The economy's numbers are right. They answer a different question

TIM: Jobless claims, out last week: a hundred ninety-four thousand. Near a fifty-seven-year low. And a lot of people still feel broke. That can’t be the whole story. So today’s question: are we watching the wrong numbers? My bet: we’re using old metrics to gauge new phenomena, and the numbers that tell you what’s really happening are hiding in plain sight. It’s Monday, November second. Let’s get into it.

TIM: Fortune says Cisco, the router company much of the internet runs on, just handed one of its biggest jobs to a twenty-seven-year-old.

ZURI: With “unlimited AI tokens,” in Fortune’s headline, to rewrite its old code. Fortune’s word, “rewrite.”

TIM: And the same story says entry-level jobs are vanishing. Here’s my take, and someone famous already said it.

ZURI: No, because it’s always a podcast.

TIM: Mark Cuban, the investor behind Cost Plus Drugs, said it best on Andreessen Horowitz’s podcast last week.

CLIP: Mark Cuban: “So, that’s part one. You can’t just get rid of all entry-level positions. Part two, you have to enable those people and everybody else, white-collar oriented, how to use AI because AI doesn’t replace jobs. People who know how to use AI replace the jobs of people who don’t know how to use AI.”

TIM: The jobs go to whoever knows how to use Super Intelligence. Cisco’s twenty-seven-year-old is my exhibit A.

HOLLIS: Can I see exhibit A’s résumé?

DEV: He cofounded a translation startup in twenty twenty-three. Cisco bought it in twenty twenty-five, and he came with it.

MARGO: A former CEO. Jeetu Patel, Cisco’s president, found him “five levels” below him, doing “something unstrategic.”

MARGO: Cisco’s org chart has a basement.

TIM: Okay. So he came in the side door.

MARGO: Through an acquisition.

TIM: I wrote that in January: when the front door closes, the people who find side entrances have less competition.

ZURI: He’s quoting himself. On his own show.

TIM: It’s a good line.

TIM: But one hire is an anecdote. Is the front door actually closing?

MARGO: The New York Fed counts it. Grads twenty-two to twenty-seven: about five and a half percent out of work. Everyone: about four.

HOLLIS: And every year from nineteen ninety to twenty eighteen, a new grad was less likely to be out of work than the average worker.

MARGO: Then in twenty nineteen it flipped, and the gap’s widened since. Now it’s a point and a half.

TIM: Twenty nineteen. So it started before this year’s headlines.

HOLLIS: Years before. And forty-two percent of the grads who are working are in jobs that don’t need the degree.

MARGO: Flip the five and a half, Hollis. Ninety-four percent of grads who want work have it. Nothing’s vanishing. The head start is what’s gone.

DEV: Here’s the weird part. By major, the lowest unemployment is nurses, teachers and civil engineers, about two percent or less.

TIM: And the coders?

DEV: Computer science, seven percent. Computer engineering, almost eight. Those are twenty twenty-four numbers. They graduated into the slump.

TIM: While software job ads come back?

MARGO: On Indeed, software postings are up about a quarter in a year. All postings, up three percent.

HOLLIS: On the twenty-seventh we said software was still about twenty percent under early twenty twenty. Still true, still climbing.

DEV: So learn SI and you’re in.

ZURI: The computer science kids learned to code. They’re the seven percent.

DEV: SI alone isn’t the job. And this one’s a goal, not a result. Two people, unlimited tokens, ninety-eight percent of the old code, two years.

MARGO: And nobody’s said what “unlimited” costs. That’s the one number I want.

DEV: He gave himself four weeks to prove it. Old code does things nobody wrote down, so the four weeks is the real test.

ZURI: And look what Cisco got. Jeetu Patel calls every intern “kind of a reverse mentor” and goes to campuses for “hungry” students.

HOLLIS: So it’s recruiting.

ZURI: No, because Cisco just got cool.

ZURI: Cisco. The router company.

TIM: So Mark Cuban’s point is who gets the work. So far, though, the re-sort is by field. Cisco’s guy came in the side door.

HOLLIS: And the head start went in twenty nineteen. Nobody was quoting that number.

TIM: Then let’s go find a number that moves before the headline.

MARGO: The number everyone quoted: in mid-October, the Nasdaq had its worst week since late June, down about four and a half percent.

TIM: Here’s my take. The companies building data centers have financing so complicated I can’t read it.

DEV: Nobody can from outside. Leases, guarantees, joint ventures, debt against the chips.

DEV: Legally.

TIM: So maybe the cracks in the credit market are where we should really be looking.

MARGO: The Compound’s guest from Barclays used your word, cracks. Then he drew a line.

CLIP: a guest from Barclays: “It’s about the fact that actually AI financing does have limitations. It may not apply today to hyperscalers who still are cash volcanoes. But it is certainly applying to, you know, levered sort of plays on forward compute.”

ZURI: “Cash volcanoes.” The limits land on the “levered sort of plays on forward compute.”

TIM: And did credit move first?

HOLLIS: This time, yes. a16z’s newsletter says the cost of insuring the build-out’s debt “has soared since the middle of the summer.” Stocks noticed in October.

TIM: So credit’s worth watching.

ZURI: Peter Berezin, BCA Research’s chief economist: tech is shifting “from being a cash generator to a cash incinerator.”

ZURI: Volcano to incinerator. Same fire, worse marketing.

MARGO: One sector’s debt, though. And a16z says part of it is supply: so much fresh issuance.

TIM: So what tells me if it’s spreading?

MARGO: The whole market’s fear. The Fed publishes it every month.

DEV: It’s called the excess bond premium.

DEV: I read the paper.

TIM: In English?

DEV: A spread’s the extra interest a company pays over the government. Subtract what its stock and debt say about default. What’s left is nerves.

TIM: So it reads the complicated part for me.

DEV: Default risk explains about seventy percent of the ups and downs in spreads. The leftover is the part that’s warned before recessions.

MARGO: September: minus point four. Lower than ninety percent of months since nineteen seventy-three.

TIM: Minus means calm?

MARGO: Calm. Since July, spreads widened a little, and the fear part fell by the same amount. Lenders are pricing each borrower, not panicking.

HOLLIS: RenMac, the research firm, put it plainly: “implied recession risk from the excess bond premium is under 10 percent.”

MARGO: On the twenty-second we said calm in the middle, choosy at the edges. Still true. The middle got calmer.

DEV: And it’s been early before. In July 2007 it turned positive, five months before the recession started.

HOLLIS: It cries wolf. By our count, a spike means a recession ahead a third of the time, one already here a third, nothing a third.

TIM: Nothing?

HOLLIS: The Fed’s own named false alarm is 2002. Accounting scandals, spreads jumped, no recession.

ZURI: No recession for who?

HOLLIS: Fair. That July WorldCom, the phone and internet giant, filed America’s biggest bankruptcy yet. The telecoms’ network build-out had lost much of its value.

TIM: And the bear case now?

HOLLIS: A strategist on Bloomberg Surveillance fears the “trickle of selling becomes a bucket getting dumped.”

MARGO: Right now the gauge is pointing the other way. That’s a market sorting borrowers in the middle of a boom. Which is what you want.

TIM: So credit’s the place to look, just not one sector’s spreads. The market’s nerves are steady. The edges are where the builders live.

ZURI: Steady nerves. Tell that to anyone still paying for six streaming apps.

TIM: Netflix. From its June twenty twenty-five high to October twenty-first: down about half.

ZURI: And it raised prices in March.

TIM: Here’s my take. What is any of them doing that’s interesting? Their own apps, higher prices, and they all feel the same.

HOLLIS: All of them?

TIM: I cut subscriptions the way I skip junk food at the grocery store. There’s just too much content.

ZURI: Your agents cancelled your Netflix. Then resubscribed to research it.

HOLLIS: Animal Spirits, the investing podcast, said it last week.

CLIP: a host of Animal Spirits: “Because right now, they’re just another tile on the screen. There’s nothing special.”

MARGO: Another tile on the screen.

TIM: So are they growing by being better, or by charging more? Margo?

MARGO: In March the ad-free plan went from eighteen dollars to twenty. Up about eleven percent.

TIM: And revenue?

MARGO: In the US and Canada last quarter, up eleven percent.

TIM: So in America the growth is about the size of the price rise.

MARGO: About. And members grew too: Netflix credits “increases in memberships and pricing.”

TIM: And the hours?

MARGO: Netflix’s own count: viewing this first half was a record, and up two percent on last year.

DEV: And revenue over those same halves, up about fifteen.

MARGO: So for every hour you watched, Netflix got about twenty-three cents last year. This year, about twenty-six.

MARGO: Same couch, higher rent.

DEV: And flat hours are what the stock’s afraid of. Netflix’s letter admits “investor concern about our view hours growth.”

ZURI: And people are cutting. Deloitte’s survey: forty-three percent cancelled a streaming service in the last six months. Gen Z, fifty-seven.

TIM: There it is. The junk-food cut.

ZURI: Just not Netflix. Netflix says its churn improved from a year ago, in the US too.

MARGO: Alicia Reese, Wedbush’s media analyst, asked whether cancelling is rising: “Across services, not Netflix, though.”

TIM: So people are cutting the other apps and keeping that one.

DEV: Then the problem’s the ads. Netflix cut this year’s ad forecast by about a tenth, on a slower rollout.

TIM: Can a better algorithm fix it?

DEV: Yes. Better recommendations, more hours. That’s a code problem.

ZURI: The algorithm is why it’s a tile. Animal Spirits asked how to be “more like Apple and HBO and less like an algo.”

TIM: Okay, here’s my maybe. People on GLP-1s, the weight-loss drugs. Healthier, off the sofa. Does that touch this?

ZURI: It’s a good hunch, and nothing measures it. Gallup, this summer: about one adult in nine is on one.

HOLLIS: And the only data close to it?

ZURI: Morning Consult: GLP-1 users are more likely to consider streaming services, not less. Its words: “not ascetics.”

TIM: So it’s a hunch. Fine. Then what’s the catalyst? A merger?

HOLLIS: Already happened, to the others. Paramount closed its takeover of Warner Bros. Discovery on October sixth. HBO Max and Paramount Plus, one company.

MARGO: With nearly ninety billion dollars of debt.

HOLLIS: And a correction. On the twenty-fourth we told you Paramount won Warner Bros. What we didn’t say: it had already closed.

TIM: Noted. And mergers as a catalyst?

HOLLIS: Anyone heard of AOL Time Warner? Two thousand, a hundred sixty-four billion, sold as old media’s future.

ZURI: AOL. The landline years.

HOLLIS: Two years later, a loss of nearly a hundred billion dollars. Old media’s future arrived early.

TIM: And the most famous one went badly.

DEV: And Netflix? If the hours start falling, the price has nothing to stand on.

TIM: So Netflix grows on price and members, and so far it’s holding.

HOLLIS: Speaking of numbers that hold. Jobless claims, still near a fifty-seven-year low. One of The Compound’s hosts quoted it too.

CLIP: a host of The Compound and Friends: “We had initial jobless claims down to 194,000, which is the fourth lowest level going back to the 1960s.”

TIM: That’s the number everyone quotes. Here’s my take. We may not be tracking the right numbers.

ZURI: Which ones?

TIM: There’s a ton of churn: lots of options, but horizontal moves. Participation’s too low. And housing’s slowed.

MARGO: And people still feel broke.

TIM: Right. When I go look at a job site in Atlanta, what I see doesn’t match what those numbers say.

DEV: So the data’s off?

TIM: The numbers and the reality are different. That’s a data-collection problem: we’re hanging on to old metrics instead of what’s on the ground.

ZURI: No, because the feeling is the data. It’s giving recession, and the claims say almost nobody’s getting fired.

TIM: So the jobs look good on paper, and everybody’s discontent. The numbers don’t match what people feel.

MARGO: Then take horizontal moves. The Atlanta Fed tracks the same people a year apart.

ZURI: And?

MARGO: Switch jobs, you got about four and three-quarters percent. Stay, about three and a half.

HOLLIS: A gap of about a point and a third. In late twenty twenty-two it was nearly three points.

MARGO: So the reward for moving halved.

TIM: Halved. So moving barely pays.

MARGO: It still pays. The thirty-year average gap is under a point. Early last year it was below zero. That was the sideways year.

TIM: Then what’s the churn?

HOLLIS: There isn’t much. It’s low hire, low fire. Claudia Sahm, the former Fed economist: “The low-hire, low-fire economy trudges on.”

DEV: On the twenty-seventh we said steady can just mean stuck. Still stuck.

MARGO: Now set the stayer against prices. A bit under three and a half. Prices, a bit over.

TIM: So the stayer’s running in place.

MARGO: About even with prices. Most people stay. That’s who feels broke.

TIM: Here’s the real discontent, I think. Most Americans had never lived through a noticeable amount of inflation before the last few years.

MARGO: Here’s the record. From ninety-two to twenty-twenty, not one year averaged four percent inflation.

HOLLIS: And half this country is under forty. As adults, they’d never lived through a four percent year.

MARGO: Since twenty-twenty, prices are up about twenty-nine percent. As much as the fourteen years before that, combined.

ZURI: It averages out, though, right?

TIM: Maybe it averages out to two percent over time. That doesn’t change how it feels to someone who’s twenty-five.

ZURI: Tara Sinclair, who chairs economics at George Washington, on Marketplace: for people changing jobs to get a higher wage, “those numbers aren’t looking so good.”

TIM: So moving pays, if you can get the move. And participation? I said too low.

DEV: Overall, about sixty-two percent. But ages twenty-five to fifty-four, about eighty-one percent have a job. Employ America calls that healthy.

MARGO: So who’s missing?

DEV: Partly retirees. It’s an older country. And in January the survey reset its population estimates, which pulled participation down.

TIM: So what would you watch instead?

DEV: Paychecks. ADP reads its actual payroll records: stayers up three percent, changers almost five.

ZURI: How do you know what’s in people’s paychecks?

DEV: ADP runs the payroll. It reads the checks.

HOLLIS: And ADP only made base pay its headline this year.

HOLLIS: Your new metric needs a metric.

ZURI: And housing?

HOLLIS: Census: homes completed in September, down fourteen percent on the year. Builder confidence this month, lowest since twenty twelve.

TIM: And rates are like inflation. They feel high next to the last twenty years, but historically they’re very medium.

HOLLIS: Freddie Mac, last Thursday: the thirty-year mortgage, seven and a quarter percent. The last twenty years averaged under five.

MARGO: Since nineteen seventy-one, it’s averaged about seven point seven. Half the weeks on record were higher than this one.

TIM: So it’s not high. It’s higher than we’re used to.

MARGO: Here’s the upside. Bank of America’s card charge-offs fell, and late payments improved six quarters running. People are paying their bills.

HOLLIS: For the record, this summer’s low was the lowest since nineteen sixty-nine. That year, the recession came in December.

MARGO: And in twenty twenty-two they got this low too, and nothing came. A stuck market with jobs beats a moving one without them.

HOLLIS: Fine. Base rate, not a forecast.

HOLLIS: I’ll widen the interval.

TIM: So the old numbers are right about firing, and quiet about the person who stays.

MARGO: One price in this economy keeps falling. Here’s Big Technology’s host, last week.

CLIP: a Big Technology Podcast host: “This is from Max Woolf. OpenAI and Anthropic cut their LLM prices by up to 80% within a few months.”

DEV: Max Woolf, a data scientist. His eighty percent is OpenAI’s cheapest model. And a token’s a piece of a word, sold by the million.

MARGO: Anthropic’s top model, Opus, went from twenty-five dollars to twenty, per million output tokens. Twenty percent.

TIM: Here’s my take. Token costs go to zero eventually, and I’m not the only one saying it.

ZURI: Zero zero?

TIM: We’ll stop tracking them, like we stopped counting cell phone minutes. They’ll be like cloud storage.

ZURI: Okay, you’re in company. Sam Altman, OpenAI’s chief executive, in twenty twenty-four: “towards intelligence too cheap to meter.”

HOLLIS: And Ethan Mollick, the Wharton professor, on Big Technology last week.

CLIP: Ethan Mollick: “I think that the token cost concern will go the way of, you know, of vibe coding and rag and other kinds of things because costs will decrease even as we get more compute online.”

TIM: Costs fall as compute comes online. So let’s do storage. Margo?

MARGO: Opus output was seventy-five dollars in May last year. Now twenty. Down about seventy-three percent in sixteen months.

TIM: And storage?

MARGO: Amazon’s standard list price: a little over two cents a gigabyte a month. Last cut in December twenty sixteen.

TIM: Ten years, same sticker.

MARGO: And nobody counts it. A terabyte’s about twenty-three dollars a month.

TIM: So storage never went to zero.

MARGO: It went too cheap to watch. That’s your end state. And Anthropic says it cut prices “while preserving our margins,” per Tom’s Hardware.

HOLLIS: Remember the storage war? March twenty fourteen. Google cut its storage price about sixty-eight percent.

DEV: And Amazon?

HOLLIS: Answered the next day, about half off. HP shut its public cloud less than two years later.

ZURI: HP. The printer people?

HOLLIS: Back then, yes. It announced the shutdown two weeks before it split in two.

HOLLIS: Price wars don’t kill demand. They kill whoever can’t keep up.

MARGO: And All-In’s version of that risk, for the labs.

CLIP: an All-In host: “So the price will keep going down, but the market size keeps increasing. I think the biggest risk factor to Anthropic, I do think it is some version of the open source models, maybe not catching them, but sort of saturating intelligence at a low price point”

DEV: DeepSeek, the Chinese lab with open models, lists its pro model at about two dollars per million output tokens, off-peak. Opus is twenty dollars.

ZURI: Off-peak?

DEV: It charges more in its busy hours.

ZURI: Nights and weekends? Tim, your cell phone minutes came back.

HOLLIS: And the flat plans?

ZURI: Not unlimited. Every Claude plan still has a weekly limit. And Max Woolf says OpenAI halved its two-hundred-dollar coding allowance in September.

MARGO: On the twenty-ninth we said Google’s top model price doubles December first, to twenty-four dollars for a million in and a million out.

HOLLIS: Same as Opus. The top isn’t falling to zero.

TIM: In February I said imagine all the costs go away, you don’t need people to buy tokens.

DEV: You said that about running the model on your own phone. Own the machine, no meter. Rent it, the meter stays.

TIM: Then it’s not zero. It’s storage. Metered, and too cheap to watch.

TIM: Except by my agents. They count everything.

HOLLIS: At the bottom. At the top, weekly limits and a price doubling in December. Somebody’s watching that meter.

TIM: Old metrics, new phenomena. The old numbers aren’t lying. They’re answering a different question. Claims tell you who’s getting fired, and almost nobody is. Netflix’s stock tells you what investors fear, not who’s cancelling. One sector’s bonds aren’t the whole market’s nerves. The numbers that say what’s going on sit right next to the headline: the grads’ head start, gone since twenty nineteen, and the raise for the person who stays, about even with prices. That’s most of us, and it’s not the number anyone quotes.