whatimreading

Tuesday, June 23, 2026 · 110 newsletters

Starmer Quits, Iran Pauses

uk politics · iran · ai models · china · polymarket · creator economy · critical minerals · voting rights · nyc · markets

Published on Tuesday, June 23, 2026.

Pulled from 109 newsletters in yesterday's inbox. Two governments wobbled, one model leapfrogged, and the people writing about all of it spent the day arguing about what just happened.

UK: Labour's Landslide Lasted Eleven Months

The dominant story across the desk. Bloomberg broke it on a live blog, then framed it everywhere else: Keir Starmer went from landslide win to swept aside inside a single political cycle, with the resignation expected imminently all weekend before he set the timetable on Monday. The pound moved, the bookmakers moved faster, and Andy Burnham opened up as the next PM, possibly unopposed.

The climate read is the one to watch. Bloomberg Green walked through what the resignation does to UK net-zero commitments at exactly the moment London Climate Week opens. David Callaway made the same point harder: the climate file was the one piece of Starmer's program that had real teeth, and now it goes into the same blender that ate his predecessors. Burnham's Manchester record is greener than the Labour right wants and bolder than the Reform-haunted center will tolerate. Pick one.

The Madho take: this is the seventh UK leader in ten years, and the through-line is no longer Brexit. It is that the British center cannot hold a narrative against Reform's pricing power on immigration, and every PM who tries to govern around it gets sanded down. Burnham inherits the same room with the same furniture. The interesting trade is in gilts, not in the polls.

Iran: Sixty-Day Oil Waiver, Inspectors Back In

The other big foreign story, told mostly through Bloomberg's evening sweep on major progress after all-night talks and the morning follow that had oil falling on the progress. Matt at WTF Just Happened Today had the cleanest summary of the mechanics: a 60-day Treasury waiver letting Iran sell oil, including to US buyers, while negotiators try to land a final deal; 78% of Americans saying the war should end now.

JD Vance is the face of this deal, and that is the point. Bill Kristol and Andrew Egger at The Bulwark flagged that Vance, not Rubio and not Hegseth, came out of Switzerland to announce IAEA inspectors will be allowed back in as a deal condition. That is the first significant Iranian concession of the cycle, and the administration is letting the heir apparent take the credit. Brian Beutler at Off Message sharpened the domestic angle: the loudest pro-war Jewish commentators of the last six months spent yesterday arguing nobody promised them this exit.

The takeaway: Trump wants the optics of having ended a war he started, oil prices want it more, and the negotiating leverage that came from bombing has a half-life measured in weeks.

AI: China Closes the Gap While Anthropic Sits Out

Three independent threads converged on the same point: the American lead is the smallest it has been in a year, and the proximate cause is policy, not capability.

GLM 5.2 is the model the open-weights crowd has been waiting for. Linas called it the "ChatGPT moment for local AI," citing Matt Velloso describing it as the first open model that "passes the bar as a daily driver" and Jeremy Howard putting it on par with Opus 4.8 and GPT 5.5. One-million-token context, MIT license, top of independent open-weight leaderboards, number one on Arena AI for frontend coding once Anthropic's Fable came off the board. News Items ran the Economist read on top: Zhipu's Beijing lab shipped a model at less than a tenth the price of Fable 5, with downloadable weights, and shares in the company are up sharply.

The Anthropic shutdown is the Chinese tailwind nobody planned for. The Algorithm at MIT Technology Review walked through the timeline: Anthropic flagged Mythos in April as a cybersecurity-grade coding model, released the safer Fable variant on June 9, the federal government slapped export controls on it three days later citing national security, and Anthropic revoked access within hours. James O'Donnell's point is the one that sticks: the doomers got their government intervention, but it came over a coding model, not a bioweapon, and the practical result is that European and US enterprises are now looking hard at GLM. Amazon's Andy Jassy reportedly lobbied the government that Fable was dangerous, which is convenient for Amazon's own Anthropic-backed and Anthropic-competing stack.

The application layer says it sees friends, not frenemies. The Information AI Agenda had Canva's ecosystem chief Anwar Haneef on the record arguing Claude Design and ChatGPT integrations have added millions of new users to Canva, the chatbots handle ideation and Canva handles the brand-kit-and-share work. Figma is reportedly less calm. The honest read: Canva is correct until the day Anthropic or OpenAI ships their own templates-and-team-assets layer, at which point the partnership math inverts.

AI Operators: The Loops Are Getting Real

A second AI cluster, this one about how working teams use the tools instead of how the tools are priced.

Loops, not prompts. Lenny Rachitsky ran Claire on how to design AI agent loops in Claude Code and Codex: heartbeats, crons, hooks, and goal-based scheduling, with a working daily PR-review loop and a weekly skills loop that spawns its own subagents. ByteByteGo's Part 2 with Shah Rahman, Meta's global head of autonomous ML iteration for Ads, did the same thing one altitude up: how do you organize a team when the engineer is now an orchestrator and the four core practices are context engineering, spec-driven development, critic patterns, and continuous evaluation.

The career-audit genre is here. Katie Parrott at Every pointed Codex at her own quarter of OKRs, got an unflinching read on what she shipped versus what she said she would ship, and wrote the most useful piece of the day on what AI is actually doing to knowledge work. Marily Nika called prompt engineering dead in the same breath, arguing the skill that matters is product framing.

The platform moves, meanwhile, are not happening in the EU. Ben Thompson tied Apple's first real iPhone price increase in years to the memory chip shortage, then made the harder point: Apple Intelligence is not shipping Siri AI in the EU. The DMA is now a feature gate, and the geography of who gets the good model is being drawn at the regulatory layer.

Connect the dots: the open Chinese model gets cheaper, the closed American model gets blocked or geo-fenced, and the operator-level conversation has moved from "which model" to "which loop." The American advantage is shrinking on both ends.

Markets: Polymarket's Fake Wash and an AI Bond Binge

Matt Levine put a methodical hit on Polymarket: a fresh report found roughly $130 million of wash trading on the platform across recent months, much of it tied to a handful of accounts gaming volume and rebates. His structural point is the one to bookmark: traditional markets separated exchanges from brokerages partly so the embarrassing retail behavior would not stain the exchange itself, and Polymarket collapsed the two. The fake traders are not a bug of decentralized prediction markets, they are the load-bearing wall the design refuses to admit it has. Tech Brew ran a shorter read in the same direction.

The other market story is debt. Bloomberg had SpaceX sliding ahead of an expected AI-financed bond binge, then followed up in the evening with the thinning-out frame on the eventual IPO. Read alongside The Information's reporting that OpenAI's balance sheet has $750 million in lease liabilities and zero debt heading into its IPO question, and the pattern is hard to miss: AI infrastructure is being funded with debt and equity tricks that the public markets have not yet had to underwrite. They will.

Politics: Trump Wave, DOJ Slump, Abortion Remembered

Bloomberg framed Colombia's presidential election as a Trump-aligned win that extends the administration's sway across the hemisphere. Paul Krugman used Phillips O'Brien's frame to argue the US military was already rotting before Trump II and Hegseth has accelerated it, citing heavy aircraft losses and a Pentagon culture optimized for warrior-spirit cosplay over readiness.

Domestic politics ran the opposite direction. Democracy Docket flagged two pro-voter rulings in a single day plus a Supreme Court decision not to hear a challenge to an anti-voting Arkansas law. Marc Elias made the broader case in the same outlet: the DOJ is losing in court on voter-database, mail-ballot, and citizenship-verification suits, and the cumulative effect could save the midterms even if Congress does nothing. JVL at The Bulwark and Bill Kristol both spent the day on the Reflecting Pool farce, with arrests and a peeling "American Flag Blue" coating, as a case study in shrinking authoritarianism. Rick Wilson called the cabinet a zombie presidency.

The piece worth flagging: The Bulwark on swing voters and abortion. The conventional wisdom from 2024 was that the issue had cooled. The focus-group evidence in 2026 is that voters in states with very restrictive bans still talk about it unprompted, in a "pro-life but pro-choice" register that maps to swing districts. Democrats over-learned the lesson of one cycle ago.

And one money piece. Judd Legum at Popular Information traced a network of sham "progressive" super PACs back to the Congressional Leadership Fund through shared vendors, a Tallahassee Staples mailbox, and FEC filings submitted in a 32-second window with the same Republican compliance software. That is the rare investigative piece where the receipts are conclusive.

NYC: Mamdani's First Real Test, AI Takes the Office Space

Gothamist had two of the day's NYC stories. The first: Mayor Mamdani faces his first political-strength test in Tuesday's congressional primaries, which he has explicitly framed as a referendum on what it means to be a Democrat in New York. The second: AI firms are eating Manhattan office space at a pace that has finally pushed the post-pandemic vacancy story off the front page.

A third piece worth noting: the Supreme Court restored the conviction in the Etan Patz murder, ruling the Second Circuit exceeded its authority in overturning Pedro Hernandez's 2017 conviction. NYC stories that close at the federal level have a long tail.

The take: Mamdani's primary slate is a leverage play, not a referendum. If his picks lose, he loses none of his mayoral capital. If they win, he gets a House caucus that owes him.

Critical Minerals: From Hypothetical to Recurring

Farrell Gregory at ChinaTalk won their economic-security essay contest with the cleanest framing of the year: after fifteen years of bipartisan attention, countless executive orders, and tens of billions in appropriations, the US is still overwhelmingly reliant on China for critical minerals, and in some categories the reliance has grown. Beijing's 2010 cutoff against Japan was supposed to be the wake-up call; the 2023 gallium and germanium controls were the second; the 2025 rare-earth restrictions finally forced US manufacturers to shut lines. The endgame argument: stop treating each round as a one-off and build redundant supply for the next decade as policy, not press release.

The Information Electric ran the practical counterpoint with Lithium Americas' Thacker Pass mine in Nevada, where GM has already claimed the entire twenty-year output of phase one, the federal government owns 5%, and production starts late next year. That is enough lithium to power 850,000 EVs or the equivalent in data-center batteries. Trivium China reported that Beijing responded by putting new export restrictions on US defense and rare-earth firms in the same week. The two stories are the same story.

Cannes Lions and the Creator Economy: Croisette Crowd

The festival opened to the largest creator contingent in its history. Marketing Brew put the count plainly. The Publish Press was on the ground with the "300 creators in Cannes this week" piece. Emily Sundberg filed from the South of France, namechecking the OpenAI partnerships team and the Day One Agency party. James Murray at Behind the CMO ran the Monday opener on the AB InBev Creative Marketer of the Year prize and the new Creative Brand Lion category.

The trend behind the trend is the one Casey Lewis at After School named: dopamine sites and tourist bait. The creator economy and the Lions are converging because both are measured in attention now, not effectiveness. The brands that used to fly to Cannes for trophies are flying to recruit collaborators. The trophies will follow.

Ideas Worth Reading

Outside Interests

Data Worth Noting

Three Takeaways for You

The British center is not coming back. Starmer's resignation makes seven UK leaders in ten years, and the proximate cause is now Reform's pricing power on immigration, not Brexit. Burnham gets the keys to the same furniture. Trade gilts, watch the climate file, and stop trying to call the bottom on UK politics.

The AI story moved from capability to access. GLM 5.2 ships at a tenth of Fable's price with a permissive license. Anthropic's most capable models are blocked by the US government. European and US enterprises that need to keep building are looking hard at Chinese open weights, and the operator-level conversation has shifted to loops and orchestration rather than which closed model is best. The American lead is now mostly a regulatory artifact.

If you only read three pieces today, I would point you at Farrell Gregory's "Critical Mineral Security: The Endgame" for the long arc, Matt Levine on Polymarket's fake traders for the structural read on the prediction-market boom, and Katie Parrott's "I Asked an AI to Audit My Own Career" for the most honest account this week of what AI is actually doing to knowledge work.