Hey Traders,

Man, we focus on intelligence an awful lot with AI.

Every launch comes with a scoreboard, and for two years the whole industry has argued over whose model is smartest. I’d argue the biggest AI story of the past year came from a company that doesn’t need the smartest model to win.

That company is Meta Platforms (META), and the product is Muse, the AI agent it rolled out in the U.S. last month. Muse books travel, sends emails, fills out forms and makes purchases for you, and it keeps working after you close the app.

This is the first of three pieces I’m sending you this week on why Muse matters. This one lays the groundwork on where the money in AI is actually heading. Later this week, I’ll get into Meta itself and how I’d think about positioning.

The IQ race is still on. I just don’t think it decides who gets paid anymore, because smarts are getting cheap, and fast.

Epoch AI, an independent research group that tracks this stuff closely, put out a report last month estimating that the cost of reaching a given level of AI performance has dropped about 13x a year since 2023. Stack three years of that and the same brainpower costs roughly 2,000 times less. Epoch says no world-changing technology has ever gotten cheaper this fast, and it puts computing from 1940 to 2001 at about 1.5x a year.

You can see it in real products. On Sept. 22nd, Anthropic released Claude Opus 5.5, cut the price 20%, and said typical work runs about 40% cheaper than on the last version because the new model finishes the job with fewer tokens (the small chunks of text AI companies bill by). Meanwhile, open-source models that anyone can download and run for nothing keep leaning on prices from below.

Anthropic even told customers to keep its pricier top-tier model, Fable, for the hardest problems and move everyday work to Opus.

Steve Martin would understand.

In Father of the Bride, his George Banks gets sticker shock at $250 a head. Martin Short’s wedding planner offers steak or chicken, and George takes the chicken.

When the almost-as-good model costs a fraction of the price, most people take the chicken too.

The labs will keep selling steak to anyone who needs the absolute best to crack a physics or chemistry problem. That’s a real business, but it’s a lot smaller than the chicken business.

That’s a problem for anyone selling smarts alone, because nothing makes a customer stay. When a cheaper or smarter model shows up, users switch in a couple of taps. Nobody’s loyal to a benchmark score (the test results every lab waves around at launch), and you can’t build a moat on one.

So where does the money go? Follow the people who spend it. The value flows to whoever owns their habits: the app they open first every morning, the assistant that knows their files and contacts and quietly knocks out dozens of chores a day.

Once someone weaves that assistant into how they run their life or their business, leaving hurts, and that hurt is the moat.

Any options market maker (the firms that stand ready to buy or sell to anybody) knows a version of this. The sharpest pricing model on the floor doesn’t guarantee you a dime. What pays is order flow, the customer who comes to you first.

AI model makers are starting to look like a pit full of market makers fighting over the same flow, and the fight gets more brutal every month. Whoever owns the front door decides where that flow goes.

Meta has 3.6 billion people using its apps every day, so it can skip the IQ contest. Muse just needs to make life easier for the people already standing in Meta’s hallway, saving them time and money and helping them earn some. Do that well and they stay put.

In the next piece, I’ll show you why Meta can afford to practically give the smart part away, and walk you through the 1990s browser war that shows how fights like this usually end. Let me tell you, that history should make a few AI labs sweat.

Here for a good time… AND a long time,

Hans