whatimreading

Thursday, July 9, 2026 · 116 newsletters

The Ceasefire That Wasn't

Iran War · NATO · AI Models · Immigration · Crypto · Data Centers · Venture · NYC · China

Published on Thursday, July 9, 2026.

Pulled from 115 newsletters in yesterday's inbox. The Iran ceasefire the White House spent June negotiating fell apart before lunch, and every other story bent around the crater it left.

War & Diplomacy: The Ceasefire That Wasn't

The truce collapsed by afternoon. Bloomberg, Semafor, and Bloomberg Opinion all filed by end of day. Trump, speaking from the NATO summit in Ankara, told reporters the memorandum with Iran was "over" after Central Command launched fresh strikes overnight in response to attacks on commercial vessels in the Strait of Hormuz. Semafor's Shelby Talcott caught the money quote: "we'll probably hit them hard again tonight. I'll give them a little warning." Oil futures jumped 6% to $75 a barrel. The Bloomberg Opinion take was blunt: another ceasefire bites the dust, and this one had barely learned to walk.

The politics is now moving faster than the diplomacy. Matt Berg at Crooked Media noted Trump spent the presser publicly ruminating on his own kill-list status while the IMF warned the war is dragging on global growth and oil is set to rise 14%. Bill Kristol and Andrew Egger's morning read at The Bulwark was that the administration's own explanations, "Iran will only reap benefits if they exhibit good behavior," read as improvised. Paul Krugman wrote from a different angle: European leaders in Ankara have quietly stopped treating Trump as a peer and started treating him as the senile uncle they need to route around, which is the actual story of a "de-Americanization" already underway across defense procurement and cloud.

Beijing chose the same news cycle to test a missile. Latika Bourke reported that a Chinese nuclear-powered submarine launched a nuclear-capable ICBM in the South Pacific, with Xinhua giving Australia two hours' notice. The Australian PM's response was, per Latika, unusually strong. Iran is not the only theater making moves while NATO is convened; the Ankara summit's real backdrop is a Pacific one.

The receipts are catching up. Judd Legum at Popular Information priced the Iran War at $103 billion across 120 days, based on the administration's own supplemental funding request, roughly triple the $30 billion OMB Director Russell Vought floated to House Appropriations last week. Three separate cost estimates in three months, none with supporting documentation. What all of this converges on is that the war's fiscal and inflationary tail is doing the real work in shaping the midterms, not the ceasefire theater itself.

AI: A Weird Week for the Model Names

The naming has completely broken. Techmeme's Wednesday cut led with SpaceXAI launching Grok 4.5 in partnership with Cursor, described by Elon as "roughly comparable to Opus 4.7, but much faster." OpenAI's GPT-5.6 is set to go live today. Anthropic extended the Claude Fable 5 preview through July 12, and Ruben Hassid's very on-brand take, "do not use Fable 5 if you don't know how," walked through why the "Mythos-level" tier is now a naming layer stacked above the Haiku/Sonnet/Opus flavors. Frontier labs are simultaneously shipping named tiers, numbered tiers, and mythical-noun tiers, which is starting to feel like a UI failure with pricing consequences.

The interesting AI story is one level below the model name. ByteByteGo wrote a clear explainer on the agent loop, framing an agent as "an LLM placed inside a loop where the model itself decides when the loop should stop," which is the concise version of why everyone is suddenly writing about harnesses. Stephanie Palazzolo at The Information's AI Agenda, filing from ICML in Seoul, quoted OpenAI's Noam Brown saying he would pick GPT-5.6 over a human research intern for most tasks. Nate at Nate's Substack wrote the counter-move: for eight dollars he ran a fake company with a review board, an appeals process, and a focus group, and shipped his wife's website despite one employee fabricating thirteen quotes. The 500-year-old trick, per Nate, is bureaucracy. You do not need to trust the agent if you can institutionalize its distrust.

The safety conversation moved from prompts to infrastructure. Ken Huang published two pieces yesterday: one arguing the security frontier for agentic AI is the infrastructure, not the model, and another decoding Anthropic's new Transformer Circuits paper on "J-space", the sparse verbalizable slice of internal activation space that lets the model report and steer selected concepts. The wrong headline, per Huang, is "Claude is conscious." The right one is that a small interior working surface is now something builders can inspect and address. Sidebar's daily design roundup pointed at "never mind the prompts, here's the thinking" and Anthropic's interpretability page in the same breath, which is the correct reading order.

Old-guard tech is getting squeezed. Ben Thompson at Stratechery wrote about Microsoft's Xbox division cutting 3,200 jobs and divesting five studios in what he called the abject failure of the Game Pass bundling strategy. Alex Wilhelm at Cautious Optimism noted the Iran War's intensity will be inversely correlated to tech exit frequency: rising oil, no rate cut, no hot IPO market. And in the same edition, Blue Origin is raising $10 billion at a $130 billion valuation, its first outside round, which is Bezos telling the room he intends to answer SpaceXAI on his own capital.

Immigration: ICE Turns Lethal

Two Bulwark pieces broke the same story from two different angles. Adrian Carrasquillo reported that Lorenzo Salgado Araujo, a 52-year-old small-business owner who had been in Houston for 35 years and had put three kids through college, was shot and killed by ICE on Tuesday morning while picking up workers, five minutes from the FIFA Fan Festival site. His son said Lorenzo had been in the process of obtaining a work permit through the legal process. Bill Kristol and Andrew Egger's morning column tied it to a pattern: ICE is shooting people again, and the ratio of enforcement escalation to actual criminal-record threshold is now well past what the administration's public messaging suggested six months ago.

The Bulwark ran a third essay pulling the frame wider. A lead editorial argued the Supreme Court's decision in Trump v. Barbara on birthright citizenship, largely covered as an immigration ruling, is actually a broader win for anyone born in America. The 14th Amendment binds the government to a rule it cannot opt out of when politically convenient. Read the three pieces together and the coherent argument is that the enforcement machinery has decoupled from any recognizable rule of law, and that the Court is, occasionally, still willing to say so.

Finance: The Crypto Legacy Bleeds Into the Real Market

Yueqi Yang at The Information Finance framed it as the market story of Trump's presidency. Beyond the personal $1.4 billion in family crypto income, the administration has legitimized stablecoins, prediction markets, tokenized equities, and now perpetual futures fast enough that they are starting to look like serious markets in their own right rather than crypto sidecars. Bankless reported Robinhood's new L2, pitched as tokenized-stock infrastructure, has been overtaken in its first days by a memecoin surge, an accidental confirmation that the underlying rails are working. Sam Boboev at Fintech Wrap Up rounded up the stablecoin banking research and a piece on tokenization's strategic role for Canadian banks. The Breakdown, in its most cerebral piece of the week, revisited the 1981 R. Foster Winans insider-trading case to ask when information itself becomes property, which is exactly the question tokenized markets keep sneaking past regulators.

Money for the Build: AI's Capital Stack Gets Weird

Dakin Campbell at The Information found the buried market of the week. Rule 144A private placements, a Depression-era safe harbor for institutional bond resales, are now the preferred vehicle for data-center developers who want public-market liquidity without full public disclosure. Three publicly traded developers tapped it in June alone. He called it "$3.5 trillion hiding in plain sight," which is roughly what it looks like when the buildout runs faster than the balance sheets can carry.

On the other side of the AI P&L, Ramp is going after inference spend. Jan-Erik Asplund at Sacra published the update: Ramp hit $1.5 billion in annualized revenue in May, up 89% year-over-year, at a $44 billion valuation, and is now positioning token spend management as a permanent CFO line item. On the venture side, Guillermo Flor at Product Market Fit has Lovable in talks to raise $300 million at a $13.2 billion post-money, roughly double its December mark, marking five rounds in twenty months. And GTMnow walked through Vanessa Larco's new fund Premise, built as a "companies OpenAI won't kill" thesis after watching Sora deprecated and ChatGPT checkout quietly retired within months. The private-market plumbing for AI is now sophisticated enough that the frontier labs are one of several exit routes, not the terminal event.

NYC: The Buildings Are Talking

A Midtown high-rise nearly came down on Tuesday. Gothamist's daily lead was the near-collapse and its uncomfortable overlap with NYC's office-to-apartment conversion craze; the same edition covered the city's search for a BQE "czar" to rebuild the ramshackle triple-cantilever section. Emily Sundberg at Feed Me filed a scene report from Bushwick's PubKey the same night, layered with a lament about how the physical residue of a heat wave disappears the moment the temperature drops, which is the local-politics version of the climate story nobody sustains attention on. Both pieces underline the same point: New York's infrastructure is aging on a shorter clock than its housing supply can absorb, and every conversion is a bet against that clock.

Ideas Worth Reading

Outside Interests

Data Worth Noting

Three Takeaways for You

The ceasefire's collapse is the day's headline, but the more durable read is fiscal. A war costing three times what the administration is admitting, oil pushing back toward $75, and no interest-rate cut in the queue. That trio determines the fall, not the diplomacy.

The AI cycle has entered a phase where naming, safety, and infrastructure are the three fronts that actually matter, and each of them looks different depending on whether you are a builder, a compliance officer, or a treasurer. Grok 4.5, GPT-5.6, and Fable 5 are the surface. Rule 144A, Ramp's token spend line, and Anthropic's J-space paper are the substrate.

If you only read three pieces, I would suggest Pathetic in Ankara by Paul Krugman for the geopolitical frame, Stop Waiting for AI You Can Trust by Nate for the operator's playbook on agents, and He Lived Here for 35 Years. Put Three Kids in College. ICE Killed Him. by Adrian Carrasquillo for the human stakes of the immigration story now unfolding beneath the noise.