — it happens when values get spoken aloud and governance gets written down before anyone needs to fight over it.

The Ice King's first customers had no idea what to do with ice — so he seeded the bartenders for free until they couldn't go back

Frederick Tudor lives near Boston in the 1800s. He learns South America has never had ice. He decides to sell it to people who have never touched it, can't conceive of wanting it, and will complain that it melts.

The logistics alone were absurd: harvest blocks from frozen lakes, pack in sawdust to slow the melt, ship for a month, arrive with 35% losses, try to unload the remainder to customers who just didn't get it. He had to explain that ice melts. That's the product.

The breakthrough wasn't logistics. It was demand creation. He went to bartenders and handed out free ice with one condition: serve rum cold. "Once you've had cold rum, you never want to go back to that warm piss again." The felt need didn't exist before he manufactured it — but once someone tasted a cold drink on a hot day, the demand was immediate and self-sustaining.

Tudor eventually put himself out of business when people figured out they could just freeze water locally. He'd solved the problem too completely. But between seeding the bartenders and that collapse, he ran a monopoly on a category that hadn't existed — built entirely by giving the first sip away free.

The next empire belongs to whoever maps the demand explosion — not the efficiency curve

Every story in this episode points the same direction: the winners weren't the ones who made things more efficient. They were the ones who correctly predicted how demand would explode once efficiency arrived — and positioned themselves in the path of that explosion before anyone else saw it coming.

Cargill didn't invent grain. Tudor didn't invent cold. The cotton gin didn't belong to the plantation owners. The opportunity in every case went to whoever understood, one step earlier than the crowd, where the demand was about to land.

That's the real AI question. Not "will this take my job" — but "when code costs nothing, where does the demand go?"

---

Topics: Cargill, Jevons Paradox, AI and jobs, middlemen, family dynasties, industrial revolution, history of technology, business strategy, demand creation, private companies