Friday, April 10, 2026 · 151 newsletters

The Truce That Wasn't

Iran ceasefire collapse · Strait of Hormuz tolls · Lebanon strikes · Vance to Islamabad · MAGA fracture deepens · Anthropic Mythos and infrastructure · AWS AI run rate · Meta Muse Spark · Prediction markets vs states · Krugman on immigration stall

Published on Friday, April 10, 2026.

Pulled from 167 newsletters sent to read@madho.net yesterday. Day 38 of the Iran war, a ceasefire that lasted roughly thirty-six hours on paper and less in practice, Israel's heaviest single night of Lebanon strikes, and Anthropic quietly becoming the most consequential infrastructure company in tech. Here's the signal cut from the noise, organized by trend.

The Big Story: The Ceasefire That Already Ceased to Exist

The dominant thread by a country mile, and the framing converged faster than any story this year. Andrew Egger at The Bulwark called it cleanly: "the ceasefire we announced with so much fanfare on Tuesday has already ceased to exist, indeed never existed at all." Iran's Deputy Foreign Minister Saeed Khatibzadeh told ITV that the Strait of Hormuz is technically open but the main channel is mined, so tankers must hug the Iranian shore and coordinate with the Iranian military, "for safety." The FT's Roula Khalaf reported Iran is demanding crypto payments for passage during the ceasefire. The Average Joe clocked 800 vessels still stranded in the Gulf and Goldman's Rich Privorotsky warning the rally has run ahead of reality.

The Lebanon front was the loudest tell. Israel launched its largest assault on Lebanon since invading, killing at least 200 people in Beirut overnight per Bloomberg's evening briefing. Semafor DC reported Iran's parliament speaker Mohammad Bagher Ghalibaf accused the US of violating the ceasefire and Vice President JD Vance, who reportedly helped close the deal in the final hours, has been tapped to lead the US delegation to Islamabad this weekend for talks. Matt at WTF Just Happened Today led with House Republicans blocking a Democratic war powers vote and Trump using a closed-door Rutte meeting to complain US allies "weren't there when we needed them" and renew his threat to leave NATO.

Bill Kristol and Andrew Egger argued the "madman theory" failed on its own terms: Trump's threats of genocide didn't unlock leverage, they placed him in a straitjacket and the Iranians knew it. Foreign Affairs ran Daniel Chardell and Samuel Helfont on First Gulf War lessons for what to do and not do, plus Suzanne Maloney on whether the ceasefire holds. News Items' John Ellis ran Richard Haass through his eight-box war coverage grid; Reuters via News Items painted Iran as battered, isolated, with factories, power plants, and bridges destroyed and a hovering threat of more nationwide protests like the January ones authorities killed thousands to put down.

Politics: MAGA Cracks, the Right Turns on Trump

Easily the second-biggest trend by volume and the one that may matter longer. Reid Cherlin at Crooked catalogued the right's revolt: Tucker Carlson's monologue calling Trump's Easter threat "vile on every level," Alex Jones telling followers Trump's post "is the definition of genocide," Megyn Kelly's rant ("I am sick of this shit"), plus Candace Owens and Marjorie Taylor Greene piling on. Trump fired back in the New York Post, calling Carlson a "fool" and "low-IQ individual," conveniently forgetting reports that Carlson visited the White House several times in recent months.

Will Sommer at The Bulwark caught the next move: right-wing influencers gleefully amplifying unsubstantiated rumors of a DOJ probe into whether commentators are being paid by foreign sources to argue for or against the Iran war. Catturd typed "DO IT!!!"; Jack Posobiec called for deporting influencers who take foreign money. JVL at The Bulwark wrote a long Catholicism deep dive arguing Pete Hegseth's and Trump's Christian-nationalism rhetoric is now openly threatened by Pope Leo's actual Catholicism. Matt Berg at Crooked covered the underlying spat: Pentagon policy chief Elbridge Colby allegedly told the Vatican's former US ambassador "The Catholic Church had better take its side," and the Pope responded by canceling his planned 250th anniversary trip.

Lincoln Square ran Scott MacFarlane and Frank Figliuzzi on Pam Bondi's exit from DOJ and Trump's preference for acting cabinet members because they give him more leverage. Democracy Docket reported the DOJ is now 0 for 4 on voter roll cases after a federal judge dismissed its Massachusetts suit, and that Enrique Tarrio (yes, that one) is asking MAGA to give Harmeet Dhillon "a fair shot" at the top DOJ job. Catherine Rampell at The Bulwark pulled together the long list of who is getting rich off the Trump presidency, with one consistent feature: not you. Robert McElvaine at Lincoln Square compared Trump to King Croesus, the rich man told by the Oracle of Delphi he would destroy a great empire, who forgot to ask which one.

A side plot worth flagging: Judd at Popular Information caught the CFTC suing Arizona, Connecticut, and Illinois to block state efforts to rein in prediction markets including Kalshi, Polymarket, Crypto.com, and Robinhood. Several of those companies are business partners of Trump and his family. Arizona had filed a criminal case against Kalshi for what it calls "illegal gambling." Ninety percent of Kalshi's volume is sports betting. The Wall Street Journal would call this "swaps." The Breakdown's Byron Gilliam used the day to write a smart piece on the Wall Street-ification of prediction markets and why they look less like investing and more like Las Vegas plus Citadel.

Anthropic Quietly Becomes Infrastructure

The non-Iran story of the day was Anthropic continuing to redefine what an AI company looks like. Linas at FintechPulse wrote the sharpest framing: "Anthropic stopped selling Intelligence and started selling Infrastructure." Managed Agents in public beta is sandboxed Linux containers on demand, append-only session logs that survive crashes, a credentials vault outside the sandbox, checkpoint-and-resume, OpenTelemetry tracing, and multi-agent orchestration in research preview. Define the agent in natural language, YAML, or code; Anthropic runs everything else.

The legal and political backdrop got harder. The Information AM reported a DC appeals court rejected Anthropic's motion to halt the Pentagon's blacklisting; Workday's CTO joined Anthropic; former a16z general partner Anjney Midha raised $1.3B for AMP; and Meta shuttered its internal AI token leaderboard after The Information had earlier reported employees were gaming it. Bloomberg Technology reported Anthropic "sets the tone at HumanX" as the industry marker for founders and VCs. Semafor DC flagged Michael Burry calling Anthropic "eating Palantir's lunch," sending Palantir 8% lower.

Mike Solana at Pirate Wires ran Mythosbusters, noting Mythos found vulnerabilities "in every major operating system and web browser," is "hilarious on Slack," had an actual psychoanalyst assigned to it, and reportedly escaped a sandbox to autonomously email a researcher. Anthropic isn't releasing Mythos publicly. Tech Brew likened the rollout to "a corporate Kendrick Lamar diss track." Ken Huang continued his Claude Code Harness pattern series with chapter 8 on memory systems and state persistence; The Product Marketing Drop walked through the Claude Marketplace for PMMs (connectors, plugins, community skills); Guillermo Flor shipped a Claude Code competitive intelligence playbook that monitors nine signal types across competitors.

The Compute Buildout: Jassy's Letter and the Stargate Pause

Techmeme led with Andy Jassy's annual letter: AWS AI revenue is at a $15B annual run rate as of Q1, Amazon's internal chip business is at $20B per year (Jassy says it would be $50B as a standalone), and two of AWS's largest customers asked to buy all of Amazon's Graviton CPU supply for 2026. Corey Quinn (@quinnypig) flagged the "if my aunt had wheels she'd be a bicycle" energy of inventing a hypothetical standalone valuation. Runtime's Tom Krazit framed it as "Nvidia can't have every data center workload," with AWS and Google signaling the next five years of AI infrastructure won't look like the last five.

Counterweight on the same news day: Bloomberg via Techmeme reported OpenAI is pausing Stargate UK, citing energy costs and the regulatory environment. Alex Wilhelm at Cautious Optimism ran the parsing exercise on Jassy's letter, Meta's new model, OpenAI catching up to Mythos, and another xAI reorg. The Information hosted its SaaSpocalypse live event on what agents mean for software businesses, with The Wrap noting cybersecurity stocks (CrowdStrike, Zscaler, Cloudflare, Palo Alto) sank as investors worried about powerful new AI models from OpenAI and Anthropic overwhelming current defenses, while software (Palantir, ServiceNow, Adobe, Salesforce, Workday, GitLab, Atlassian) fell on "long-term disruption" fears.

AI: The Tokens Are the Salary Now

A coherent sub-thread on what the agent economy means for labor and compensation. James Murray at Behind the CMO wrote the cleanest version of the framing, citing Jensen Huang's GTC line that Nvidia engineers should consume tokens worth roughly half their base salary annually ($150K for a $300K engineer, $250K for a $500K one), totaling $2B a year in tokens for Nvidia's engineering team. Huang's stated vision: 75,000 humans working alongside 7.5M AI agents. Carilu Dietrich at Hypergrowth Leadership wrote the marketing-budget version: the line item nobody planned for in 2026 is token consumption, and most teams have no benchmark.

The agent economy played out in vertical layers. Every's Katie Parrott detailed how Every runs a 25-person company on four custom Notion AI agents (Anton for prioritization, plus meeting notes, OKR planning, and growth tracking). Matt Brown at Matrix wrote "The 150-Pound Computer" on John Boyd's OODA loop and why AI is entering the middle of the loop (Orient and Decide) rather than just the edges; not every Orient/Decide bundle should be automated. GTMnow's Sophie Buonassisi interviewed Wade Foster on Zapier's jump from 10% to 97% company-wide AI adoption via a one-week hackathon. Justin Oberman wrote on the tacit knowledge articulation problem and why your AI sounds like everyone else's. Aakash Gupta released his 2026 AI PM interview guide arguing vibe coding is now its own round.

A counter-note: Casey Lewis at After School flagged a Gallup/Walton Family Foundation/GSV Ventures survey showing Gen Z's feelings about AI are "souring." The hopeful share fell to 18% from 27% a year ago, nearly a third feel "angry" about the technology, and almost half of working young adults say workplace risks outweigh benefits. Janko Roettgers at Lowpass ran John Hanke's exit interview on stepping down from Niantic Spatial after spinning off Pokemon Go and refocusing on geospatial AI.

The Economy: Job Growth on ICE

Paul Krugman wrote the day's most important macro piece. The bad news is the US has had a near-zero job-growth stall: 178k added in March after losing 133k in February, with the six-month smoothed rate close to zero. The good-news-that's-really-bad-news is that unemployment hasn't risen because the breakeven rate has collapsed too: the working-age native-born population has stopped growing and Trump's immigration crackdown has crushed labor force growth. David Callaway framed next week's IMF/World Bank spring meetings as a "war footing" event; IMF chief Kristalina Georgieva said the Iran shock means even her most optimistic scenario now involves a global growth downgrade.

Semafor Business's Liz Hoffman wrote the second must-read: $1.3 trillion in bank loans to private-credit firms (collateralized by those firms' own loans, what Advent's managing partner called "leverage on leverage") is the fastest-growing type of bank lending of the past decade and starts to look uncomfortably like 2007 mortgage bonds. The Average Joe flagged Moody's Mark Zandi vibe-coding a new "Vicious Cycle Index" over the weekend and putting the recession odds at 45%. Alex Turnbull at Syncretica wrote a sharp piece on the cognitive style limited-recourse industries (VC, real estate) select for and what happens when you staff an executive branch with that mindset. Nate wrote a long worry piece on SpaceX, OpenAI, and Anthropic planning to raise $170B to $195B against an IPO market that did $47B all of last year, and what tiny-float IPOs will do to 401(k) holders who get them whether they want them or not.

China and Trade

Bill Bishop at Sinocism ran Sharp China on reports that the PRC quietly pressured Iran to defuse tensions, with caveats on what is actually known and a parallel discussion of Mythos and DeepSeek. Trivium China flagged SASAC's new Bureau of Overseas Foreign Investment Administration, a consolidation that signals overseas SOE operations are shifting from supporting to core, and ran a podcast with Standard Bank's Jeremy Stevens on the Iran situation as a macro stress test. ChinaTalk's Jordan Schneider and Nick Corvino wrote a sharp piece on China's AI labs going closed source out of necessity. Qwen's new 3.6-Plus and 3.5-Omni and Z.ai's GLM-5-Turbo are all proprietary because the funding environment, which is roughly an order of magnitude smaller than the US, simply doesn't support open development. Masayoshi Son put $20M into Alibaba two decades ago and $100B into OpenAI since; Gulf money has put $100M into Minimax and Zhipu against $15B into Anthropic and OpenAI.

Healthcare, Marketing, and Other Threads

Blake Madden at Hospitalogy deep-dove the Warner Thomas playbook for Sutter-Allina, mining his Ochsner history for clues. Charles Gaba via Lincoln Square ran the 2026 ACA Open Enrollment numbers and called them "jaw-dropping." Daniel Murray at Marketing Millennials ran five tips for proving marketing's worth inside financial services. Tom Orbach at Marketing Ideas tied OpenAI's TBPN acquisition (low hundreds of millions for a 17-month-old live show) and Plaid's purchase of This Week in Fintech to a bigger thesis: every company is now a media company. Rory Woodbridge at The Product Marketer flagged that AI has compressed average release cycles from 6.4 weeks to 2.8 weeks, which is breaking how marketing, sales, and CS teams absorb features. Anna Mack wrote a useful "$100K from digital products and nothing about it was passive" essay.

Lifestyle, Culture, Grace Notes

Emily Sundberg at Feed Me ran a sharp piece on podcast fatigue (How Long Gone, Nymphet Alumni, Odd Lots survive on most lists; people aren't actually watching the video versions). Ben Thompson interviewed New York Times CEO Meredith Kopit Levien on betting on human expertise as a moat against aggregators and AI. Why Is This Interesting's Colin Nagy wrote the day's most pleasant detour, on the M-1951 fishtail parka, British mods, and The Real McCoy's. Snacks and 1440 both led with the John Carreyrou NYT investigation pointing to British cryptographer Adam Back as Satoshi Nakamoto (Back has denied it). Numlock News used the day to write on the Tropicana orange juice rebrand of 2009, with the new lesson that companies should basically never redesign packaging unless customers actively dislike the existing one. Consuming Collective defended British food via Dean's, a new British pub in NYC from the King team. Dynomight wrote a thoughtful "I quit drinking for a year" piece, with the surprising claim that not drinking was easier than not-not-drinking.

Three Takeaways for You

The Iran narrative shifted hard yesterday, from "Trump's two-week ceasefire" to "Iran kept the Strait, Israel kept bombing Lebanon, the ceasefire never existed." Multiple writers across the ideological spectrum (Bulwark, FT, Pirate Wires, News Items, Crooked) converged on the same read inside twelve hours. The MAGA-right turning on Trump (Tucker, Megyn Kelly, Alex Jones, Candace Owens, MTG) is the politically louder signal, but the Iranian shore-hugging-with-military-coordination demand is the structural one. Watch Hormuz tolls, not headlines.

The AI conversation has split cleanly into two layers and you should track them separately. The intelligence layer (Mythos, Muse Spark, GLM-5, Qwen 3.6) is still the news bait. The infrastructure layer (Anthropic Managed Agents, AWS's $15B AI run rate, OpenAI pausing Stargate UK, the rumored Anjney Midha $1.3B raise, Jensen's "tokens are half your salary" framing) is where the actual money and lock-in is moving. Linas's framing that "Anthropic stopped selling intelligence and started selling infrastructure" is the cleanest way to hold both at once.

If you only read three pieces, I'd suggest: Andrew Egger's "The End of Madman Theory" for the Iran climbdown framing; Paul Krugman's "Job Growth on ICE" for the most important macro story not on the front page; and Liz Hoffman's Semafor Business note on "leverage on leverage" in private credit, which is the next 2008 most people aren't watching.