Monday, May 18, 2026 · 101 newsletters
Anthropic Pulls Away
anthropic-revenue · agent-economy · trump-china · cassidy-purge · agentic-payments · iran-hormuz · ai-skepticism · fintech-ant · creator-economy · lifestyle-nyc
Published on Monday, May 18, 2026.
Pulled from 48 newsletters sent to read@madho.net yesterday. A lighter Sunday inbox than most weekdays, but a strikingly coherent one. Two gravitational stories: Anthropic's blowout Q1 numbers and the agent-economy reckoning that arrived with them, and the political aftermath of Trump's Beijing trip set against a Louisiana primary that purged Bill Cassidy from his own party. Here is the signal organized by trend.
The Big AI Story: Anthropic Pulls Away, and the Agent Premise Gets Rewritten
By volume and by significance, the dominant thread of the day. Alex Banks at The Signal led with a remarkable disclosure set: Anthropic's Q1 2026 revenue grew 80x year on year, with annualized run rate now estimated above $44B, $200M committed to a four-year Gates Foundation partnership, and a first-ever lead over OpenAI in verified business customers (34.4% versus 32.3% per Ramp's May AI Index). Funding talks reportedly approach a ~$900B valuation. PwC formally expanded its alliance to certify 30,000 staff on Claude, with a global rollout targeting its 364,000-person workforce.
It is worth saying out loud what the run-rate number means. A year ago the question was whether Anthropic could survive as an independent foundation lab; today the question is whether its enterprise lead is durable enough to justify a valuation roughly equal to the GDP of Switzerland. The PwC and Ramp data points matter more than the headline revenue, because they suggest the lead is being earned at the buying-decision layer where companies actually pick a default model, not just in benchmarks or press cycles.
The model wars produced a clear winner this week. Peter Yang ran a long interview with Anthropic research PM Alex Albert on how the team turns user feedback into training, decides which capabilities to ship, and shapes Claude's personality. Bloomberg's AI issue carried a feature on the Claude frenzy driving record Mac Mini demand as users roll their own agent stacks on Apple's entry-level desktop. And Every's Context Window ran "Opus 4.7 Reels Us Back In" by Laura Entis, in which Cora general manager Kieran Klaassen made Opus 4.7 his default for synchronous work after weeks of Codex dominance. The story everyone keeps trying to write is that the leader changes every quarter; the story they should write is that the customer reaction time is now measured in weeks. (Wall Street has not priced this in.)
The personal-agent thesis is being rewritten in public, which is the more interesting story. Every's Brandon Gell and Willie Williams published a retrospective titled "We Gave Every Employee an AI Agent. Here's What We're Doing Differently Now," arguing the problem was not the OpenClaw harness, it was the premise that every employee needs a personal agent. They are now rebuilding Plus One 2.0 around shared, reliable coworkers. Pair this with The Vibe Marketer's writeup of Hermes, the open-source agent that hit #1 on OpenRouter (224B tokens in a single day, surpassing OpenClaw's 186B), and you can see the industry quietly migrating from fragile single-user agents to stable, memory-equipped, multi-channel infrastructure. The 2024-2025 vision of every knowledge worker getting a copilot turned out to be wrong in the same way "every car gets self-driving by 2020" was wrong. Sorry, "Autopilot." The shared-infrastructure version is much less photogenic but probably much more durable.
Skepticism is now a coherent literature, not a contrarian pose. Ruben Hassid on outsourcing your brain to AI: you can outsource your thinking, but you can't outsource your understanding. MIT Technology Review surfaced two unsettling pieces: AI chatbots are giving out people's real phone numbers via Google AI, and three things in AI to watch from Nobel-winning economist Daron Acemoglu, who is more cautious than the consensus about the jobs-apocalypse story. Hannah Zhang at Nonlinear News ran the practical counter: build an AI second brain, a folder of markdown files telling agents who you are and how you sound, before the slop drowns you. The fact that the most interesting AI writing this week is about how to use it well, not about what it can do, tells you where the curve has bent.
Agentic Payments: Eight Standards, One Fed Warning
A surprisingly tight cluster, which usually means the story is more advanced than the consensus thinks. Rich Turrin at Cashless mapped eight competing agentic-payment standards launched in 2025 and argued Google has the structural edge, while flagging that the Fed has elevated AI to No. 3 on its financial threat list with concern doubling in six months. Simon Taylor's Fintech Brainfood led with Anthropic's small-business finance agents and asked whether Anthropic becomes "the world's biggest fintech company in a year or two." Zack at Tearsheet covered LendingClub becoming Happen Bank, plus a micro case study on how American Express is underwriting AI agent error via its ACE Developer Kit, verifying intent before execution. Sam Boboev at Fintech Wrap Up ran a long deep dive on Ant International's converged stack: 2 billion consumer accounts, 150 million merchants, 300+ payment methods, 220+ markets. It is the most coherent vision of agent-era commerce infrastructure anyone published this weekend.
The interesting thing about a standards war is that eight standards means there will eventually be one or two. The fragmented map Turrin describes is not the destination, it is the friction phase. Underwriting agent error, as the Amex move does, is what tells you which side is taking commerce seriously; the standards talk is what you do while you wait for the liability rules to settle.
Politics: The Cassidy Purge and the Math of the Map
Saturday night's Louisiana primary was the political through-line. Sen. Bill Cassidy, one of seven Republicans who voted to convict Trump in the 2021 impeachment trial, finished third in his own primary per John Ellis at News Items, with Trump-endorsed Rep. Julia Letlow leading 45 to 28 over state treasurer John Fleming. George Bounacos at Gov Brief Today framed it as the second Trump-driven primary purge in twelve days, with Rep. Thomas Massie likely to lose Tuesday in Kentucky and Trump now publicly threatening Lauren Boebert. Lincoln Square covered Rep. Darrell Issa's H.Res. 1211 to expunge both Trump impeachments from the House record, sponsored with 22 GOP cosponsors and Judiciary Chair Jim Jordan's support. Jonathan Cohn at The Bulwark ran a long piece on all the ways Trump policy is making it harder, not easier, to have a child in America, the Moms.gov launch notwithstanding.
Bruce Mehlman's six-chart Sunday gave the cleanest read on the redistricting math: prediction markets have moved Democrats' odds of retaking the House from 86% on April 3 to 74% today, with Cook's Amy Walter forecasting a ~10-seat swing toward the GOP from redistricting alone. Six weeks ago Democrats held 217 safe-or-leaning seats and needed one of 16 toss-ups; now they hold 207 and need 11 of 18. The map is the message: the same political environment that helped Democrats six weeks ago has been partially neutralized by structural drawing of lines, which is the story of American politics in 2026 more than any single election outcome.
The Cassidy result is being read as a loyalty enforcement story, and it is. But the deeper point is the speed: two purges in twelve days, a third teed up Tuesday, and a constitutional move to retroactively expunge impeachment is not a faction inside a party, it is a party rebuilt as a faction. The Republicans who think this stops with Cassidy are reading the wrong precedent.
China: The Meeting Was the Message
A two-source convergence worth flagging. Bruce Mehlman read Trump's Beijing summit with Xi as positive vibe, few specifics, and per the China Theory of Everything reframe, the actual deliverables Trump wanted are unglamorous but real: Iran pressure on Hormuz, tariff truce extension, ag exports, rare-earth and magnet access. Dexter Roberts at Trade War read it harder: state visit ends with few deliverables but heavy pageantry, Boeing's 200-plane order disappoints market, Nvidia still has no green light to sell H200s into China, and Beijing stiffed Washington on Iran help. Both pieces noted Xi's Taiwan warning and Trump's framing of Taiwan arms sales as a "bargaining chip."
The honest read is that decoupling is still the long-term strategy and constructive engagement is the tactical pause. (The timing here, set against the Iran war and the inflation print, is not subtle.) Calling Trump's Beijing reception "a bargaining chip" presidency rather than a coherent China policy is harsh, but probably accurate.
Iran and the Strait: A New Tax on Global Trade
Two notable items. Gov Brief Today led with Iran's announcement of a plan to charge tolls for Strait of Hormuz passage and bar US Navy operations as the blockade enters month two; 78 ships have been redirected and four disabled per CBS News. The USS Gerald R. Ford returned from the longest carrier deployment in 50 years, spanning the Maduro capture and the Iran war, with a 30-hour onboard fire damaging 600 bunks. Latika Bourke from Tallinn interviewed Estonia's foreign-intelligence chief Kaupo Rosin on Russia's weakening hand in Ukraine, China's growing leverage over Moscow, and the consequences of Trump's NATO posture.
A toll regime on Hormuz is not a normal escalation; it is Iran reframing the war as a tax on global trade rather than a regional conflict. If even a fraction of it sticks, the inflation print becomes structural rather than transitory, which is the part the markets have not absorbed yet.
Fintech and the Reframing of the Bank
A coherent supporting cluster. Jason Mikula at Fintech Business Weekly ran a long post-mortem on Synapse with new on-the-record material from Yotta's CEO ("Synapse is gonna f everything up," said in advance of brokerage migration), plus updates on Varo, Evolve, and Plaid's intelligent finance survey (55% of consumers have used AI for money tasks in the last 12 months). Zack at Tearsheet made the smart point on LendingClub's Happen Bank rebrand: the name change is the brand catching up to what the business already became, and the "Motivated Middle" feedback loop (debt consolidation lifts cash flow, cash flow lifts credit scores, scores unlock pricing) is a strong product thesis for digital banking past the neobank-as-checking-app era. Paired with Sam Boboev's Ant International piece, the takeaway is consistent: the interesting digital banks now are the ones with a converged stack, not a single product.
Ideas Worth Reading
Paul Krugman on the European decline narrative. A useful follow-up to his earlier essay arguing that the standard productivity-growth comparisons are misleading, with replies from Noah Smith and Luis Garicano. The data Krugman keeps surfacing on this is more interesting than the consensus take.
Noah Smith with dating advice for men. An economics blogger writing dating advice should be embarrassing; this one is not. Smith makes a serious case that the dating recession (one-third of young adults projected never to marry on current trends) is a public-policy question and worth treating like one.
Eleanor Warnock at Every on "Socrates as a Service." The argument: in a world where AI can search anything, the people who know how to extract tacit knowledge (the stuff that isn't on the internet) are getting more valuable, not less. Seven interviewing techniques, plus a case for taste and judgment.
Shane Parrish on Brain Food. This week's three tiny thoughts: urgency as a predictor of success, self-care as the input to everything else, and detailed long-horizon planning broken into daily checkpoints. Shorter than usual, which is a feature.
Hannah Zhang's AI Second Brain build. The most practical "personal AI infrastructure" piece I've seen this year. A folder of markdown files that any agent on your machine can read, telling it who you are, how you write, what you're working on, and what rules to follow. 45 minutes to set up.
George Bounacos on the 471st consecutive nightly news brief. A reminder that the best news products in 2026 are the ones run by one person who refuses to stop. The independent newsletter as civic infrastructure is real and underrated.
Outside Interests
Brick on the Union Square Greenmarket protein guide. Part three of her Union Square survival series. American Pride Seafood for $10/lb whole sea bass, $12-a-dozen oysters, and the cheapest live lobsters in the city in late spring. Useful regardless of whether you cook fish.
The Culturist on how to become a writer. Four principles for becoming a writer in the Substack era. The frame is romantic without being precious.
DrawTogether on Dorothea Lange's trick for seeing more closely. Wendy MacNaughton on how Lange used captions and field notes as a tool for attention. A short piece about the discipline of looking.
Polina Pompliano profiles Marcello Hernandez. The GQ profile of SNL's breakout Gen Z star, framed by Polina at The Profile as the comic built for both TikTok-era fame and old-school stand-up.
Neil Pasricha interviews Dr. Laura Markham on peaceful parenting. Chapter 46 of 3 Books, recorded in a Park Slope brownstone. On choosing presence over reaction with children. (A minority interest, I know, but the questions are good.)
The GIST on PWHL Finals pre-game rituals. Ottawa Charge vs Montreal Victoire for the Walter Cup, with players from every postseason team on their superstitions. The PWHL's audience is doing the thing where it doubles every year and nobody quite notices.
Data Worth Noting
Anthropic at 34.4% of business AI customers vs OpenAI at 32.3% (per Ramp's May AI Index). Up from 9% in May 2025. The first time Anthropic has led on this metric, and the rate of change is the point.
Hermes hit #1 on OpenRouter with 224 billion tokens processed in a single day, past OpenClaw's 186 billion. 140,000+ GitHub stars in three months since the February launch. The fastest open-source agent adoption curve I've seen.
Maritime Analytica on Maersk Q1: more cargo, less margin. Container shipping is once again the leading indicator no one watches. Hapag-Lloyd's Q1 swung negative; CMA CGM is dropping $820M into Kenya; Maersk's CEO is warning Hormuz consequences are not priced in. If you want to know what global trade is doing, watch the carriers, not the customs data.
Three Takeaways for You
The AI story has visibly bent from "what can the model do" to "what infrastructure does an enterprise actually want." Anthropic's Q1 numbers are headline-grabbing, but the more important data is the migration from personal agents to shared coworkers (Every's Plus One 2.0), the rise of Hermes on stable open-source rails, and the pivot to AI second brains as personal infrastructure. The agent economy is being rebuilt around durability, not novelty. That is a much better foundation than what we had a year ago, and the market is still a quarter or two behind it.
American politics is now operating on a two-track speed: the conventional one (inflation, war, primaries) and a much faster constitutional one (impeachment expungement, redistricting math, mid-cycle purges of incumbents who voted against the leader). The Cassidy result is not an outlier, it is a velocity reading. The Republicans who treat the next twelve days as a normal political cycle are misreading what is happening to their party in real time.
If you only read three pieces, I'd suggest: Every's "After the Personal Agent" (the agent thesis is being rewritten), Bruce Mehlman's Six-Chart Sunday (the cleanest single read on the midterm-China-Iran convergence), and Rich Turrin on the eight agentic payment standards (the most under-covered structural story of the quarter).