Good morning, happy Friday, and the most interesting thing I heard this week is that one horse does not equal one horsepower. Go figure?

James Watt invented the term “horsepower” around 1780 to sell his steam engines, and he set the bar at what a horse manages over a full day of plodding—a deliberate lowball to make the machines look good. A real horse, in a short burst, puts out closer to 15 horsepower. The whole unit was just a marketing number.

Moving on. In the next 5 minutes over your coffee: how gaming hours are key to training AGI, does AI actually kill jobs, how FIFA profits from which teams advance, why AI is arming the rebels against Amazon, and the bird-speed app that somehow hit 300K users.

General Intuition, a lab spun out of the gaming-clip app Medal, just raised $320M at a $2.3B valuation. They also turned down a $500M buyout from OpenAI. The bet behind the valuation is that text models like ChatGPT and Claude are good with words but clueless about how things move through space and time, and billions of hours of gameplay are the missing training data. In one demo, a robot trained on their game-built “world model” navigated an office it had never seen with just 8 minutes of footage.

Maybe the resource in AI is shifting from text (which is all mostly scraped already) to data about physical cause-and-effect. And video gamers generate tons of it for free. It's the same as LLMs learning language from the internet, but pointed at physical AI and robots. It’s also why Bezos, Fei-Fei Li, and Yann LeCun are all pouring money into “world model” labs right now.

Proprietary data is the moat, not your model. Everyone rents the same models so the edge is a dataset nobody else has. Ask, “What data am I generating that is unique to my product?”

Maybe you don’t want to write a newsletter for yourself—but your product almost certainly needs a direct line to its people. That's the founder move: use beehiiv to build in public, own your subscriber list (no algorithm deciding who gets to see you), and turn it into a real channel you can monetize when you're ready.

Monetize with paid subscriptions, one-click ads, boosts, or digital products—and keep 100% of all of it.

It's the full-stack open operating system I run The Diff on, and if you use my code THEDIFF30, you get 30% off your first 3 months.

Marc Andreessen recently spoke about the lump-of-labor fallacy—the belief that there's a fixed amount of work to go around, so every machine that does a job destroys one. His example was the car wiped out the entire horse economy (stables, farriers, feed) and yet employment exploded, moving into building cars, paving roads, motels, road trips, tourism. The work just moved.

Both extremes are usually wrong. “AI changes nothing” ignores the real, painful displacement (ask a farrier in 1910). “AI ends all work” assumes the pie is fixed, which it has never been. New capability creates new demand which creates whole new categories we can't see yet. This basically says specific jobs go, the aggregate grows.

Don't bet your product on “AI replaces X.” Bet on the new work it creates—every automation wave spawns adjacent jobs and tools (the people paving the roads, not the ones shoeing the horses). Ask what new workflows and roles your product could own once AI makes the old task cheap.

When OpenAI took ChatGPT mobile, they ran the subscriptions on RevenueCat and went from signed deal to #1 in the US, UK, and Germany App Stores within a week, absorbing millions of new subscribers without breaking a sweat.

Their infrastructure already handles 2B+ API requests a day, so traffic spikes are a non-event. As OpenAI's finance lead put it: “with RevenueCat, we never had to slow down.” If it scales for ChatGPT, it'll scale for whatever you're building.

I found out the other day that FIFA runs its own resale exchange for the World Cup, and takes a handsome ~30% of every marked-up ticket, with prices floating on demand. So when Norway knocked out Brazil out (an epic game), the cheapest seat to the Miami quarter-final dropped by over ~$1,000 within hours—and FIFA's cut fell right along with it, a seven-figure swing across the stadium on one result.

The body that governs the competition now has a direct financial stake in which teams advance—Argentina or England in the bracket is worth materially more to FIFA than Egypt or Norway. Nobody's shown they're tilting games, and I'm not saying they are. But this is the same organization behind the 2015 US racketeering indictments and the Qatar bid mess, and when the house refs the match and skims the resale, you shouldn't have to take fairness on faith.

The red card worth stealing: be wary of any platform that both sets the rules and profits from specific outcomes—an app store competing with its own developers, a marketplace ranking its own products first, a model provider grading its own benchmarks. When the referee has a P&L tied to the result, incentives bend, whether anyone means them to or not.

The big e-commerce fear was AI becoming the middleman, where it buys everything for you and every brand gets reduced to an invisible API call. But data shows it's shaking out the opposite way: AI is behaving like a referral channel (a better search engine), not a distributor, and it's sending 4–8x more traffic to brand.com sites than to the marketplaces (Amazon, Etsy, eBay). AI is “arming the rebels.” if you will.

That flips the doom case—higher-intent shoppers landing straight on you is a de-risking event, the same reason Google and Meta got rich on referral ads instead of owning checkout. And the best tell for why agents won't just auto-buy everything: the milkman. Home milk delivery peaked at ~30% in the '60s and fell to zero; Amazon's Subscribe & Save is ~1–3% of GMV. People don't want their pantry on autopilot—agents help with the considered buys, not the toothpaste.

Treat AI as a new top-of-funnel that rewards being the clear, direct answer—structured product data, real reviews, markdowns made for agents to read, a clean brand.com the model can confidently point to, not a listing buried in a marketplace.

I love random apps. And Roost is one of them. It’s a “slow-cial” app where messages travel at the literal speed of a bird—send by falcon and it's quick-ish, by hummingbird it takes hours. It is beautifully and deliberately painfully impractical. Built solo by a Ticketmaster PM with Claude Code, it went from 10,000 to 100,000 users in three days after one mom's Threads post, and it's about to cross 300,000 in five weeks.

This is the pull of absurd novelty. The idea is so far off the map of solving a real problem that it's fun and viral—and fun is what makes people screenshot it and text it to a friend. A slightly-better messenger never goes viral; one that runs at pigeon speed does. Novelty is one of the last things that still earns free attention at this scale.

Novelty is a spike, not a moat. A message that takes two days to land is a great gag, not a problem solved, so the viral wave probably fades once the joke gets old. Use the weird hook to get in the door, then give people a real reason to stay—or just enjoy the ride and bank the spike. No problems there.

› For anyone traveling this summer—don’t let the days blur together, and don’t worry about having to journal everything down. Just use the photos you’re already taking, and capture one little moment in under 60s everyday. (Download the app, and get a movie at the end of my trip).

› Nobody likes the guesswork on how you’re doing compared to others—and access to data that answers that for your product can be a chore to sift through, which is why I did it and packaged it for you in our free set of tools— the app revenue/economics calculator, an app health score meter, and the competition heat calculator from up top (See how your app compares).

› My go-to stack with the vetted tools I trust everyday — the products I genuinely lean on to build things at work, my apps, and run my newsletter business (Explore my personal app store of recs).