Vol. I, No. 10
Covering 24 April - 7 May 2023
Monday, May 8, 2023
Late edition · A Relentless publication
The Relentless Dispatch
All the fortnight that mattered, in technology and in the world, read next to what we were building at the time.
BANKING

First Republic fails and is sold to JPMorgan before dawn; the Fed raises rates and hints at a rest

The third large American bank to fail in two months went to the biggest bank in the country, which was not supposed to be allowed to get any bigger.

Regulators seized First Republic Bank in the early hours of May 1 and sold it, the same morning, to JPMorgan Chase. It was the second-largest bank failure in American history, larger than Silicon Valley Bank in March, and it followed the same script: a bank stuffed with uninsured deposits and low-rate assets, a wealthy customer base that could move its money with a tap, and a run that regulators could not stop once confidence went. First Republic had limped through eight weeks on a $30 billion lifeline from eleven larger banks, but when it reported at the end of April that more than $100 billion in deposits had fled, the end came within days.

JPMorgan, already the largest bank in the country, got First Republic's branches, its wealthy clients and its loans at a discount, with the government absorbing much of the risk. Its chief executive, Jamie Dimon, said "this part of the crisis is over." The rules are supposed to stop the biggest bank from getting bigger by acquisition; in a failure, over a weekend, with no other bidder willing, the rules bent. The lesson three failures deep is uncomfortable: the safest place to keep money in a panic turned out to be the institution too big to be allowed to fail, which is exactly the concentration the post-2008 rules were written to prevent.

Two days later, on May 3, the Federal Reserve raised its benchmark rate a tenth time, to a range of 5.00 to 5.25 percent, the highest since 2007, and Jerome Powell removed from his statement the language promising further increases. It was as close to "we may be done" as a central banker gets. Whether he is done depends on inflation, which is easing slowly, and on how many more First Republics are hidden in the regional banking system, which nobody knows.

HOLLYWOOD

The writers go on strike, and for the first time a union is bargaining over AI

Eleven thousand screenwriters put down their pens on May 2. Among their demands is one nobody would have written into a contract two years ago: that a machine not take the work.

The Writers Guild of America went on strike on May 2, its first walkout in fifteen years, shutting down late-night television immediately and threatening the autumn season. The dispute is mostly about money, and about the way streaming has hollowed out the middle of the profession: shorter seasons, smaller rooms, residuals that no longer add up to a living. But one demand marks the fortnight as a turning point. The writers want guardrails on artificial intelligence: that studios not use it to write or rewrite scripts, and not feed writers' work into models that would then compete with them.

The studios' initial response, that they would meet annually to discuss advances in technology, was read by the room as a refusal, and it landed as one. This is the first time a major union has sat across a table and bargained over whether a generative model may do its members' jobs, and both sides know that whatever gets written into this contract will be quoted in every other industry within the year. The writers are not, mostly, afraid that a chatbot will write a great script. They are afraid that a studio will have a chatbot write a bad first draft cheaply, and then hire a writer at a lower rate to fix it, and call the writer's real work a rewrite. That is not science fiction. It is a pricing strategy, and it is the exact move I expect to see attempted in every knowledge business, including my own.

CORONATION

Charles III is crowned

Charles III was crowned at Westminster Abbey on May 6, in a ceremony a thousand years old and trimmed for a diminished, sceptical age: shorter than his mother's, more faiths represented, fewer aristocrats, an invitation for the public to swear allegiance from their sofas that many found presumptuous and declined. Some 2,000 guests watched inside; a smaller crowd than expected stood in the rain outside; anti-monarchy protesters were arrested under a public-order law passed days before, which troubled even supporters of the crown. He is 74, and has waited longer for the job than anyone in British history.

AI

The godfather quits Google to warn about his own life's work

Geoffrey Hinton, whose work on neural networks in the 1980s and after made the current wave of AI possible, told The New York Times on May 1 that he had left Google so that he could speak freely about the dangers of the technology he helped create. He said he now feared that the systems were becoming more capable faster than he had expected, that they would flood the world with false text and images, and that they might, further out, pose a genuine threat. Three days later Vice President Harris met the chief executives of OpenAI, Google, Microsoft and Anthropic at the White House and told them they had a responsibility to make their products safe. The word from a founder carried further than the meeting.

IN BRIEF

Chegg; the debt clock; a state bans an app's rival; Sudan

Chegg, a homework-help company, lost nearly half its stock-market value on May 2 after telling investors that ChatGPT was hurting its business; it was the first publicly-traded casualty named as such. The US Treasury warned that the government could run out of money to pay its bills as soon as June 1 unless Congress raised the debt ceiling, turning a slow standoff urgent. IBM's chief executive said the company would slow hiring for roles it believed AI could do, some 7,800 jobs over time. In Sudan, the war entered its third week with no ceasefire holding and the UN warning of a humanitarian collapse.

The Column

Too big to fail, and now bigger

The third bank failure of the spring was solved by handing the wreckage to the largest bank in the country. The cure and the disease are starting to look like the same thing.

By Rebecca Cho

When regulators seized First Republic in the small hours and sold it, the same morning, to JPMorgan Chase, they solved one problem by deepening another, and the trade is worth staring at because it is the defining bargain of modern finance. The immediate problem was a bank run, and it was solved cleanly: the depositors were protected, the panic contained, the failure absorbed before markets opened. But the price was that the biggest bank in America got bigger, absorbing a wounded rival at a discount with the government backstopping the risk, and the rules that were supposed to prevent exactly that, the post-2008 architecture built specifically to stop the too-big-to-fail institutions from growing through crises, bent quietly out of the way because in the moment there was no better bidder and no time to find one.

Here is the pattern, and it is the one this paper keeps finding under different headlines. We spent fifteen years building rules to reduce the concentration that made 2008 so dangerous, the handful of institutions so large and so interconnected that their failure threatened everything. And then a crisis arrives, and in the crisis the fastest, safest, most rational move is always to make the concentration worse: to hand the failing thing to the strongest survivor, because the strongest survivor is the only one big enough to swallow it over a weekend. Every individual decision is correct. JPMorgan was the right buyer; the depositors had to be protected; the deal did stop the panic. And the sum of all those correct decisions is a banking system more concentrated after each crisis than before it, which is to say more fragile in exactly the way the rules were written to prevent, assembled one sensible emergency at a time.

The deepest irony is what it teaches ordinary savers, and it is the wrong lesson delivered by the right outcome. The message of this spring, three failures deep, is that in a panic the safest place for your money is the institution too big to be allowed to fail, because that is the one the government will always save, because saving it is the only option. So money flows toward the giants, which makes them bigger, which makes them more essential to save next time, which makes the concentration the rules feared grow with every round of the very crises the rules were meant to prevent. First Republic did not just fail this fortnight. It fed the thing that will make the next failure harder to contain, and everyone involved did the right thing at every step, which is precisely why it is so hard to stop.

Field Notes
A Relentless build, told plainly

Shipping a construction-management app to the Salesforce AppExchange, and what the market taught us

The column is about writers fearing a machine will do the cheap first draft. Here is a product I took from idea to market, and the first-year numbers, because building the thing teaches you what no strategy deck will. No client is named.

By Alosh Paul

A few years back I led a product from a blank page to a live listing on the Salesforce AppExchange. It was a construction-management app, built because the industry it served had a specific, expensive, unglamorous problem: projects ran over. Budgets slipped, timelines slipped, and the slippage was usually invisible until it was too large to fix, because the information about it lived in a dozen spreadsheets and site foremen's heads and never reached the person who could have acted a week earlier. We built the software that made the overrun visible while it was still small. On the projects that used it, overruns came down by about thirty-five percent, and in its first year on the market the product generated on the order of a hundred thousand dollars in revenue, with a companion iOS app that put it in the hands of people who were never at a desk.

I ran the whole arc: the architecture, the development team, and the part most engineers underrate, the go-to-market. And the go-to-market is where I learned the thing I want to write down, because it is the opposite of what you expect. The hard part was not building the software. The hard part was that construction firms did not want an app. They wanted their overruns to stop, and an app was a cost and a change and a training burden standing between them and that outcome. Every instinct in a product team is to talk about features. Every buyer I met wanted to talk about the number, the thirty-five percent, and whether it was real, and what it would cost them in disruption to get it.

That reframed how I build. A product is not a set of features; it is a promise about an outcome, and the features are just the machinery that delivers the promise. The AppExchange listing that worked was not the one that listed what the software did. It was the one that led with what stopped happening to your projects when you used it. The demo that closed was not the one that showed the cleverest screen; it was the one that ran on a project like the buyer's own and showed them their own kind of overrun getting caught in week two instead of month three.

There is a reason I am telling this story in the same issue as the writers' strike. The screenwriters are afraid a machine will do the cheap first draft and devalue the real work. Maybe. But the reason my construction app sold was never the code, which a capable model can increasingly write. It was the years of understanding why construction projects overrun and what a foreman will actually open on his phone at seven in the morning. That understanding is the product. The code was always the easy half, and I say that as someone who has shipped a lot of it.

The Ledger
AI
A Google founder quits to warn about AI; the White House summons the labs; a public company blames ChatGPT for its collapse; a union strikes partly over it. The technology moved from novelty to labour question in a single fortnight.
Data centers & power
The compute demand under all of this keeps climbing quietly. Nvidia reports later this month, and the number everyone is waiting for is data-centre revenue.
Rates
Fed at 5.00-5.25 percent and signalling a pause. Three bank failures in eight weeks have tightened credit on their own. The debt-ceiling X-date of June 1 is now the market's main fear.
Real estate
First Republic's failure was a deposits story, but the commercial-property loans on regional books are the slower one underneath. Mortgage rate 6.39 percent (May 4).
India tech
IBM's plan to slow hiring for AI-replaceable back-office roles is the shape of the thing India's services industry has to answer. The roles most exposed are the entry-level ones the industry runs on.
What we called wrong
In No. 6, on Silicon Valley Bank, we asked whether the deposit guarantee would stop the run at one bank and left the answer to the markets. Three failures later the answer is plainly no; our trust-and-confidence framing held, but we did not foresee the shape of the cure, that a crisis born of concentration would be resolved by concentrating the wreckage into the largest bank of all. The run we saw coming; the consolidation we did not.
The Back Page

Talk to the hand

Jerry Springer died on April 27, at 79, and the tributes were more affectionate than his show ever pretended to be. Before the chairs were thrown he had been the Democratic mayor of Cincinnati, a serious local politician who resigned over a scandal, won re-election anyway, and might have gone further. Instead he spent three decades hosting the loudest, lowest, most-imitated hour on American daytime television, a carnival of paternity tests and folding-chair brawls that he introduced with a shrug and closed, every single day, with a straight-faced little sermon he called the Final Thought. He knew exactly what it was. "I just want to get to heaven on a technicality," he liked to say. He apologised, more or less, for coarsening the culture, and then took a bow, because the audience was his and it was enormous and it loved him. Take care of yourselves, he signed off, every day for twenty-seven years, and each other.