Souk's Notes - Unofficial Summaries

2026-07-26 · open the original ↗

• Nicaragua's Ortega, in power since 2007, will permanently suspend elections. The speaker is skeptical Nicaragua is strategically important enough to warrant US intervention on the scale of Cuba, Venezuela or Iran.
• He flags Rubio calling the suspension a "matter of national security." Nicaragua's relevance, in his view, comes mainly from China's infrastructure investment there and reported interest in an alternative to the Panama Canal, plus its role as a source of migratory labor to the US.
• He reads Ortega's path from socialist leader to dictator as vindication. He contrasts this with the US, which he says retains rule-of-law checks on its own president despite being accused of authoritarian drift itself.

2026-07-26 · open the original ↗

• Nvidia's founding 3D-graphics algorithm was "exactly wrong," and the company relearned the field from library textbooks after 35-40 rivals already had a head start. Realizing the error in 1995, Huang bought OpenGL textbooks at Fry's Electronics and handed them to his engineers; his lesson is that the specific technology a company starts with matters less than its willingness to confront that it was wrong and relearn.
• A moment of radical honesty with a client kept Nvidia alive during its near-failure. Unable to deliver the graphics chip Sega had contracted for its post-Saturn console (which became the Dreamcast) under a $12 million deal, Huang told Sega's CEO the technology didn't work but still asked to keep $5 million of it; Sega paid, betting on the team rather than the product, and Nvidia went public in 1999 at a $300 million valuation, now worth more than a trillion dollars.
• AI automates tasks, not jobs, Huang argues, citing data across three fields. Software engineering jobs are up roughly 10% year-over-year despite coding being automated, radiology jobs are up about 20% despite scan-reading being automated, and paralegal roles are growing despite legal AI tools like Harvey -- in each case he attributes the growth to a backlog of unmet demand that automation lets the field finally work through.
• Nvidia's "physical AI" business -- robotics and self-driving -- is already roughly $10 billion and, Huang predicts, will become the company's next $100 billion business within less than 10 years. Self-driving was chosen as the first major application because it combines a large enough market with standardizable technology; Nvidia supplies chips to Waymo, Tesla and Mercedes, and open-sourced its Alpamayo self-driving stack for smaller adjacent markets like agriculture and warehouse robots.

2026-07-26 · open the original ↗

• A tanker leaving Iran's approved Hormuz route hit a mine. He reports that on Sunday, July 26, a tanker exploded in the Strait of Hormuz after striking a mine. Iran had agreed to let it pass on the approved Iranian route, but it apparently veered off course.
• A fragile truce is already fighting over its own wording. He reports that the Houthis and the Saudis are still fighting, and that Iran has agreed to stop retaliatory strikes as long as the US keeps not striking it. Both sides are already claiming the other caved, and he expects this cycle to repeat "a half a dozen times" before it ends.

2026-07-26 · open the original ↗

• API providers should patch breaking changes automatically, not just announce them. Drawing on experience with over 50 API vendors and noting that external API and package changes caused over 30% of service downtime at AWS, the author argues the fix is an agent that scans affected codebases and opens a pull request when, for example, Stripe ships a breaking change.
• Agentic coding tools have made this newly feasible. The author points to Claude Code, Devin, and Replit as proof that developers now routinely grant external tools codebase access that would have been unthinkable two years ago, and suggests this could work as per-provider update agents or as a neutral third-party service tracking changes across vendors, "like Dependabot, but for APIs."

2026-07-26 · open the original ↗

• Equating opposition to the president with hating America is authoritarian. Responding to an opening clip whose captions are unintelligible, he argues the founders fought a revolution against a king precisely so loyalty to a ruler would not equal loyalty to the land, and made the right to criticize the powerful the First Amendment.
• The claim collapses when applied evenly or to another name. If opposing the sitting president meant not loving America, opponents of Biden, Obama, Bush, Reagan and Lincoln would all be traitors, branding about 150 million people enemies on a rotating four-year schedule. Swapping in any senator or governor, he says, makes the sentence sound absurd.
• Supporting Trump is not un-American either. He concedes millions back Trump out of their own reading of the national interest, and argues patriotism belongs to no party, so no politician gets to gatekeep it. Real love of country, in his view, means holding it to its own promises rather than falling in line.

2026-07-26 · open the original ↗

• Red Sea and Strait of Hormuz disruptions pushed Brent above $100 a barrel, complicating central banks' "last mile" inflation fight just as new US tariffs added trade friction. Brent eased to close the week at $97, but rising energy, insurance and freight costs threaten fresh inflationary pressure, while new 10-12.5% US tariffs — with extra levies on Brazil and Canada — replace those phased out after a Supreme Court ruling, raising USMCA and retaliation questions.
• Heavy tech capex-driven debt issuance pushed borrowing costs to multi-decade extremes across currencies and bonds. The US 10-year closed at 4.68%, Germany's at 3.15% and the UK's at 5.03% (each briefly higher intraweek), the yen touched a 40-year low of 164 per dollar, and Japanese government bond yields hit 31-year highs, with the pressure on equities hitting the Nasdaq hardest.
• The week ahead centers on the Fed's July 28-29 meeting, where markets price roughly a 40% hike probability, alongside BOJ and BOE decisions and mega-cap tech earnings. He expects the Fed to hold while closely monitoring inflation; consensus looks for PCE headline inflation to ease to 3.7% YoY and core to 3.1%, while Amazon, Apple, Meta and Microsoft face scrutiny over returns on their AI capital spending.

2026-07-25 · open the original ↗

• The US is asking to end a war and being refused. The US has halted strikes on Iran and reportedly put three offers through mediators, all rejected. He lists the objectives as unmet: Iran fires more advanced missiles than at the start, there is no deal, the strait opens and closes like a broken valve, and the regime is confident enough to say no.
• Ukraine's Caspian strike exposed a Russia–Iran arms pipeline. Ukraine struck a Russian warship and cargo vessels carrying Iran-linked military hardware on a route Kyiv says moves weapons between the two countries. He notes this came a day after Trump said it was not happening and that he trusted Putin.
• Russian resupply explains why Iran's arsenal improved under bombardment. His reading is that every degraded stockpile was topped back up, so the enemy's ammunition was never finite. He calls it Moscow's cheapest, most effective military aid in modern history, pinning the US in the Middle East for five months, and says past presidents avoided war with Iran for this reason.
• The war turns on endurance and supply, where the US is outmatched. He argues the US faced a patient coalition, Iran fighting and Russia arming, rather than one isolated adversary. He predicts that unless the US gets out, the Ukraine and Iran wars merge, and calls the position near checkmate.

2026-07-25 · open the original ↗

• Canada's profile inverts the normal logic of alliance: it is a supplier seeking protection, not a protector seeking supply. A food, energy and critical-minerals superpower — potash, uranium, nickel — with roughly 41 million people spread across the world's second-largest territory and no independent ability to project force. The paper reduces it to one line: Canada has everything a bloc needs except the ability to defend it. The only currency it holds in abundance with which to buy security is resources.
• Every single-partner substitute for the United States fails the security test. China is the threat the security factor exists to guard against; the Gulf states sit behind a Strait of Hormuz that closed in early 2026, making any alliance routed through it self-defeating; a Europe-only bloc shares Canada's values but is an ocean away, still militarily dependent on the US itself, and cannot cover Pacific or Arctic approaches; and 41 million people cannot self-fund blue-water and continental defence simultaneously. The conclusion is that no patron is on offer, so the alternative must be built rather than swapped into.
• What survives contact with all five factors is a four-ring lattice, paid for in resource surplus. Inner maritime-industrial (Japan, Korea, Australia) supplies shipyards and Pacific patrols; the values ring (EU and UK, with CETA as the existing spine) supplies economic depth and standards; Four Eyes preserves intelligence independent of the US question mark inside Five Eyes; and a Nordic-Arctic ring covers Canada's own approaches. The paper is unusually candid about its limits: even a perfect lattice cannot replace the US on continental defence given shared border, airspace and scale, so the realistic goal is survivable dependence rather than substitution — and the geometry is offered explicitly as a falsifiable lens, not a forecast.

2026-07-25 · open the original ↗

• Tariff policy is revealing which dependencies are real, one exemption at a time. Trump imposed tariffs on 60 nations with Canada topping the list at 50%, without specifying the goods. Two carve-outs followed almost immediately — aluminium, because Middle East supply shut down, and phosphate, because American farmers need it — while a 50% tariff on hockey sticks stands. His point is that the exemptions map the genuine chokepoints, and the rhetoric from Bessent and Navarro that Canada "owes" $3 trillion really means Canada never had to build a deep-water fleet.
• The security guarantee that made global trade cheap was a gift, and its withdrawal restores an older logic. Before eighty years of American sea-lane protection, every empire that shipped goods maintained a battleship fleet — that was the entire arrangement. His data point for the return: Ukrainian strikes on Russia's ghost fleet have pushed Russia into providing naval escorts. He is careful to frame this as a signal worth watching rather than a prediction that convoy warfare is back.
• He accepts the lattice framework but draws a blunter conclusion than the paper does. Canada can genuinely diversify counterparties — Japan and Korea for shipyards and patrols, Europe for market access, Four Eyes for intelligence, the Nordics for the Arctic — and he credits Carney's agreements as coherent rather than performative. But geography does not negotiate: shared border, shared airspace, and scale mean the dependence can be reduced, not escaped. His analogy is American rare earths, where $23T over 25 years still does not sever reliance on China, and his own forecast is that the US, Canada and Mexico eventually function as one entity.

2026-07-25 · open the original ↗

• The Houthis' refinery strike deepens a trap the US built itself. He reports it confirmed that the Houthis hit Saudi Aramco's 400,000 barrel-a-day Jizan refinery on the Red Sea coast, their first hit on a Saudi refinery in years, plus facilities at Yanbu, where 75% of East-West pipeline oil exits. They also announced a Red Sea blockade, with about 5 million barrels a day of refined fuels already off the market.
• Stopping direct attacks on Iran cannot end a franchised war. That night the US ran no strikes on Iran and, he says, spent it pressing to return to the MOU, despite Trump's pledge to hold Iran responsible for Houthi actions. With Saudi Arabia now bombing the Houthis in Yemen, he argues too many hands are on the throttle for the US to de-escalate alone.
• Reviving the June MOU would be surrender to Iran. He argues its deliberate ambiguity handed Iran control of Hormuz and pulled the US deeper in, and he doubts Marco Rubio, J.D. Vance or Trump ever read it. Washington wants it as a face-saving exit, but in his view it does nothing about the Houthis, Saudi Arabia's war, the blockade or $100 oil.
• Every option left carries a heavy cost, none a victory. Keep fighting and bleed against an enemy seeking exhaustion; revive the MOU and retreat while proxy war continues; or escalate and risk both chokepoints closing, oil off the charts and global recession. He calls it entirely avoidable, born of believing a $950 billion military could win quickly.
• Falling crude will not bring falling gasoline. Oil fell over the weekend on deal talk, but he argues each refinery knocked out removes fuel from the market, so gasoline prices may climb. He expects a possible oil glut when this ends that cannot be turned into gas.

2026-07-25 · open the original ↗

• Pure software startups are now "basically dead" as durable businesses, one YC partner argues, because writing software stopped being the hard part. Tom says he would not start something like Calendly or DocuSign today since the code is too easy to replicate; durable startups now need a genuinely hard bit instead -- a regulatory barrier (he says obtaining a banking license "almost killed me"), a hardware problem, or brutal B2B sales into a resistant industry.
• AI won't push startups toward solo-founder billion-dollar companies, Tom argues, because coordination costs, not software costs, set company size. He compares team costs to Metcalfe's law and predicts AI will instead compress companies toward Dunbar's number, roughly 150 people, rather than one; he cites funding a strong solo founder building self-improving agents who hit a "crisis of confidence" six weeks into the batch with no co-founder to steady him.
• The single biggest trait separating top-performing YC startups is launching early and often, partners say, since founders' instinct to keep researching or building is usually a mistake. They point to Cursor's founder publicly showing a string of near-zero-upvote Hacker News launches before the company's eventual large acquisition, and frame startups as an empirical cycle -- hypothesize, launch, gather real-world data, adjust -- rather than something solved by sitting and thinking harder.
• As software gets cheap, ambitious hard-tech founders still need proportionally more capital, not less, Tom says. He cites a YC-funded founder building small nuclear reactors who will need to raise roughly $800 million, arguing venture capital's role shifts toward funding startups tackling harder physical and regulatory problems rather than software that can now be built for a fraction of the previous cost.

2026-07-25 · open the original ↗

• Compliance costs grow faster than revenue as companies expand, YC argues. It points to financial compliance being stitched together from spreadsheets, siloed point solutions, and expensive specialized headcount, a burden it says compounds most sharply for businesses navigating state-by-state licensing, renewal cycles, and audits across a varying regulatory patchwork.
• Most compliance work is inherently AI-native, YC argues, since it is mainly monitoring, flagging, and reporting. Tasks like tracking regulatory changes, flagging anomalies, generating reports, and keeping audit trails are things YC says AI can already do "faster and cheaper" than humans, yet most existing tools are still built around manual review bottlenecks.
• YC says it wants to fund founders who rethink compliance operations around AI, not just automate the existing manual process. It predicts that founders who consolidate today's fragmented tools and reduce reliance on specialized headcount will become "essential infrastructure" for any business operating globally, and is soliciting applications from founders building in this space.

2026-07-25 · open the original ↗

• Verifying that someone online is actually human is an urgent problem. YC calls it one of the most important problems of the next decade, citing a case where a finance worker wired out $25 million after a video call where every other participant turned out to be a deepfake, and arguing that cheap, realistic voice and video clones have broken the trust signals the internet was built on.
• The opportunity goes beyond fraud prevention to bots, dating, and reviews. YC envisions social platforms with no bots in the replies, dating apps where every match is a real person, and reviews written by people who actually bought the product, saying the ideal solution verifies humanity without forcing anyone to give up privacy — and whoever builds it becomes the layer banks, apps, and video calls check before trusting anyone.

2026-07-25 · open the original ↗

• Amilabs is betting that large language models are fundamentally ungrounded, and is building "world models" trained on raw sensory data instead. Alexandre Lebrun, its CEO, argues an LLM is "like someone who has never left the room they were born in" but read every book, learning about the world only as a proxy through text; Amilabs, co-founded with Yann LeCun, instead trains on video, audio, and robot-collected touch data.
• Lebrun calls today's vision-language-action robotics models "a very bad hack," too slow and expensive for real deployment. He argues VLAs misapply LLM-style architecture to robotics, making them imprecise and unable to hit the latency a home robot needs, and expects world models to cost roughly as much to train but be far cheaper at inference once built.
• Amilabs raised roughly 1 billion euros in what Lebrun calls the largest seed round in European history, and says the real cost was not dilution but expectations. He warns that raising that much money and showing nothing publicly for two years would be very hard to survive, and the company deliberately built its research teams in Paris, New York, Montreal and Singapore rather than San Francisco, partly to select for hires making a genuine long-term commitment.
• Large companies structurally can't take the risks that let a startup like OpenAI pull ahead, Lebrun argues from his own time at Meta. He describes an internal Meta chatbot that couldn't ship after it answered a test question inappropriately, and says the core transformer research behind large language models existed as early as 2018 but only OpenAI, unconstrained by that same risk aversion, chose to scale it into ChatGPT.

2026-07-25 · open the original ↗

• Landau founded Encord on the "bitter lesson" that scale beats hand-engineered domain expertise, a pattern he saw across three careers. In particle physics and then ten years as a quant, he crafted domain-specific features before applying machine learning; he says AI's shift toward just adding more data and compute convinced him and his co-founder this was "the technological paradigm shift" of their generation, prompting him to quit trading during the extreme volatility of COVID-era markets.
• Encord's first two sales teams failed before its third one worked, and Landau kept one bad hire for nine months despite YC warning him against that exact profile beforehand. He says the company has since gotten faster at trusting its gut on people decisions, since there is rarely a single clear reason a hire isn't working, just a feeling that takes too long to act on.
• Encord's pivot from general multimodal data annotation into physical AI followed the market catching up to the product, not a deliberate repositioning. Landau cites the fact that 80% of economic activity involves moving or manipulating physical objects, and says robotics insiders expect more robots than people within five to ten years; Encord now processes more training data than was reportedly used for GPT-4, multiple petabytes across hundreds of customers in robotics, autonomous vehicles, logistics and manufacturing.
• Landau's advice to founders is to stop trying to smooth out the highs and lows of running a company and instead "ride the roller coaster." He says he only learned to enjoy the volatility relatively recently, after realizing he had more control over his own mental state than he assumed, framing each new crisis as something to run toward rather than dread.

2026-07-25 · open the original ↗

• Falling sensor costs now make dense physical-world data collection feasible. YC cites Gecko Robotics, which uses robots to collect data in hard-to-reach places to build predictive models, and Source Air, whose autonomous weather balloons feed atmospheric data the US government uses for forecasting — arguing this unlocks industries like energy, agriculture, logistics, and construction that still rely on limited data and intuition-based models.
• Modeling a physical system precisely enough eventually lets it be controlled. YC argues this is the payoff of denser real-world data, citing steering hurricanes, reversing desertification, and cooling the planet as the kind of outcomes becoming possible.

2026-07-25 · open the original ↗

• Measure-and-share analysis rarely drives a decision. The speaker says most naive analyses run two steps, measure and share. The lesson teaches six (decide, ask, hypothesize, measure, share, act) and later adds an explicit validate step to make seven. The running example is an e-commerce "save for later" button: savers buy more, so should the button be more prominent? The number may be correct and still not support a decision.
• Pin down the decision's stakes and the question's definitions first. Deciding includes the cost of the work the team won't do instead, and someone has to own the call. On "buy more", a room the speaker polled gave revenue per session, revenue per customer, orders per month, AOV per user, repeat orders, subscriptions and referrals. Five analysts, models or AI runs could return five technically correct answers. The session defines a buyer as a user with at least one order of any status, ever.
• Bare Claude Code returns a convincing correlation it cannot check. With one three-column file (user ID, saver flag, ever-purchased flag) and no skills or agents, it reported 68.6% of savers vs 52.6% of non-savers had ordered: a 16-point gap, about 1.3x or 30% more. It flagged on its own that this is a correlation, not an effect, but had no data to test that. The speaker argues the friction AI removed used to force the "is this worth asking?" step, so validation now has to be made explicit.
• Write the disproof and the funding threshold before looking. The hypothesis is that saving causes some buying; it would be disproven if the same people buy more for another reason, such as tenure. The bar was set at an expected 5-point lift, because a bar set afterwards gets talked into whatever turns up. In a gated "workbook" repo with a skill, recorded definitions and added columns (signup date, channel, country, device), the system flagged tenure (700 vs 616 days on average), refused post-saving variables, and stopped at gates for the user's confirmation. Within tenure buckets the gap fell to about 5–7 points, so the speaker would probably not fund the change. They note this is still not causal and that a randomized experiment would settle it.
• AI leads only the measuring; the human owns everything else. In the speaker's two-lane chart, measurement is the one AI-led stage. AI can draft the rest and run validation checks, but those are checks the analyst wrote in advance, and the analyst is accountable for the number just as with a spreadsheet or Snowflake. Their single takeaway is to ask "what decision is this for?" before typing anything.

2026-07-24 · open the original ↗

• Software for physical-world labor has not evolved in 20 years. Despite 80% of the global workforce not working at a desk, the essay says construction, maintenance and fleet-operations software still just dispatches, tracks, and bills — nothing built for the new mix of workers now arriving.
• Three worker types — agents, robots and humans — now share the same job site. It argues today's operating systems can't route work among an AI agent that quotes and schedules jobs, robots newly deployed in the field, and human workers using wearables to record their work, nor handle safety when robots and people work side by side.
• The opportunity dwarfs existing software spend, YC argues. These industries spend 10 to 100 times more on labor than on software, incumbents only sell dashboards for human coordination and visibility, and a startup capturing end-to-end data on how the work actually happens would have information no frontier model, robotics startup, or software incumbent currently holds.

2026-07-24 · open the original ↗

• Datadog got rejected by Y Combinator twice before its own success. Olivier Pomel recounts applying, getting an interview, then a rejection email from Paul Graham — followed by an invitation to a YC dinner that was rescinded once the rejection was remembered.
• Early fundraising failed because investors didn't understand the category. Pomel says New York VCs didn't grasp infrastructure companies while Bay Area VCs balked at backing one outside the Bay Area; the angel round only came together after Datadog repositioned from a vague "DevSecOps collaboration platform" to the more legible category "infrastructure monitoring."
• A December surge in AI coding convinced Datadog to restructure around agents. Pomel says his co-founder told the engineering team the company would stop writing code within two quarters, after developers were seen rebuilding entire systems solo in days that once took a six-person team six months.
• Move faster on hiring and firing, even before finding the "perfect" candidate. Pomel's advice, from scaling Datadog to roughly 8,000 employees, is to hire a promising candidate immediately and fire quickly if it doesn't work, since waiting for certainty wastes cycles a faster hire would resolve.

2026-07-24 · open the original ↗

• The administration admitted cutting blue-state energy funding over votes. He reports it conceded in court that it cancelled loans and grants to blue states because they did not vote for Trump. He focuses on $6.2 billion specifically designed to lower energy bills, much of it in the PJM network.
• The largest cancelled loan was transmission, not renewables. A $4.9 billion loan would have carried existing cheap Midwest power over high-voltage lines into PJM's most congested areas, cutting delivery costs he calls the network's main pain. He stresses no new generation was involved, so an anti-renewables rationale does not apply.
• Trump chose rising bills he can blame on Democrats. Trump promised to halve everyone's energy bills; he concedes these projects would not have halved them but would have lowered them and stopped the climb. His reading is that Trump preferred cancelling them and blaming Democrats to taking credit, out of indifference, retribution and political expedience.

2026-07-24 · open the original ↗

• Whether America is a "Christian nation" changes nothing legally. Whatever label is used, he argues, the laws and the Constitution stay the same: Christians, Jews and Muslims keep the right to worship and atheists keep the right to sleep in. Freedom of religion means the government stays out of it rather than picking a winner. A label that does not lower costs or fix roads, he says, is pointless.

2026-07-24 · open the original ↗

• No test yet separates real AI help from techno-solutionist theater. Asked for one, Okolo says the evidence base on long-term AI deployments in rural, low-resource areas is growing but not strong enough. From her ICTD research she says the older pattern was a Raspberry Pi or mobile-app pilot run for the study's length, weeks to a year, then abandoned by the community for lack of technology transfer, resources and infrastructure.

2026-07-24 · open the original ↗

• Iran has struck US bases in Bahrain, Kuwait and Jordan. Two US soldiers have died and significant equipment has been destroyed; the speaker contrasts this with the US Navy's aircraft carriers and strike groups, which he says remain largely unscathed because they're moving targets, unlike fixed bases.
• He argues the US military treats base losses as acceptable collateral. With a ceasefire in place since April 2026, he says the military had four months to better defend these bases — the fact they're still being hit implies the losses are tolerated, not unavoidable.
• By his read, the US is escalating while Iran is not. He says the scale of US strikes on Iran has been increasing even as Iran's own strikes on Persian Gulf countries have decreased compared to earlier in the year.

2026-07-24 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-24 · open the original ↗

• The setup: 64 sugar cubes fill a 4×4×4 block, each colored. Every cube is white, blue, or brown, and the arrangement of colors across the 64 positions can be anything at all.
• The task: prove six same-colored pairs always exist, no matter the coloring. That means finding 12 individual cubes total — grouped into six pairs, each pair a single matching color — regardless of how the 64 cubes happen to be colored.
• A further constraint: the six pairs' distances between centers must all match. For each pair, measure the distance between the centers of its two cubes; the puzzle requires that all six of these distances come out equal to each other.
• The puzzle comes from the 2024 British Math Olympiad. It's presented as part of Grant's recurring "puzzle of the month" series, posed without a solution in this short.

2026-07-24 · open the original ↗

• Iran's flat ceasefire rejection signals it believes it is winning. Iran refused a US offer carried into Tehran by Iraq's prime minister, demanding a permanent end to the war rather than temporary truces, then paraded ballistic missiles through its capital. He argues no one rejects a ceasefire after nearly half a year of bombing from desperation.
• Iran's missiles are now beating US base defenses. Citing an assessment from war-studies analysts rather than his own read, Iran has adapted with high-speed maneuverable missiles built to defeat the interceptors protecting American bases. He asks whether the program evolved under bombing or was resupplied, and calls either the opposite of victory.
• Unconfirmed reports suggest outside help is widening the war. He believes reports that Iranian troops have deployed to Yemen to support the Houthis; he is unsure of reports of Chinese and Russian missile help. He also notes Iran's new letter of intent and MOU with Iraq.
• The promised quick war has no plan and cannot be won. A one-to-two-week campaign is heading into its sixth month, with every response more strikes, blockades and American deaths. Iran's theory of victory is endurance, he argues, citing eight years against Iraq with half a million dead and four decades of sanctions.
• The US should get out, having refused ground troops. He argues fighting a country four times larger than Iraq with air power alone was never viable, while costing service members' lives, rising gasoline and diesel prices, and missiles not deterring anyone in the Pacific. He is explicit that he is not endorsing ground troops.

2026-07-24 · open the original ↗

• Anthropic's attempt to block Open Code backfired into a growth spike. Jay V says Open Code started 2026 at roughly 650,000 monthly active users; when Anthropic began rejecting Claude Code subscription requests that mentioned "open code" in the system prompt in January, the crackdown put the two products "on the same pedestal" and drove new users to check out Open Code, which reached about 13 million monthly actives by the end of June.
• A free tier is Open Code's CAC, paid in tokens. V frames the AI-era customer-acquisition cost as tokens instead of ads: Open Code eats the cost of a free tier to give users the "aha moment," then converts them to its $10 open-source-model subscription (about 160,000 subscribers, roughly $18 million annualized) and a pay-per-token inference business that reached about $38-40 million annualized within eight months of launch, with volume discounts on tokens becoming the company's margin.
• Usage data contradicts the DeepSeek-vs-GLM hype cycle on Twitter. Open Code's published token-volume data shows DeepSeek Flash and DeepSeek Pro leading usage even as GLM 5.2 draws more online buzz; V says users, unlike Silicon Valley's "token-maxing" culture, actively switch to cheaper models like DeepSeek Flash once they approach usage limits, in usage spanning 180 countries with China at 17% of traffic.
• The name "Open Code" was a deliberate bet on displacing closed leaders. Drawing on a prior project called OpenNext, V says that when a market has one or two dominant closed products, it coalesces around a credible open alternative, so Open Code claimed support for 70-plus models and providers at launch and built a companion open-source project, models.dev, to back that claim with real data.

2026-07-24 · open the original ↗

• Loaded tanker traffic through Hormuz is far below normal. As of 24 July 2026, the host says very few loaded tankers are transiting the Strait of Hormuz — not zero, but very few — a problem since the US-Iran war began.
• A Houthi embargo threatens Saudi Arabia's Red Sea workaround. Saudi Arabia had shifted crude via the East-West pipeline to Yanbu on the Red Sea to avoid the Gulf; the Houthis, siding with Iran, now say they will embargo Saudi-linked ships through the Bab el Mandeb Strait.
• Avoiding Bab el Mandeb makes Asia-bound voyages far longer. A tanker from Yanbu would have to go north through the Suez Canal and the Mediterranean and then around Africa, turning a two- to three-week trip into about a month and a half, per the host.
• A drone attack on the CPC pipeline adds a third bottleneck. The host reports the CPC pipeline at Novorossiysk on Russia's Black Sea coast was recently hit by drones in the Russia-Ukraine war, halting shipments he puts at 1 to 1.5 million barrels of crude per day; he presents the three bottlenecks together as his explanation for the current oil price.

2026-07-24 · open the original ↗

• Kimi K3's 2.8 trillion parameters are mostly idle per token. The host reports Moonshot AI's model is a mixture of experts activating 16 of 896 experts per token, with a 1 million token context window, native vision and always-on reasoning; Moonshot attributes roughly a 2.5x scaling-efficiency gain over K2 to its Kimi Delta attention and attention residuals. He relays Bank of America analysts' view that it shows scale plus architecture still delivers step-change gains despite China's compute constraints, which he reads as US export controls proving "a leaky dam".
• K3 is a top-tier model but not the top one. Moonshot itself says it trails Anthropic's Claude Fable 5 and OpenAI's flagship overall while beating Claude Opus 4.8 and GPT-5.5 on coding and agentic benchmarks; the host cites a score of 57 on the Artificial Analysis index against a median of 31 for its price tier and Arena's top ranking for front-end coding. Against that, Opus 4.8 still leads independently verified coding benchmarks, Moonshot's agentic scores are unreproduced, and Ethan Mollick calls it strong but uneven.
• Its pricing undercuts the US labs but not Chinese rivals. The host gives K3 at $3 per million uncached input tokens, $15 output and 30 cents for cached input, against $5 input for Opus 4.8 and $5/$30 for OpenAI's flagship; Z.ai's GLM 5.2 costs under a third as much per output token and DeepSeek V4 is 94% cheaper. Because reasoning is locked at maximum and billed as output, hidden thinking tokens can cost more than the visible answer on simple tasks.
• It is not yet open source until the weights ship. The API went live in mid-July, with full weights promised under a modified MIT licence by July 27; if delivered, the host says enterprises could run frontier-class AI on their own infrastructure with no per-token fees.
• Chinese models are pushing frontier AI toward a cheap commodity. The host reports OpenAI and Anthropic loosening usage limits in response, Cursor, DoorDash and Thinking Machines already using earlier Kimi versions, and Anthropic's February accusation that DeepSeek, Moonshot and MiniMax distilled Claude, which Beijing calls groundless. His own take is that every price war subsidises the applications practitioners build.

2026-07-24 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-24 · open the original ↗

• YC replaces vague ambition with concrete, recalibrated targets. The PhotoRoom co-founders say arriving at YC turned a fuzzy "billion-dollar company" dream into something real once they were surrounded by founders already operating at that scale — citing a retreat dinner where two founders compared how many satellites their 20-30-person startups had launched — and argue the recalibration keeps happening through ongoing YC office hours and demo-day comparisons, not just once.
• A hiring anecdote captures why European founders undersell themselves. The moderator describes making equal offers to equally-skilled engineers in Paris and the US: the French candidate turned the job down, saying the team was too strong and he'd fail, while the American candidate accepted and demanded two compensation levels higher than offered.
• Ambition can be trained with specific thought exercises. The panel recommends listing your current level, then the level reachable with "a magic wand," then 10x that, then a "king of the universe" ceiling; a related YC office-hours exercise asks founders to explain how their business could reach $100 million in revenue, and another asks what they'd still do to hit a goal with half their current resources.
• PhotoRoom grew by narrowing focus, then hit an "impossible" target. The co-founders say each 10x growth phase came from cutting scope — first from video-and-photo down to just photo, then down to e-commerce photo — and that an explicit $20 million ARR goal, which the team initially called unrealistic from a few million, was hit almost exactly a year later.

2026-07-24 · open the original ↗

• YC expects every startup it funds to eventually run on crypto rails. The essay argues this will happen "from capital raising to payments," often without founders realizing it, citing stablecoin adoption by major financial institutions, new regulatory clarity, and tokenized stocks as evidence the shift is already underway.
• Bear markets make better builders than bull markets, YC argues. Because prices are "decoupled from reality" when crypto is hot, a downturn removes the competitive pressure from scams promising "infinite yield" and instead attracts founders focused on building and reaching an early liquidity event.
• YC names Stripe, Coinbase and Axiom among its crypto bets. It also cites Beliani Pay and Infinia, which build developer interfaces and money-movement rails for Latin America, and Sporeum, which moves money to India via stablecoins, as examples of the infrastructure it's funding.

2026-07-23 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-23 · open the original ↗

• The administration neither foresees nor cares about the coming cost shock. He says the situation is without precedent, so no one knows what follows, and believes the administration lacks a clue and does not care because those hit hardest rate lowest with it. They have already lost safety-net, SNAP and health benefits and now face rising fuel, diesel, housing, electricity and food costs.
• His warnings cannot be fearmongering to people already afraid. He answers critics on comfortable incomes, around $150,000 a year, that he is speaking to people who cannot absorb an extra $500 a month from a week of heat.
• Officials seek war funding while cutting aid and enriching themselves. He says the administration is asking Congress for $1.5 trillion to continue a war he calls the dumbest in history, launched under false pretenses, while cutting children's cancer research and food aid. His deepest worry is being right, given how it has handled big problems.

2026-07-23 · open the original ↗

• AI finally makes products for the aging population possible. YC notes that by 2030 one in five Americans will be over 65, the US faces millions of unfilled caregiver jobs within a decade, and 53 million family members already provide unpaid care, while even voice assistants like Alexa and Google Home are frustrating for most seniors to use.
• New AI product categories span voice interfaces, safety monitoring, and robotics. YC lists conversational voice interfaces, safety monitoring for independence, physical-assistance robotics, and care-coordination software for family caregivers as newly possible, calling this one of the largest and most underserved markets in the world.

2026-07-23 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-23 · open the original ↗

• Selling volatility, not direction, is the trade in SpaceX despite disliking the stock. One host argues that 100-110% implied volatility on a trillion-dollar-plus company isn't sustainable and favors selling puts, calls, strangles or call ratio spreads; he frames being long the stock while trading its volatility as two separate decisions.

2026-07-23 · open the original ↗

• Iran's campaign is now about controlling oil, not military retaliation. Oil jumped nearly $7 in a day to over $101 a barrel. He reads the shift as a campaign to control global oil prices and flows and punish countries worldwide for the attack on Iran, a view he says he held while many analysts did not.
• Selective passage for Chinese tankers turns blockade into chosen access. Reports, which he believes, say two Chinese tankers transited the Bab el-Mandeb while most shipping avoids it, echoing early Hormuz. He warns China's resumed imports end the reduced buying that eased prices around June, and could shift the geopolitical balance.
• Iran's refusal of "fake ceasefires" means no recovery pauses. Iranian officials say they will no longer accept truces that let the US and allies stabilize flows and restock before fighting resumes. He reads this as continuous pressure designed to maximize economic damage.
• Uncertainty creates a shortage before supply physically runs out. Refiners, shippers, insurers and governments holding reserves cannot plan, he argues; Murban crude from the UAE jumped over $20 a barrel intraday. With Hormuz effectively closed, Bab el-Mandeb dangerous and Novorossiysk disrupted, rerouting breaks the supply chain links between oil in the ground and fuel.
• Iranian troops in Yemen mark a full-blown energy war. Iran left the roughly 10 million barrels a day of Hormuz bypass routes alone at first, he says, but the Houthis have now shut the Bab el-Mandeb and Iranian troops have deployed to Yemen to hold it. He expects every family buying fuel or food to feel it.

2026-07-23 · open the original ↗

• Chuck likes the screen, but it only works with recent Pis. The Raspberry Pi Touch Display 2 (10-inch, portrait) costs $80, which he considers comparable to other screens, and supports only the Raspberry Pi 5 and the Compute Modules, so a Pi 3 or 4 will not work. Raspberry Pi sent it to him "no strings attached".
• The main gain over the 5- and 7-inch versions is size. The specs he reads out are 10.1 inches diagonal, 1200 by 1920 native portrait, 224 PPI, IPS, 400 cd/m² (60% brighter than the original), anti-glare, 24-bit colour and 16 mm thin, with the driver board now inside the enclosure. Touch is 10-point (the 7-inch did five) with a 35 ms maximum response. The box holds only the screen, a DSI ribbon cable and a GPIO power cable.
• A fresh OS image is not enough: update the Pi's EEPROM firmware. On a never-booted Pi 5 the display showed nothing, and swapping the Pi, SD card and power supply did not help. Updating the EEPROM firmware made it recognise the touchscreen. He found the DSI ribbon cable flimsy and hard to seat.
• Touch and the auto-popup keyboard work well, but the desktop is cramped. He found touch very responsive in games, and the on-screen keyboard means no mouse or keyboard is needed. On the other hand, desktop targets are hard to hit, raising the text size broke the keyboard, there are no speakers, and nothing is provided for wall mounting; he screwed it to the wall with the screws holding the panel.
• Chuck argues a Pi makes a better wall dashboard than a tablet. He runs TouchKio, an open-source kiosk app (a locked-down full-screen browser) that comes back up after a reboot, showing three custom Home Assistant dashboards made through its WebSocket API. The Home Assistant server stays on a Proxmox VM, but he says everything could run on the Pi. His AI, which has full access to the Pi, builds new graphs on request.

2026-07-23 · open the original ↗

• The real problem is where the money goes, not technology. Answering the point that crypto was once going to save Africa and now AI is, the speaker (the passage also appears in the Okolo interview of 2026-07-17) cites Nigerian budget allocations that go to agencies without trickling down, politicians hoarding pallets of rotting cash, and health money spent on cars for hospitals instead of beds and medicine. The advice to African governments is to fix the basics first, then use AI to augment, rather than adding it to understaffed clinics without internet or stable electricity.

2026-07-23 · open the original ↗

• NASA teaches team scale, but its roles are rigid. One speaker who worked on the Mars Perseverance rover notes that 30,000 to 100,000 contractors touched that program, teaching large-scale coordination, but says NASA assigns each person a narrowly defined role on 10- to 20-year timelines — the reverse of the fluid, undefined work expected at an early-stage startup.
• Startups make researchers better by forcing faster iteration under pressure. A speaker argues the "pressure cooker" of running a company teaches you to abandon ideas lacking clear evidence of success and to tolerate discomfort in a way pure research roles don't demand.
• Scientists, not business-school graduates, may be best suited to found startups. One speaker says they used to assume entrepreneurship belonged to business people, but now argues NASA scientists are well positioned because a startup is "a series of long experiments" pursued without certainty of success, plus the added task of convincing others to care about what's being built.

2026-07-23 · open the original ↗

• European gas prices hit their highest since the Iran war began. As of 23 July 2026, the host says European natural gas crossed €62 per megawatt hour, which he attributes to a structural supply deficit.
• Europe will likely miss its pre-winter storage target. Citing Equinor, the host reports European gas storage at 54% of capacity, the second-lowest late-July level in 15 years, and Equinor's warning that Europe will likely fail to reach its 80% target before winter.
• Qatar's extended force majeure removes LNG during procurement season. Qatar Energy extended force majeure through mid-October after Iranian drone attacks in March damaged processing facilities at Ras Laffan, removing 20% of global LNG supply during the critical winter procurement window.
• Europe is losing replacement cargoes to higher-paying Asian buyers. Buyers in China, Japan and Pakistan, among others, are paying more for spot LNG, so US shipments are shifting from European ports to Asia.
• Reopening Hormuz cannot undo the deficit already built up. The host argues that injection is physically capped by compressor capacity, so three months of missed LNG cannot be forced into a four-week window; his read is that the shortfall is locked in and Europe enters winter with its thinnest gas cushion in years.

2026-07-23 · open the original ↗

• The US destroyed bridges linking Bandar Abbas to the rest of Iran. Bandar Abbas handles most of Iran's shipping and imports; he argues this differs from earlier strikes on nuclear, missile or drone facilities in that it lacks a clear military target and instead threatens the civilian supply chain for roughly 90 million Iranians.
• He distinguishes ordinary Iranians, who largely oppose the IRGC, from the regime. He cites the protests ongoing since 1979 as evidence that most Iranians are not IRGC supporters, and says he feels conflicted describing a strike that, in his view, punishes civilians rather than the government he wants to see fall.
• He frames the US as holding a severe targeting asymmetry over Iran. The US reportedly knows exactly what and where Iran can strike, while Iran cannot predict US targets among thousands of possible sites, which he warns raises the risk of a broader humanitarian crisis.

2026-07-23 · open the original ↗

• Stablecoin treasury holdings aren't a systemic risk, Sosnoff argues. Stablecoin issuers are becoming major holders of short-term Treasury bills, but he says that pool is tiny next to total Treasury debt, so it neither stabilizes Treasury demand nor creates institutional danger; Lossdog itself plans to launch its own stablecoin as a fintech product by mid-August.
• Analyst price targets on the newly-covered SpaceX span an implausible range. Initiations clustered within days of each other (July 7–17) produced targets from roughly $200 to $950–$1000, while Sosnoff cites the options market's own one-year expected move — priced by "high-frequency" firms — at about $79 against a ~$113–114 share price, calling the analyst spread evidence the targets are largely noise.
• A caller's backtest of 8,400 tickers over five years found no edge in entry timing beyond trend. The founder tested liquidity-sweep-and-reclaim entries and got no benefit over plain trend-following; Sosnoff and co-host Tony Bautista agree, saying the only things they can actually control are position size and price discovery, and everything else in entry timing is close to random.
• Rising leverage costs aren't yet a stock-market threat, the hosts argue, while the bond market is already repricing hikes. They cite Citadel's Q1 2026 report noting higher debt costs and short-term financing spreads up to 130 basis points, but say this mainly hits institutions, not retail; separately they read Fed Chair Kevin Warsh's comments as more worried about inflation than eager to raise rates, even as they note bond prices have fallen roughly five points in three weeks as the market shifts from pricing cuts to pricing hikes.

2026-07-23 · open the original ↗

• Sosnoff says his contrarian style was forced, not innate. He recalls that on his first day on the CBOE trading floor in 1981, survival as a market maker meant only taking the opposite side of whatever public customers wanted — a contrarian stance adopted out of necessity, not natural disposition.
• He treats "positive drift" as separate from his premium-selling strategy. He cites the market rising roughly 53-47 historically, closer to 60-40 over the last two decades, but says selling premium is a time-value and risk-mitigation tool rather than a directional bet, and claims that despite trading through a mostly directional market for 15 years, one big move has never wiped him out.
• He argues market outcomes form a wide bell curve, not winner-take-all. In his hypothetical of 100 traders, he expects roughly 50 to find some success — 34 moderate, 16 outsized — against a similar number of total failures, and attributes most retail losses to traders not believing in their own edge and betting too large a size on single positions.
• He claims a spotless long-term trading and business record. He says he's never had a losing year across 45 years of trading or a losing month across the businesses he's built, though he declines to give profit figures, noting only "millions" of both winning and losing trades. He closes by pitching Lossdog's new trade-feed feature as a source of ideas for viewers watching his live trades.

2026-07-23 · open the original ↗

• Three major oil arteries are now being choked at once. He reports Hormuz closed (20 million barrels a day of crude and products before the war), tankers driven off the Bab el-Mandeb (about 4.5 million barrels a day, roughly 70 vessels a day before the Houthi attacks on two Saudi tankers), and Caspian Pipeline Consortium loadings near Novorossiysk halted after Ukrainian attacks on tankers — more than 1 million barrels a day, about 80% of Kazakhstan's oil.
• One global market means each disruption tightens every other. He argues India, Europe and Asia will all bid for the same Atlantic Basin barrels, while ships rerouted around the Cape of Good Hope lose 30 to 40 days, burn more fuel, remove tanker capacity and triple delivery cost. He cites an estimate that Persian Gulf, Red Sea and Black Sea instability threatens a quarter of world supply.
• The shock absorbers are already spent. He puts the US Strategic Petroleum Reserve at 310 million barrels, the lowest since 1983, with US and European refiners at their limits, Russian diesel exports gone and refining margins at $70 a barrel or more. He lays out the sequence he expects: tanker rates, then insurance, physical premiums, margins, wholesale fuel, and finally pump and grocery prices — with margins and wholesale already moving.
• Iran wins by making the war too costly, not by fighting. In his reading, fear does the work: a missile costing a few thousand dollars threatens a tanker carrying $100 million of oil, so owners, crews and insurers enforce the blockade for Iran. He counts eight successful tanker attacks in the Iran conflict in two days, plus roughly 40 Ukrainian strikes on tankers in Russia over the past week.
• Losing Fujairah would leave Gulf oil no way around Hormuz. The UAE's pipeline to Fujairah carries a couple of million barrels a day from Abu Dhabi to the Gulf of Oman, within range of Iranian missiles and drones; he calls its loss "checkmate" because supply would then be rationed by price. He says markets complacently expect diplomacy and reserves to cover it, and with US exports back at pre-war levels, he expects more tankers coming for remaining US barrels.

2026-07-23 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-23 · open the original ↗

• Verticalized AI products lose their edge as models keep improving, Stan argues. He says the value of vertical AI "scaffolding" built around weaker models "just disappears" as frontier models get better, pointing to Anthropic's move into Legora's legal-tech space as a sign that only genuine network effects, not model wrapping, will keep vertical products defensible.
• Dust deliberately raised a small seed under "No GPU before PMF." Stan says Dust raised just $5 million at seed — a contrarian choice while peers were raising $100-200 million for training runs — to avoid what he calls the "coffin corner": raising more than a company can productively deploy and ending up forced into a down round.
• A latency comparison suggests frontier labs run roughly 70-80% margins. Stan reasons that because Claude Code and GLM 5.2 run at similar latency — implying similar underlying model size and cost — but GLM is priced about 9x cheaper, the gap reveals how much margin frontier labs capture at the token level.
• Model-agnosticism is a structural advantage the labs themselves can't copy. Stan compares locking into one model provider to buying factory machines that only plug into one energy company; Dust bets no single lab wins outright and has shifted from flat seat-based pricing to usage-based credits as agent loops and token consumption have grown, to keep its margins from compressing.

2026-07-23 · open the original ↗

• AI is the biggest platform shift yet for consumer products. YC notes that three years in, the only new icon on the home screen is ChatGPT, compared to Google and Airbnb from the web, or Instagram and DoorDash from mobile; it argues intelligence is now good enough to treat an agent like a person, and inference costs — currently around $1,000 per user per month in tokens — are falling roughly 10x a year, putting the next consumer boom "very soon."
• Whoever builds the next-generation AI-native consumer app now will own the category. YC points to getting things done, getting around, learning, staying healthy, managing money, and connecting with friends as categories ripe for reinvention.

2026-07-23 · open the original ↗

• A "gold swap" lets a struggling country raise cash without appearing to sell reserves. Because an outright gold sale signals insolvency and can trigger currency panic, a central bank instead swaps physical gold for dollars with the Bank for International Settlements, reclaiming the metal later by repaying the dollars plus interest.
• The swap functions as a collateralized loan used to defend a currency or pay foreign debt. The government uses the BIS dollars for exactly that purpose while the gold's title temporarily transfers away.
• Accounting rules let the swapped cash and physical gold share one balance-sheet line. That means a citizen or analyst reviewing the country's reserves sees a normal-looking total with no visibility into how much of the gold has actually been pledged away.

2026-07-23 · open the original ↗

• The video argues the US uses the same gold-swap mechanism itself. Under the Gold Reserve Act of 1934, the Treasury Secretary's Exchange Stabilization Fund can deal in gold, foreign currency, and credit instruments to defend the dollar without congressional approval or real-time disclosure.
• Historical and recent precedent is cited to support the claim. The Fed and Treasury used central bank swap networks through the 1960s-70s, and swaps reportedly worth billions occurred again in August 2024.
• Because the BIS doesn't disclose which banks are swapping, the public is described as structurally last to know. Blended accounting hides pledged gold until reserves are already depleted — the video's stated rationale for holding physical gold and silver directly.

2026-07-23 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-23 · open the original ↗

• Supabase bet on Postgres because no company owns it. Copplestone says that because no single vendor controls Postgres, every hyperscaler is forced to build its own competitive Postgres offering, continually improving the underlying database — and Supabase's positioning only took off once it repositioned from "real-time Postgres" to "open source Firebase alternative."
• Supabase obsessed over "time to value," which AI redefined. Copplestone timed launching, connecting to, and writing a row into an AWS RDS instance at about 8.5 minutes and set out to beat one minute (Supabase is now around 5 seconds); he says for AI agents, time to value is no longer about raw speed but about needing as little re-prompting input as possible.
• Two AI-driven surges reshaped Supabase's growth, the second larger. Copplestone says Supabase was "DDoS'd" by early AI app-builders Bolt and Lovable in December 2024, leading it to build "Supabase for Platforms," now used by over 50 companies; a second surge after the most recent holidays, which he attributes to Claude Code, pushed the company from about 6.5 million to 10 million developers while conversion and activation rose alongside growth, despite Supabase spending nothing on marketing.
• Supabase's next bet is "self-driving databases," not app-building. Copplestone says most agent effort today goes into launching apps, which is becoming easy to replicate, so Supabase is focusing on databases that manage security patches, uptime and scaling on their own, since operating a database long-term is the harder, more defensible problem.

2026-07-23 · open the original ↗

• YC wants to fund startups that move data centers onto the ocean. It argues this escapes the land, water, and permitting constraints that can delay or kill new data centers for years, noting the ocean covers 70% of earth's surface and offers abundant sunlight, no permitting process, and a natural heat sink; it envisions "compute flotillas" of standardized modular vessels operating together as one global cloud.

2026-07-23 · open the original ↗

• Both Middle East chokepoints are now effectively shut. He reports that, as of July 22, a second Saudi tanker was struck in the Red Sea. He reads that as effectively closing the Bab el-Mandeb Strait, with the Strait of Hormuz already closed, and calls it the nightmare scenario.
• Russia's Novorossiysk closure doubles the oil off the market. He reports that Russia closed Novorossiysk, its largest export facility, which ships over 2 million barrels a day. He says this leaves twice as much oil off the market as during the worst part of the war a couple of months earlier. He expects Trump to ask Zelenskyy to "lay off a little bit."
• He warns that even brief persistence of this outage turns severe. If the outage is sustained, he expects it to be "really bad." He says he cannot give a timeline but that it is "not that long," and that even a couple of weeks would be very damaging.

2026-07-23 · open the original ↗

• A device someone depends on should be repairable by them. The interviewer argues it should be open source and repairable, not something a big company controls and one person knows how to fix. The maker says its first product shipped conventionally, under warranty, but in practice the company fixed everyone's machines and sent its repair manual to a customer whose unit kept getting stuck in customs between the Netherlands and Britain.
• The newer scientific model's chassis was redesigned to come apart. The maker says repair is being formalized with it, while warning it is "not a simple task" and not like a Framework laptop. The comparison offered is a car: having spent a lot on it, you have the right to clear a clogged carburetor even though it is not simple.

2026-07-22 · open the original ↗

• AI missed its "multiplayer moment" that made past tools dominant, YC argues. It cites Google Docs beating Microsoft Word and Figma beating Photoshop by turning solo tools into shared team spaces, while AI work today stays single-player — a private chat window, or at best a read-only transcript link teammates can't edit.
• Long-running agent tasks make multiplayer collaboration necessary, not optional. As agents start running work that takes hours, days or even weeks, the essay argues that scale of work "was never meant to be done alone," and teams need to be able to jointly watch, redirect and hand off a live agent session the way they would with a human colleague.
• YC wants "multiplayer by default" agents built across specific team functions. It names engineers coding together, sales teams working a deal, support teams resolving a ticket, lawyers drafting a contract, analysts building a model, and marketers shipping a campaign as the use cases it's looking for founders to build shared agents around.

2026-07-22 · open the original ↗

• The Houthis have now shown they can close Bab el-Mandeb by force. A fully laden Saudi VLCC, too deep for the Suez Canal unless roughly half unloaded, U-turned north, then turned back toward the strait and was struck and set burning in the Red Sea well short of it. He says the strait had been shut without a shot until now; oil jumped over $1 on the report.
• The strike pulls Saudi Arabia, and its refineries, into the war. He assumes Saudi Arabia will respond, which he expects to move its roughly 3.5 million barrels a day of refining capacity up Iran's and the Houthis' target list, with 5 million barrels a day already lost elsewhere. If Saudi infrastructure were hit the way Russia's has been, he says "the sky's the limit" on oil and gas prices.

2026-07-22 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-22 · open the original ↗

• Local-for-local architecture insulates the selling, not the sourcing. PepsiCo manufactures heavy, low-value, taste-local products inside the markets where they are sold, so a tariff wall between blocs does not stop it making and selling chips or bottled drinks on either side. That makes it structurally unlike an arbitrage-global manufacturer shipping assembled goods across borders — but the briefing's point is that trade fragmentation threatens the input layer instead, which the finished-goods insulation does nothing for.
• Three input chains are already under pressure in 2026, and none can be localized away. Section 232 duties on aluminium (25%, up to 50%) drove a reported 65–70bp operating-margin headwind in the final quarter of 2025; smelting runs up to ~40% power cost, so primary production clusters, roughly 60% of it in China. A 10% duty applies to concentrate imported from Ireland. Phosphate and potash sit under export controls, with phosphate output ~44% China and reserves ~68% Morocco, and China having halted new phosphate export applications in December 2024. FY2026 core EPS growth was stepped down to mid-to-high single digits with tariffs named as a driver.
• The concentrate spine in County Cork is the one genuine arbitrage-global node, and the decision to make. Two plants — Little Island (1974, ~650 site employees) and Carrigaline (2003, ~€100m build) — produce the concentrate for Pepsi, Pepsi Max, Gatorade, 7Up and Mountain Dew and export to over 100 countries, fanning out into hundreds of finished SKUs. The recommended response is effectual rather than predictive: size a reversible second-node pilot by affordable loss, lock domestic and recycled aluminium and rPET supply while reclaimers are closing cheaply, and co-commit bottlers, can-makers and retailers as risk-sharing partners rather than forecasting where duties land.

2026-07-22 · open the original ↗

• PepsiCo's "local-for-local" manufacturing looks like deglobalization resilience but isn't the real story. Chips and bottled drinks are made inside the market where they're sold ($92B revenue, 200+ countries), so tariffs and trade walls don't stop production or sales at the finished-product level — the strength critics credit isn't actually where the exposure sits.
• One layer down, at the input level, PepsiCo is surprisingly concentrated. Two plants in County Cork, Ireland (operating ~50 years) supply all global concentrate for Pepsi, Pepsi Max, Gatorade, 7Up, and Mountain Dew to 100+ countries; roughly 40 factories across two Mexican brands supply most of the chips — a two-spot global chokepoint beneath an apparently diversified SKU footprint.
• Mondelez's structural risk is worse and more singular: chocolate itself. Nearly all global cocoa comes from just two countries, Ghana and Cote d'Ivoire, both under climate-change pressure — driving Nestle, Lindt, and others toward lab-grown chocolate. Mondelez ($38.5B revenue, 150 countries, 79% non-US) has one dominant choke point versus PepsiCo's two.

2026-07-22 · open the original ↗

• Coding agents have made small, bespoke software easy to build. YC argues existing clouds still make it hard to deploy and share, since incumbents like Azure and AWS were designed for software that scales to many users, at a cost of complexity unnecessary for one-off tools built for a single team.
• A cloud for small software still needs to solve permissions and sharing. YC's bar: sharing a small internal tool with non-technical colleagues should be as easy and secure as sharing a Google Doc.

2026-07-22 · open the original ↗

• The US Army says it has scrapped its old acquisition playbook. Secretary of the Army Dan Driskill says it's now seeking commercially developed, modular open-system technology faster, and is actively funding low-cost interceptors and components that lower the cost per kill, next-gen sensors, software, payloads and hardware built on an open architecture, drones, resilient logistics, and advanced manufacturing — all able to survive extreme climates from the Arctic to the subterranean.

2026-07-22 · open the original ↗

• Passive index funds now hold 54% of total assets under management. One host cites that figure alongside retail equity volume near 30%, but attributes much of that volume to options-market hedging and index arbitrage rather than stock-picking, leaving him unsure whether "the tail is wagging the dog."
• The hosts split on whether passive dominance helps or hurts price discovery. One argues cap-weighted index funds are becoming heavily concentrated in a handful of stocks and questions who would buy if that concentration breaks, without concluding that real price discovery is actually being lost.
• The other host says more capital always means more efficiency, full stop. He points to markets trading a penny wide as proof and argues it makes no difference whether that capital is passive or diversified — a position he calls "not open to debate."

2026-07-22 · open the original ↗

• YC frames a fictional AI tutor as now within reach. The Primer, a book from Neal Stephenson's novel The Diamond Age, adapts completely to a child over years, teaching not just facts but reasoning and ethics; the essay argues AI finally makes something like it possible, though it says "we're a long way from building one."
• The pitch universalizes one-on-one tutoring, historically reserved for the few. Citing Aristotle's tutoring of Alexander as the model of the best education, the essay argues AI can extend that same personalized approach to every child rather than just the privileged.
• YC wants a near-term product: AI teaching kids to read. It frames this narrower product — teaching reading, writing and arithmetic at consumer scale, aimed at parents rather than schools — as a supplement to teachers and as the entry point toward the far larger ambition of something resembling the fictional Primer.

2026-07-22 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-22 · open the original ↗

• The US and France each arm Ukraine for different self-interests. The US wants Ukraine producing its own Patriot missiles so American stockpiles are freed up for defending against Iranian drone and missile attacks in the Middle East.
• France's real motive, he says, is protecting African interests from Russia. He cites France losing military influence in seven West African countries to Russia's expansion into nine, with Russia now holding about 43 African defense contracts; France is also giving Ukraine about 16 Rafale jets by 2028.
• He reads a planned US munitions surge as a more aggressive posture. The US reportedly aims to raise Tomahawk production from about 90 to 1,000 a year and Patriot interceptors from 600 to 2,000, which he says leaves China's allies Iran and Russia both "toothless."

2026-07-22 · open the original ↗

• PostHog is rebuilding itself as "self-driving software," not analytics. James Hawkins says the company decided it's better to fully solve a customer's job than describe it, so PostHog now clusters product, support and session data to autonomously ship pull requests that fix bugs and improve metrics, with engineers increasingly reviewing and merging AI-generated fixes rather than writing them.
• Splitting the company by energy, not title, let him focus entirely on AI. Hawkins and co-founder Tim swapped their CEO/CTO roles as their interests shifted; when Hawkins wanted to make PostHog AI-first, Tim took over hiring, finance and scaling so Hawkins could spend a year and a half working almost solely with engineers on AI.
• Ambitious ideas are easier to fund, staff and market than narrow ones, he argues. Strong investors chase outsized upside asymmetrically, top talent wants to work on inspiring problems, and remarkable products spread by word of mouth in a crowded market — while modest "point solution" SaaS struggles to stand out even if it can still succeed.
• PostHog reached its current product only after roughly five pivots, tested by real commitment. Hawkins says they judged an idea's validity not by talk but by costly effort — like taking two trains and a bus to close an early $300/month customer — and abandoned ideas fast once attempts to raise the price showed customers didn't truly value them.

2026-07-22 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-21 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-21 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-07-21 · open the original ↗

• The Adjacency Premium — reachable-but-costlier beats cheap-but-blockaded — is near-zero under guaranteed sea control and decisive once that guarantee erodes. The proof case: Gulf aluminum (UAE/Bahrain/Qatar, ~20% of U.S. primary aluminum imports) was cheap because Gulf energy is cheap, but became worthless when Feb. 2026 military action closed the Strait of Hormuz — Hydro's Qatalum smelter shut down for an estimated 6–12 months. Tariffed but adjacent Canadian aluminum was the correct trade all along.
• Carney's 20-deal, four-continent trade diversification (aiming to double Canada's non-U.S. trade to ~$600B over a decade) diversifies partners without diversifying risk, since most new routes cross chokepoints Canada can't secure. Ranked by severance exposure: U.S./Mexico and the opening Arctic passage are lowest-risk; EU, Japan/Korea/Australia moderate; China/India high; UAE/Qatar highest — the strait that just closed.
• The deeper framing is Hardin's tragedy of the commons versus Ostrom's polycentric governance: as the U.S. Navy's single-power guarantee of open seas retrenches, "adjacency" (enclosure — each actor securing only what it can reach) is the rational individual response but fragments the system collectively. Canada's defense-bundled deals with Japan and Korea are read as early, harder-to-build polycentric co-governance rather than mere navy-renting — the only path the author sees as not ending in fragmentation.

2026-07-21 · open the original ↗

• The tariff authority is being rebuilt on new legal ground. USTR Greer announced 12.5% tariffs on roughly 99% of imported goods under a 1974 statute, justified by forced-labor claims, because the interim 10% tariffs the Supreme Court allowed after striking down the prior authority expire Friday.
• The Canadian aluminum saga shows what "adjacency premium" costs when ignored. After Trump's 50% tariffs on Canadian aluminum, US buyers shifted ~20% of sourcing to cheap Bahrain/Qatar smelters; the Iran war then destroyed those Gulf plants (out for 3–5 years), forcing buyers back to now-pricier Canadian aluminum plus the original tariff — a trade that was wrong the moment US sea control could no longer be assumed.
• Diversifying trade partners isn't the same as diversifying risk. Carney's campaign (20 deals across four continents, aiming to cut US trade share from ~75% to ~50% and roughly double non-US trade to $600B over a decade) still depends on sea lanes Canada cannot secure on its own, so the "adjacency premium" — near zero under a guaranteed US sea umbrella, decisive once that guarantee is withdrawn — can't be arbitraged away by cheaper unit costs.

2026-07-21 · open the original ↗

• Saudi Arabia struck the Houthis, Iran's last functioning proxy, in Sana'a. Hezbollah has been defanged by Israel and Hamas claims to have stepped down, leaving the Houthis as Iran's only proxy capable of offering meaningful support if Israel enters the war directly.
• He corrects himself: Israel hasn't actually been attacked by Iran. Unlike Qatar, Kuwait and Bahrain, Israel has been spared, which he speculates may trace to Israel's roughly 80-ton bomb dropped on Lebanon two weeks earlier — a fraction of Hiroshima's yield but still, in his view, a likely deterrent.
• He predicts Oman may eventually abandon neutrality for the US. Iran has struck US naval-support facilities in Oman despite Oman's professed preference for diplomacy, which he thinks could eventually push Muscat toward open cooperation with Washington.

2026-07-21 · open the original ↗

• A regulatory gap let the shortfall go undetected until it was too late. The retail cash-for-metals market lacks the capital reserve requirements of banks or brokerages, so by the time the SEC and New York Attorney General began investigating Rosland's IRA products and sales practices, its vault was already empty.
• A named community member's six-figure order illustrates the harm. She ordered gold and silver in February, was told of supply-chain delays, then had her refund calls refused; she is now working with the FBI and multiple state attorneys general and seeking other victims for a class action.
• The presenter frames this as one dealer's failure, not the industry norm, and discloses using ITM Trading (Arizona) personally for about seven years without issue — a personal endorsement, not independent verification.

2026-07-21 · open the original ↗

• Rosland Capital filed Chapter 11 on July 2 owing customers over $49 million. Employees were fired and phones went offline while the website stayed live, leaving IRA holders — including members of the presenter's community — with undelivered orders.
• Markups and commissions ate deeply into customer funds. Numismatic coins were marked up 50-100% over spot despite contracts capping the spread at 33%, and sales staff earned 15-35% commission — meaning up to $35,000 of a $100,000 IRA order went to a salesman.
• Delayed order fulfillment is described as the mechanism that sank the company. Rosland used new customers' funds to fill months-old orders, which worked while gold prices drifted sideways but became unsustainable once gold spiked, leaving over $60 million in accumulated losses and no metal in its vaults.

2026-07-21 · open the original ↗

• A debt-interest-to-tax-revenue ratio is used as the crisis indicator. The video cites Weimar Germany (1923), Hungary (1946), and Zimbabwe (2008) as precedents where rising ratios preceded a currency reset.
• The US is placed at 22% today, against Hungary's 33% breaking point. The video projects the US could reach that threshold in about four to five years, sooner if AI-driven job losses cut tax revenue or spending and yields rise — a stated projection, not a reported fact.
• 33% is framed as a historical pain threshold, not a fixed trigger, with the recommendation to keep stacking gold and silver patiently while tracking the metric.

2026-07-21 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.