Aug 31, 2026 · 58:32
Unqualified Fact Check
held up
Two guys who told you in the first minute they weren't masters of the Fed did considerably better on the Fed than on the Greeks. The universities one is going to sting for a while.
On this week's episode I tell Dan about something that had been haunting me. The events in the Strait of Hormuz feel eerily similar to a story captured 2,800 years ago. I had just seen Nolan's version of The Odyssey [1] and been left with the feeling that Hormuz feels a lot like a modern-day Troy. A trade civilization built on a handful of honor rules. A horse rolled up to the gates. Somebody breaks the one rule that made the whole thing work, and everything that comes after comes through that gap. Dan's answer, which I'm still chewing on: "this sounds like post-World War Two." We came back to Troy twice more. It ends the way it always ends. Rome cleans up.
That's the back half of a conversation that starts somewhere much drier. Dan opened by asking what on earth is going on with the Fed and the Treasury, and I did the responsible thing and pretended nothing was happening — thirty-year rates only back at 2007 levels, one arm of the government telling markets to speak freely and the other quietly buying bonds so they don't speak too freely. Then Dan said the quiet part out loud in the best line of the episode: Warsh says let the market speak, and Bessent says not that loudly. There are charts, an argument about whether what Bessent did counts as QE, and a detour on why a thirty-year mortgage is priced off the ten-year note — which I sent straight to the fact-checkers, because I was guessing.
Then it turns into a build episode, and Dan makes an argument about agents I haven't been able to shake: you start a new conversation, and the thing knows everything until it knows nothing. It has to be born into existence every time. Which, he points out, is also us — we just happen to run about eighty years, and none of us can checkpoint our context either. That leads to his digital Library of Alexandria, my record store you can walk into on the internet, his rebuild of the whole investment process at Kabu, and the saddest sentence in modern software: all I've got is this little blinking cursor. We close on whether "artificial" is doing any work in "artificial intelligence," and whether a person is allowed to hold two contradictory ideas at once. Reader, we are.
Go build something and contradict yourself.
[1] If you haven't yet, you really should go see it in the theater.
Sean: Said above 5.3 percent, at levels unseen since 2007, from memory at the top of the episode. The 30-year Treasury touched 5.33 percent on August 17-18, 2026 — a 19-year high, and the highest since 2007. Correct to the decimal.
Sean: Put the Treasury action at around the middle of the week, August nineteenth. Treasury announced the upsized buyback on August 19, 2026, raising the maximum operation size to at least 4 billion dollars from 2 billion, aimed at 10- to 30-year maturities, with operations scheduled September 9 through November 4. One nuance skipped on air — the 19th was the announcement, and the actual buying had not started yet.
Sean: Treasury relaunched its regular buyback program in May 2024, the first since the early 2000s. This matters because it is the load-bearing beam under Dan's both-sides-are-implicated point, which was the whole purpose of the segment.
Dan: He apologized for his pronunciation twice and then got both titles and both authors right. The Myth of America's Decline — Politics, Economics, and a Half Century of False Prophecies by Josef Joffe, Liveright 2013. When Reason Goes on Holiday — Philosophers in Politics by Neven Sesardic, Encounter 2016. Word for word.
Dan: Said the Tyler Cowen episode had just been released the previous week. It went up August 19, 2026, four days before we recorded. And it is in fact the Rick Rubin show, which he got to on the third try.
Sean & Dan: Dan asked, Sean agreed, and Sean described it as putting dollars out into the ecosystem. Treasury buybacks are funded by issuing other debt, mostly shorter-dated bills, so no new money enters the system — it is a duration swap, much closer to Operation Twist than to QE, and only the Fed can create reserves. Where they are right, and it is the part that matters: it does pull long duration out of the market, which is exactly the pressure point on the long end. Credit for the mechanism, docked for the money-printer framing.
Dan: The best six-word summary of Warsh's position anybody has produced, and not a Warsh quote. What he actually said at the July 29, 2026 press conference was that market participants are learning to play the ball, not the referee, and that the central bank need not always and everywhere be the center of attention. Same argument, better sentence. Fair paraphrase, wrong set of quotation marks.
Sean & Dan: The one Sean sent to the fact-checkers himself. Dan said the typical mortgage hold is about seven years and Sean guessed not much over eight. The seven-year average life is real — it is the prepayment convention that makes agency mortgage-backed securities trade off the 10-year instead of the 30-year, which means Dan's actual point about pricing is correct. But homeowner tenure has stretched well past it: Redfin's 2026 read puts the typical US homeowner at about 12 years in the house, thanks to everyone with a 3 percent mortgage refusing to move. Right convention, right conclusion, stale number.
Sean: Sean had Odysseus smuggling in the Spartans. The force inside the horse was Achaean broadly — Odysseus of Ithaca leading, under Agamemnon of Mycenae — and Menelaus of Sparta is usually on the list, but the Spartans is a stretch. Then: civilization collapsed after the war was won, by Rome. The Late Bronze Age collapse is real and roughly contemporaneous, with Troy's destruction layer around 1180 BC, but Rome's traditional founding is 753 BC, four centuries later, with Archaic Greece rebuilding first. He nailed the hard parts though — Odysseus, Agamemnon, the feigned departure, and that scholars put Troy in modern Turkey at a site beginning with H.
Sean: The claim was that historians think early Christian communion wine may have carried hallucinogens. That is Brian Muraresku's The Immortality Key from 2020, which builds on psychoactive residue found at a site in Catalonia and on the Eleusinian mysteries. It is a real argument that got a real hearing. It is not the consensus of classicists or historians of early Christianity. He hedged twice, which is exactly the right amount of hedging, so he keeps the yellow.
Sean & Dan: Dan answered instantly — in the elites — and Sean specified the top 2 percent wanting to be the top 1 percent. The direction is well supported: Crane Brinton's The Anatomy of Revolution calls it the desertion of the intellectuals, and Peter Turchin's elite overproduction is the modern version. The crisp 2-percent-envying-1-percent formulation isn't from anywhere in particular, though — it is a good compression, not a finding. Sean flagged his own uncertainty out loud, which is why this isn't red.
Sean: The claim was that when you go there it is supposed to be the repository of universal knowledge. Lovely, and wrong. It comes from universitas magistrorum et scholarium — the whole body of masters and scholars. In medieval Latin universitas meant a guild or corporation, not a universe of subjects. The original university was a labor union with better robes.
Bessent keeps intervening in the long end at increasing scale — bigger and more frequent buybacks — through the November 2026 midterms, treating the long rate as a political variable to be held down rather than a market signal to be read.
Democrats take control of both the House and the Senate in the November 2026 midterms.
“get everyone sworn in after the midterms in February. The Dems are now in control of both houses. And you can just point the finger at them now. Their problem. It's their fault.”
Inflation runs hotter from here, not cooler — both hosts expect another leg up rather than a glide back to target.
The structural fiscal and institutional problems do get addressed — by Gen X and millennials — but not for another 10–15 years, and only after enough financial pain forces the conversation. Nothing gets fixed at the moment it should be fixed.
“I really hope that the gen X and millennials still have enough oomph… to say that we're actually going to try to right some of these holes one day. And it may still be 10 years in the future, 15 years in the future." (On the mechanism): "When you have large collective problems like this, they almost never get fixed when it's time to fix them. It's pretty much never. You have to start feeling pain. Like the bloodletting has to start before the reaction sequence can begin.”
Every executive precedent set in this period gets reused by the other party rather than retired — the Obama-era precedents Trump has used, and the Trump-era precedents the next Democratic administration will use. The ratchet only turns one way.
“we can talk all about how many different precedents were set by Obama that have now been used by Trump, or how many precedents Trump has now set that I'm sure the next Democrat will try to employ to our own detriment, right? So this is where… precedent matters.”
Better and more abundant digital records will not let societies escape the cycle of forgetting — abundance fragments the record rather than settling it, so future generations pick among competing "truths" that are bought and sold, instead of converging on one.
Prompt quality becomes decisive again — not for humans typing into a chat box, but because agents now execute prompts on your behalf, unread and at scale, so the prompts buried inside an agentic system become the highest-leverage artifact in it.
“prompts really mattered for a while and the prompt engineering thing became a thing. And then prompts felt like kind of went away. And then you realize as you build systems, the prompts matter again because the agents are running the prompts on your behalf. And those prompts matter a great deal." (On his own failure): "I was not reading word for word every single prompt that these agents are given in the sort of investment committee process I've built.”
The AI-writing controversy is short-lived. The market segments cleanly and stays segmented: people keep choosing human prose where the reading is the experience — fiction, storytelling, art — while academic papers, how-to articles and reference writing go to machines without much resistance.
“this sort of hubaloo about it is gonna be short-lived. It's not important. Where it is important, people continue to choose human prose where it's fiction and storytelling and weaving and art and life and it's something to be lived while consuming it. And then there are academic papers and how-to articles.”
The word "artificial" falls out of common use — the field and the culture settle on "intelligence" or "machine intelligence," on the argument that what these systems do is real intelligence arriving from somewhere else.
Warsh's reduced transparency — fewer FOMC meetings, less telegraphing, a Fed that is harder to read — is a deliberate strategy to *protect* Fed independence rather than an erosion of it; he is an antagonist to Fed convention, not to the institution.
“I think everyone's up in arms about how this is ruining Fed independence. And I'm kind of thinking it's protecting the Fed. … He might be an antagonist, but he's an antagonist to how things have been done, not to the actual institution itself. … I don't believe Warsh wants that. I think he wants to not completely be understood. Like this isn't a mistake is what I'm trying to say.”
Long-duration bonds bought at current yields are a good entry — clip the coupon while waiting, and capture price appreciation when rates whipsaw back down over the following six to twelve months.
“That's why I'm saying it's time to go ahead and buy some bonds. Maybe I clip a couple of nice coupons along the way, and then in six, twelve months I get my an extra thirty percent in — Not thirty, but I don't know.”
The agent interface becomes the operating system — the primary layer through which people touch every other tool — and it must be a *horizontal* layer that survives changes of model provider and harness, not a vertical stack per vendor.
