Souk's Notes - Unofficial Summaries

2026-05-01 · open the original ↗

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

2026-05-01 · open the original ↗

• AI-augmented programming produces personal software that needs a home. Hanselman, who says they have changed role to work more with the Windows teams on making Windows a great place for developers, describes building "tiny tools" for an audience of one, such as a click-to-show-desktop tool, maximizing a window to a virtual desktop, and blood sugar management. Tiny Tool Town (tinytooltown.com) is where such tools can live, and Hanselman reports 325 submissions so far.
• The site runs entirely on GitHub, with no database. It is an Astro site in a GitHub repository. Tools are submitted through a GitHub issue form or a CLI.
• Free, licensed, open-source tools may get a Build shout-out. The rules are a license, a public GitHub repo and nothing commercial or promotional. Hanselman will feature favourites on the pre-Build show for Microsoft Build, June 2–3 in San Francisco, and asks for submissions within "the next couple of weeks". Hanselman and Mark Russinovich will do "a bit of a vibe check" there, and Hanselman says AI-augmented software engineering is more interesting than vibe coding straight into production.

2026-05-01 · open the original ↗

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

2026-05-01 · open the original ↗

• Imperfect consumer sensors pull people away from their own signals. The speaker relays that their cardiologist hates the Apple Watch because its ECG is one lead, not several. For every person it saves by catching AFib, the cardiologist says, a thousand come in worried about a resting heart rate or an Oura ring reading, and the advice is "just listen to your body."
• The feedback loop should run through self-reflection, not validation. The speaker worries about outsourcing judgment to measurement systems and wants to think, read the signals available, and decide without checking with Apple Health or ChatGPT. The example is a Post-it note reading "Talk to more people."

2026-05-01 · open the original ↗

• Imperfect consumer sensors pull people away from their own signals. The speaker relays that their cardiologist hates the Apple Watch because its ECG is one lead, not several. For every person it saves by catching AFib, the cardiologist says, a thousand come in worried about a resting heart rate or an Oura ring reading, and the advice is "just listen to your body."
• The feedback loop should run through self-reflection, not validation. The speaker worries about outsourcing judgment to measurement systems and wants to think, read the signals available, and decide without checking with Apple Health or ChatGPT. The example is a Post-it note reading "Talk to more people."

2026-05-01 · open the original ↗

• US hardware moves far slower than China's because of iteration speed. YC contrasts Shenzhen, where a team can go from design to a physical part in a day, with the US, where the same loop takes weeks, attributing China's edge to dense supplier networks and tight coordination between design and production.
• YC wants to fund startups that make US hardware iteration dramatically faster. It points to a startup already building faster turnaround from design to manufactured mechanical part as an early example of the kind of missing infrastructure it's looking for.

2026-05-01 · open the original ↗

• Safe districts hand the real election to the primary. He argues that when gerrymandering makes the general election a foregone conclusion, primaries dominated by the most ideologically extreme voters decide the seat, so the winners are whoever performs the most outrage rather than moderates or pragmatists who can work across the aisle. He grants Republicans started the current round with Texas but blames both parties.
• The result will be a Congress that performs instead of legislating. He predicts members skilled at inflammatory television and incapable of passing meaningful legislation, with the most extreme voices holding structural power because their districts were drawn to produce them.
• Moderate voters in safe seats lose representation entirely. Even a moderate incumbent must keep the primary base happy, so the broad middle has nobody fighting for it. He notes congressional approval has been below 20% for years and says redistricting will make Congress represent voters even less.
• Extremism feeds itself in a closed loop. Extreme candidates produce extreme governance, which produces polarization, which raises demand for extreme candidates. He forecasts states becoming uniformly red or blue with no purple states left, and says neither party "gets the edge" in any meaningful sense.
• Safe congressional seats are the most destructive thing in American politics. He says anyone holding one does not belong near Washington, and that the current round of redistricting is about to create hundreds more; until the maps are fixed, nothing else gets fixed.

2026-05-01 · open the original ↗

• Transformers trade recursive reasoning for parallel training speed. Chopard argues that unlike RNNs, which iterate on a compressed hidden state across time steps, an LLM computes in one feedforward pass with no compression in the time direction — every decode re-processes the full context. RNNs paid for their recursion with backprop-through-time, which caused vanishing and exploding gradients as models or sequences grew.
• Some problems are provably incompressible, unsolvable in one forward pass. He cites comparison sorting's known n·log(n)-step lower bound, plus Sudoku, mazes and rolling sums, arguing a transformer with fewer layers than the required steps cannot solve them. Chain-of-thought and tool-calling can simulate a Turing machine at test time, but only within human-demonstrated traces — an LLM taught only bubble sort, he says, could not discover merge sort on its own.
• A tiny recursive model outperformed giant pretrained LLMs on ARC-AGI. He credits the Hierarchical Reasoning Model — a 27-million-parameter model with no pretraining, trained on roughly 1,000 ARC-AGI tasks — with scoring about 70% on ARC-1 versus OpenAI's o3 scoring zero at the time. The follow-up Tiny Recursive Model, by a researcher he calls Alexia, shrank it to 7 million parameters by collapsing HRM's two networks into one, and pushed the score to 87%.
• The outer refinement loop, not the two-network design, drives the gain. Chopard says a researcher at Francois Chollet's company ran scaling ablations showing this repeated outer loop is the main source of HRM and TRM's performance, which is why TRM kept and simplified around it while discarding most of the rest. He expects the next leap to come from combining this recursion with giant pretrained LLMs rather than treating either as a substitute for the other, since today's recursive models are trained per task, not general-purpose.

2026-05-01 · open the original ↗

• Agent memory maps onto four kinds of animal memory. Richmond Alake, Oracle's director of AI developer experience (episode 985), splits agent memory into four types. Episodic memory is time-stamped, like conversation turns with role, content and timestamp. Semantic memory is world and institutional knowledge. Procedural memory is routines such as skills.md files, which he calls "SOPs for agents". Working memory is the LLM's context window. He and the host, who studied neuroscience, ground this in biology's role in inspiring CNNs through Hubel and Wiesel's experiments on the cat visual cortex.
• Storing skills in a database lets agents retrieve only what they need. Alake says Oracle is experimenting with putting skills in database tables, including vector representations, rather than files. Skills can then be exposed progressively instead of loading 100 skills into context, which he says allows scaling.
• Data engineering agents will curb junior hiring, not cut existing staff. Matt Glickman, CEO of Genesis Computing (episode 981), says no one is losing their job: existing data engineers were the limiting factor and now get "10x" more done. Junior hiring is what AI will "wipe out", and it is already starting. A few hires will be "conductors" running many agents in parallel; Glickman says a co-founder runs six while Glickman manages two and a half. Most schools still discourage AI use, which Glickman calls a "double whammy" for graduates.
• Coding is being abstracted, but AI infrastructure and systems thinking are surfacing. Linda Haviv (episode 987) notes non-technical people now talk about memory and context lag, which are infrastructure problems in other words. Engineers' edge, she argues, is knowing where systems fail, since even model training is an orchestration problem. People who know a domain should bring that context, as she did moving from media to building a news CMS.
• AI raised the floor in one school; teachers are raising the ceiling. Tracy Walker Griffith, principal of a K-8 innovation school in Boston (episode 983), describes fifth graders preparing for the state test. They write an answer, have AI in Magic School write one, explain which is better, and log wrong multiple-choice answers in a tracker the teacher checks for patterns. She says the school stays "platform-informed, not platform-loyal", also using PlayLab, which runs on Claude. AI literacy for both teachers and students comes first.

2026-05-01 · open the original ↗

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

2026-05-01 · open the original ↗

• Smith says the Iran war's costs have outweighed its benefits. The guest, Democratic Congressman Adam Smith, ranking member of the House Armed Services Committee, cites 13 US troops killed and hundreds wounded, thousands of Iranian civilian deaths, roughly 14 countries drawn into the fighting, gas prices up more than a dollar, and food-price increases still working through the system as fertilizer supply is disrupted.
• Smith says the war has left Iran's hardliners stronger, not weaker. He says the situation is now slightly worse than he expected: the IRGC and hardline regime have consolidated power since Ayatollah Khamenei's death, and the Strait of Hormuz remains effectively shut down, even though US strikes degraded — without eliminating — Iran's navy and ballistic missile arsenal.
• Smith separates a weakened Iran from a regime that's dug in. He argues shared wartime suffering may be uniting ordinary Iranians with their government rather than fueling revolt, and that with the regime still holding the guns and no armed insurrection underway, a revolution looks no more likely now than before the war began.
• Smith ties a wave of Pentagon firings to Trump demanding personal loyalty. He points to the ousted Navy Secretary (replaced by Hung Cao), Army Chief of Staff Gen. George, Gen. Hadney, and Joint Chiefs Chairman Gen. CQ Brown, and calls the rumored reason for the Navy Secretary's firing — an unbuildable battleship promised within two years, which he estimates would cost $22–30 billion — strategically wrong for an era of cheap drone warfare.

2026-05-01 · open the original ↗

• Surrounding yourself with agreement stops you from growing. The speaker argues that you will not find new solutions, and will fall into in-group/out-group bias, unless you deliberately seek out unfamiliar thinkers. Talking to strangers, such as people on the bus, disrupts your thinking patterns and gives you new information and frames, which is how decision-making changes.
• An LLM companion locks in one trained frame. The speaker suggests an LLM may lean toward a particular, perhaps neutral, way of deciding or seeing people. Relying on one while growing up, or as an adult, replaces "the randomness of a community" with that single programmed paradigm.

2026-04-30 · open the original ↗

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

2026-04-30 · open the original ↗

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

2026-04-30 · open the original ↗

• A lighthearted "bad coffee" bit built around a running dare, with no trading or career content. One host, wearing a shirt reading "iced tea is sick," is told he has to go "five for five" or face a joking forfeit; the clip is comic banter rather than analysis.

2026-04-30 · open the original ↗

• AI has erased the barriers to selling to the world's largest companies. YC argues enterprise leaders now actively seek teams that can apply AI to their problems, coding agents let a two-to-three person team ship nuanced products to large companies in months rather than years, and those leaders now understand what to build in-house versus outsource — making it common for a YC startup's first customer to be one of the largest companies in the world. This, YC says, kills the old model of startups operating in stealth for years to reach feature parity with an incumbent before launching.

2026-04-30 · open the original ↗

• Security is a subset of resilience, not its own discipline. Shortridge, chief product officer at Fastly and author of Security Chaos Engineering, calls "chaos" in the title a buzzword publishers needed. Resilience, she says, is recovering from failure of any kind and preparing for what comes next, whether that is a DDoS or traffic growth for good reasons.
• A security team's sense of control signals low resilience. Because systems and people are not deterministic, she argues that teams claiming full control of the delivery lifecycle are funding what feels good rather than what limits impact. She recommends small, low-impact experiments instead, such as a Fastly Compute function that strips cookies or auth headers from duplicated requests to test whether a login page really needs them, designed jointly with platform engineering.
• Compliance checklists and familiar metrics can actively erode security. She cites a paper by Josiah Dykstra (on HIPAA, she believes) finding that more compliance does not produce better outcomes, and a CISO who spent years telling auditors that removing SSH access had improved security. A rising vulnerability count can mean more trust between teams. In another case she describes, a security team chased 1,000 unexposed vulnerabilities while keeping its own admin credentials in a text file on a desktop.
• LLMs fit wherever they don't replace human judgment. That is her litmus test. LLMs suit tedious work like pulling facts from compliance documents, rubber-duck debugging, or testing how a resilience pitch lands across cultures. Her example: deception techniques draw "we don't want to be the bad guys" from US practitioners and "tell me more" in the EU. Asking an LLM whether your system is resilient will not go well, she says, and prose playbooks open security to more people but are a potential foot-gun.
• You own your dependencies; the vendor owns its platform. For Fastly, her own company, she says memory safety by design and multi-tenant isolation are Fastly's responsibility. Vulnerabilities in a customer's own code, or a database left open without an access control list, are not. Outsourcing specialist work such as AI bot management makes sense when it isn't your core business.

2026-04-30 · open the original ↗

• Agent tooling should widen the bandwidth between user and agents. Hinkle, CTO and co-founder of Nimbalyst, says multitasking works only with preparation: agents research and then stop, so each context switch leads into 20–30 minutes of focused review. Nimbalyst puts mockups, Excalidraw architecture, Prisma data models and annotation (circling part of a screen, clicking a table in an ERD) in one pane, so sessions don't lose context.
• Visual artifacts serve the human, not the model. He concedes the agent gains nothing from an ERD over the raw Prisma file. He is willing to spend tokens on the agent communicating better with him so that he can steer it. Files and diffs stay in front of the user, in local-first open formats on disk, which he presents as a deliberate design choice.
• One model reviewing another beats persona-based agent squads. He has Claude Code plan, Codex review the plan, Claude Code implement and Codex review the result, which usually catches errors. He uses Nimbalyst's built-in Claude "teammates" less than he expected. Their value, he says, is that each focuses his own conversation with an agent, not that the agent does better work. Hanselman shares the scepticism about "you're an expert database person" prompts and grants them only savings in the context window.
• Plans should document intent, not instruct the agent. Hanselman puts his own approach to him: "code sculpting", against a long plan.md as a "prose compiler". Hinkle says heavy plans were needed a year ago, but models have advanced. He still writes a plan for nearly everything, but it records the motivation and the points that can't be missed, so that later sessions can be checked against it.
• Agents can verify that something works but supply no taste. Nimbalyst is built in Nimbalyst. Its agent reads live logs, settings and the database, and it navigates the UI, takes screenshots and reads the DOM, looping toward a goal. Hinkle says it will make things work, and the human adds taste. He gives the status as free for individuals, an Electron app on Windows, Mac and Linux, with an iOS app first and Android coming, connected by an end-to-end encrypted tunnel through Nimbalyst's servers. His voice agent that controls coding agents is still in testing.

2026-04-30 · open the original ↗

• Semiconductor supply chains are still run on spreadsheets and calls. YC notes a single advanced AI chip takes 1,400 manufacturing steps across a dozen countries over five months, and cites 2021, when a $300 chip held up a $50,000 car and left $200 billion in vehicles unbuilt, as proof that companies see their direct suppliers but have no visibility into second- and third-tier ones.
• The bottleneck has only gotten worse. YC says Nvidia has locked up over 60% of TSMC's advanced packaging capacity, high-bandwidth memory is booked through 2026, export controls change quarterly, and the CHIPS Act is standing up new fabs in Arizona, Texas, Ohio, and New York that each need a supply chain built from scratch.
• This tooling doesn't exist yet, and building it needs deep technical knowledge. YC says real-time allocation tracking, multi-tier risk monitoring, and export compliance all require understanding wafer allocation and packaging constraints at a deep level — which is why it sees this as a startup opportunity rather than a feature bolted onto SAP.

2026-04-30 · open the original ↗

• The signal he is watching is a dollar anomaly, and he thinks it has now appeared. In a crisis the reflex is dollar up and Treasury yields down. Instead the 30-year touched 4.999% and the 10-year ran to roughly 4.43% while the dollar index rose only slightly. He calls that the first major crisis in which the safe-haven bid arrives alongside rising yields — an inversion of the pattern that held from Bretton Woods through 2019, and evidence that the bid is institutional inertia rather than structural demand.
• The apparent contradiction resolves once you stop treating Treasuries and the dollar as one trade. Japanese institutions selling Treasuries push US yields up while capping the dollar; simultaneously countries without meaningful reserves — Vietnam and similar — must buy dollars to pay for oil. Two flows running in opposite directions, both entirely consistent with what the screen shows. And the yen carry trade is a structural short of the yen, so covering it means selling Treasuries, moving into dollars, then buying yen.
• His read of the tape is that an institution moved size, and that the destination is Japan. The yen moved more than 2% at the open, which he takes as a block large enough to be repatriation of US holdings rather than ordinary flow, and asks whether it is the beginning of something larger. Given four places to put trillions, his judgement is that the yen is currently the answer — while noting China is preparing yuan-denominated gold-backed government bonds that would give large capital another route into the renminbi, which is itself a claim on dollar demand.

2026-04-30 · open the original ↗

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

2026-04-30 · open the original ↗

• Work on the side is insurance against being a number. Haviv argues that however good the people are, to the company as a whole you are "a human resource", so she never kept all her eggs in one basket; she chose software partly because it is a skill no employer can take away.
• Her personal brand fed her full-time job. Haviv says her job at Amazon came about because the company found her talking about cloud computing on TikTok, and that the side work made her better in the day job rather than distracting from it.
• Independent work lets her be more intentional than inside a company. She says committing to a single project or campaign without internal red tape is more efficient for her, and that she can keep advocating for tools she likes (Ray, AWS) from outside; the host lists her independent brand work as including Anthropic, Dell, Amazon, GitHub, Microsoft and Nvidia.
• Work is getting leaner, toward tiny billion-dollar companies. Haviv likens the present to the arrival of the internet and guesses that maybe 50% of jobs will be at small companies, with two- to three-person and perhaps one-person billion-dollar companies; she says plainly this is her own impression, "not some scientifically backed metric".

2026-04-30 · open the original ↗

• A prolonged war hurts energy importers far more than the US. The guest, who is not named in the clip, worries most about Europe and India, which lack energy independence. For the US consumer, they say gasoline makes up the budget of only 3% of households, thanks to better fuel efficiency. Frost frames the backdrop: Nasdaq and the S&P at all-time highs while Brent is up 50–65% since the war began, a split he says cannot hold.
• The wider commodity basket is the real worry, not gasoline. The guest points to helium and fertilizers and says that if the war runs long "there will be a pinch," with a weaker consumer "plausible."
• The war does not change the AI investment case. The guest's view is that "war has nothing to do with the AI food chain": the AI revolution is in its early innings, and investors still need exposure to those sectors whatever happens to the consumer.

2026-04-30 · open the original ↗

• Diversity of thought means seeking out unfamiliar people on purpose. Westerhoff, a principal AI security researcher on Microsoft's AI red team and Hanselman's colleague, keeps a daily noon reminder to talk to strangers. "If you sycophant yourself, you're never going to grow," she says. Random encounters disrupt patterns of thinking and update frames in a way that one LLM's trained decision-making paradigm cannot.
• LLMs are becoming people's introspection tool, with their context loaded. She reports a growing number of published studies showing people use AI for introspection. It is not the worst tool for that, she says, but the assessment may now come from the LLM rather than from your own mental back-and-forth. Hanselman's framing is that it is a mirror that also carries a corporation's weights and biases.
• Social skill is a muscle, and text-only practice misses the cues. She lost the muscle during COVID and got it back because she had trained methods to return to. She takes up Hanselman's analogy of the Smith machine against free weights and its stabiliser muscles. Microexpressions and other cues, picked up across many settings, are how people build dynamic skills, and rehearsing with an LLM does not supply them.
• LLMs push users to select from options instead of assessing. She bases this on decision-making research. Offering a menu suits decision fatigue, but skipping the assessment loses how your values interact with the choice, which is where the insight and better outcomes come from. An LLM widening the options (beyond Indian or Noodles) helps, but it still does not get you the answer.
• Intentional use is the line, and its trade-offs are poorly signposted. Her team studies the psychosocial impacts of AI. She grounds herself in knowing that an LLM "best friend" engages the same neurological systems as a human one, and she expects tools that build in that awareness to become the frontier for AI in relationships, coaching and education. Hanselman draws the line between deliberate companionship, such as robots for lonely older people, and drifting into a relationship like the one in the film Her.

2026-04-29 · open the original ↗

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

2026-04-29 · open the original ↗

• The guest dates the Iran invasion plan back to 2006, not this year. Chris Goldsmith, a veteran extremism researcher, cites a Marine intelligence contact whose earliest Pentagon assignment involved planning an Iran invasion, and says Hegseth aide Dan Caldwell was fired about a year ago for opposing the war — evidence, he argues, that the decision to go into Iran long predated February's strikes.
• Kurdish forces refused to serve as ground troops, undercutting the air-only campaign. The guest says the Pentagon tried recruiting Kurds from northern Iran and Iraq to backstop the bombing, but they refused, distrusting Washington after past abandonments — a mismatch he says US intelligence had already warned Trump about before he sided with Israeli PM Netanyahu's assessment.
• The guest calls Tulsi Gabbard a chronically toxic, unpredictable leader. Drawing on years working with her congressional office — marked by high staff turnover and votes that reversed positions her own team had prepared — he warns her role overseeing national intelligence risks a lasting "brain drain" as staff leave and don't return.
• The guest argues targeted economic pressure, not marching, is what actually hurts the administration. He points to canceling subscriptions to major companies (citing Scott Galloway's "resist and unsubscribe") as more effective than protest attendance alone, while predicting a hot, unrest-prone summer as war costs and a weakening economy compound going into the midterms.

2026-04-29 · open the original ↗

• Coding agents now let ordinary users become their own "forward-deployed engineers." YC notes that before AI, nearly all software shipped one interface for everyone — Netflix looks the same for every user, just with different imagery — with deep per-customer customization limited to enterprise software built by human forward-deployed engineers, and argues users can now radically customize interfaces themselves.
• This implies rethinking how software gets delivered. YC raises open questions like whether companies will need to ship source code instead of packaged binaries, and whether users' agents should be able to modify only front-end elements or also middleware on the fly.

2026-04-29 · open the original ↗

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

2026-04-29 · open the original ↗

• A small mini PC runs Project Nomad as well as a big workstation. On Nomad's own benchmark, the host's Minisforum UM890 Pro (Ryzen 9 8945HS, Radeon 780M, 64 GB RAM, 1 TB NVMe) scored 75 against 73 for his roughly 6-year-old Threadripper with 128 GB RAM and an RTX 3090 Ti. Project Nomad is Crosstalk's own offline server.
• The big GPU still wins for local AI. The host concedes the 3090 Ti is better hardware for LLM inference and chat, but says the mini PC is all you need for the knowledge-base content (Wikipedia, Khan Academy, offline maps), so a large, expensive server is unnecessary.

2026-04-29 · open the original ↗

• YC is launching a Paper Club to keep builders and researchers current on frontier AI. Framed around how hard it is even for a PhD student to keep up, the group will meet every few weeks at YC's Mountain View office to present and discuss papers, followed by an off-the-record dinner; the first cohort caps under 100 people and meets May 20.

2026-04-29 · open the original ↗

• He rejects Hasan Piker's "dying American Empire" framing as confusing change with death. He argues no plausible successor exists — not Western or Eastern Europe, South America, or Africa — leaving only a China-led order, which he doubts given China's limited power projection abroad and BRICS nations' reluctance to trust the yuan.
• He disputes Piker's framing of US imperial history as uniquely tragic. He counters with the USSR's roughly 30 million WWII dead and China's Great Leap Forward death toll, arguing no war involving the US has approached that scale since World War II, which he credits with preventing a third world war for 80 years.
• He calls Piker's claim that the Soviet collapse was a great 20th-century catastrophe indefensible. He argues no Russian, Ukrainian, Belarusian, Kazakh, Uzbek, Turkmen, Georgian, or Armenian would agree with that assessment.

2026-04-29 · open the original ↗

• Iran's navy fired on an Indian ship it had just cleared to pass. The vessel Sanmar Harold, along with a second Indian ship, was attacked in the Strait of Hormuz despite being granted passage, prompting a formal protest from India even as Iran's foreign minister, Sayyid Aragachi, had said the strait was open to commercial traffic for the ceasefire's duration.
• The IRGC navy publicly called its own foreign minister an "idiot" the same day. He reads this contradiction as proof of serious dysfunction inside Iran's government and military, and evidence the IRGC no longer controls the strait the way it once did.
• He predicts Iran's economy will be "severely hobbled" even if the blockade lifts. With thousands of IRGC targets destroyed and a US naval blockade in place, he argues Iran has lost its prior leverage over Gulf shipping, while falling crude prices are cushioning the rest of the world in the meantime.

2026-04-29 · open the original ↗

• Persia is presented as the one conquered society that fully resisted Arabization. Unlike Aramaic-speaking Syria and Iraq, Coptic Egypt, or Berber North Africa — where indigenous languages nearly vanished or gained official recognition only recently (Berber in Morocco, 2011) — Persian identity and language survived intact.
• The Rashidun Caliphate's conquest of Persia in 651 AD imposed Arabic rule for decades. Arabic became the administrative language, non-Muslims faced pressure, and the Zoroastrian priestly class was dismantled while Arab tribes settled as a ruling class above Persians — an arrangement he says held for roughly 50 years despite scattered cultural resistance.
• A Persian cultural revival is dated to 819 AD under the Samanid dynasty. Ruling with real autonomy from Bukhara, in modern Uzbekistan, under nominal caliphate authority, the Samanids patronized Persian-language writers and artists, sparking what he calls a "Persian Renaissance" that reshaped the wider Islamic world, though their attempt to revive Zoroastrianism failed.

2026-04-29 · open the original ↗

• He argues China is the only serious long-term rival to the US. Russia is tied down by a NATO-backed Ukraine, and Iran's military capacity has been degraded by more than half, by some estimates 80%, reducing it to a regional nuisance with no ability to project power abroad.
• China could out-build the US 232 to 250 times over in a year of shipbuilding, he claims. Its navy is large and impressive but still trails the US overall, and he says China has not yet demonstrated an ability to project power outside its own borders.
• He calls China's food and oil imports through the Strait of Malacca its key vulnerability. China gets roughly 80% of its oil and two-thirds of its overall imports through that strait, a chokepoint the US could blockade, sharpened by a new US-Indonesia defense deal granting American overflight access — which he says is why China currently prefers business to war.

2026-04-29 · open the original ↗

• A Pentagon official reportedly threatened the Pope over criticism of the president's Middle East actions. The official reportedly said the US "has the military power to do whatever it wants" and invoked the 14th-century Avignon Papacy, when the French monarchy stripped papal power and forced the papacy's relocation to France — described as the first such recorded threat from a US official.
• US-Catholic tension is traced to the 1800s, cementing after Lincoln's assassination. Catholic co-conspirator Mary Surratt's execution is credited with fueling suspicion that American Catholics answered to the Pope over the president, leading the US to cut diplomatic relations with the Vatican and the Holy See by 1867.
• Catholic immigrant labor organizing later got mislabeled as socialism or communism. Irish, Italian, and Polish workers organizing unions were branded as such even though — the source cuts off here — deepening the pattern of American Catholics being treated as politically suspect.

2026-04-29 · open the original ↗

• He argues "dollar collapse" headlines misread a narrower gold milestone. The real change is that gold overtook US Treasuries as the single largest reserve asset for the first time in 30 years — not that the dollar itself is being abandoned, since forex reserves (still dollar-dominated) remain about 72% of the $18 trillion in total official reserves.
• Dollar forex reserves total $7.4 trillion, or 41.1% of all reserves, versus gold's 22.2%. Comparing gold specifically to Treasuries, gold now holds $4 trillion against $3.9 trillion in Treasury holdings — the specific crossover generating the "30-year first" headlines.
• He attributes the shift to uncertainty from US tariffs and the Iran invasion. Institutions are hedging into gold as a safe haven, he argues, rather than abandoning dollar-denominated assets like Treasuries.

2026-04-29 · open the original ↗

• He rejects the consensus explanation on arithmetic rather than ideology. The line that the UAE left OPEC in order to produce more makes no sense when roughly 10 million barrels a day of capacity has been destroyed and every producer will be running flat out for three to seven years — there is no binding quota to escape. If the stated reason cannot be true, the actual reason sits somewhere else, and he treats the consensus framing as propaganda worth asking who benefits from.
• His reading is that the UAE is exiting an architecture, not a cartel. OPEC's three functions are dollar-denominated price discovery, dollar-cleared hedging, and dollar-anchored pricing formulas — all of them paper. The UAE has spent years building the parallel rails as insurance against the day staying with the dollar costs too much: CIPS, yuan pricing, gold. And it just lost 44% of throughput to a neighbour's chokepoint while pegged to a currency that same closure is degrading. Leaving the price-coordination layer of an architecture is a different act from leaving a production quota.
• He keeps the time-horizon caveat that makes the call falsifiable. Exiting at the political layer is a one-way door even where operationally reversible — the UAE could rejoin — and the bet is that recovery is slow enough that re-entry never becomes necessary. He links it back to the currency-swap analysis: papering over lost control of physical oil may work in the short run but degrades the dollar over a 12-to-36-month horizon, and he reads the exit as confirmation of that. He also notes, ruefully, discovering more than a hundred unanswered direct messages going back a year after reinstalling TikTok.

2026-04-29 · open the original ↗

• Hassabis expects AGI soon, but flags one or two open unknowns. His personal timeline is around 2030; he says continual learning, long-term reasoning and some aspects of memory remain unsolved, and puts it at "50/50" whether closing those gaps needs one or two genuinely new ideas or just incremental scaling of what already works — DeepMind is pursuing both bets at once.
• Today's reasoning models are only "fairly brute force," producing jagged intelligence. Watching Gemini play chess, Hassabis says its own thinking trace will identify a move as a blunder and then play it anyway — evidence, he argues, that models lack introspection into their own reasoning, which is why the same system can solve IMO gold-medal problems yet still make elementary math errors.
• Agents haven't truly arrived until a vibe-coded game tops the charts. Hassabis says no AI-built game has hit the App Store charts the way a studio-made hit would, and expects that to change in the next 6 to 12 months; he predicts individual builders working at roughly 1000x their prior output will appear before agents operate fully autonomously.
• The AlphaFold pattern: combinatorial search, a clear objective, plenty of data. Hassabis says that formula — a massive combinatorial search space, a clear objective function to hill-climb, and enough data or simulation to generate synthetic training data — drove both AlphaGo and AlphaFold, and advises founders to pair deep AI expertise with a hard science domain like materials or medicine as one of the few bets defensible against the next foundation-model update.

2026-04-29 · open the original ↗

• A team of specialist agents beat generalists and a single agent. Gaster, who works at Microsoft on GitHub Copilot app modernization, says he gave the same "build a snake game using Aspire" prompt to one agent, a team of five generalists and a team of specialists each fed its own large Markdown knowledge file (on App Service, Container Apps, Aspire, Functions). The single agent produced about 2,000 lines and 80 tests slowly and badly; the specialists wrote about 20% less, cleaner code with fewer tests that covered more, and ran 18–20% more efficiently than the generalist team on average over repeated test runs.
• Generalists fight over work and hallucinate; specialists stay in lane. Gaster says generalists have a "hungry hungry hippo" effect, grabbing any task and hallucinating because they carry too much knowledge, while a specialist tester will not write code. A coordinator is always required, and the team can "hire" new agents when it finds a gap: on a Swift app it asked for a front-end designer, then a marketing agent.
• Turn on the lockout policy; Gaster says it stops hallucinations. It is off by default and set with an instruction such as "set it to two": after two failed attempts at a task, as judged by a tester agent (his example is Playwright checking a colour), that agent is barred from it, others try, and if none can do it the team hires one. When the task passes, every agent writes what it learned to disk so the error is not repeated.
• Squad's memory is just Markdown files, written constantly. A "Scribe" agent in every squad writes decisions to files, so after a Copilot CLI crash (which Gaster says hits him once or twice a day) the squad can say where it was and resume; Giard adds that the files are checked into source control, so later developers see the decisions. Agents refuse to move a secret pasted into chat because it would land in a file; a storage abstraction with a SQLite vector store, built by Dina Berry after Jeff Fritz asked for one, was committed the night before recording, and longer-term-memory experiments are planned.
• Refresh the team after about a month of work. Squad starts with no more than about five members; Gaster tells a long-running squad to clear its context, report how much it cleared, then hold a candid meeting on its performance and restructuring. His overloaded Git agent "Kobayashi" was let go and the team asked for two replacements, one for CI/CD without Git ops and one for Git ops without CI/CD.

2026-04-29 · open the original ↗

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

2026-04-29 · open the original ↗

• Non-technical builders now hit AI-infrastructure problems in other words. Haviv says that once people move past toy examples to production and scale, they run into systems issues; non-technical people now talk about memory, context that "is not working" and lag, which are the same problems as KV cache and prefill-decode disaggregation without the vocabulary.
• Engineers' edge is knowing where things fail. She argues that agentic workflows sit on a distributed-systems layer, and that even model training and mixture-of-experts models are orchestration problems, so someone from DevOps or software brings real strength even without AI/ML expertise. The host's premise is that within "a couple months" it has become theoretically possible to do the whole AI pipeline in natural-language prompts.
• Domain knowledge transfers, so start where you already understand users. Haviv, who taught herself to code in 2015 and took her first JavaScript job in 2016, says her media background let her build a content management system for journalists because she had written the articles herself; the host draws the lesson that a newcomer may do better moving into development at their current employer than applying to Meta.
• Building and media are democratised, so curate what you build. She says creators no longer need a $50,000 studio or a label, and that she saw in TV news that specialists such as lawyers and doctors who came on as experts kept control of their time. The host adds that hands-on experience in a field gives more confidence in AI advice and warmer contacts than a cold email; Haviv notes junior developers struggling to find jobs out of college are also turning entrepreneurial.

2026-04-28 · open the original ↗

• Space-optimized inference chips are an emerging startup opportunity. Philip Johnston, co-founder and CEO of Starcloud (building data centers in space), argues reusable rockets from SpaceX and Stoke Space will drive a huge increase in orbital compute capacity, creating demand for chips tuned for mass, thermal and radiation constraints; he's looking to hear from chip designers with experience at SpaceX or Nvidia.

2026-04-28 · open the original ↗

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

2026-04-28 · open the original ↗

• The event is a sovereignty assertion timed for tactical effect rather than a commercial decision. Announced 28 April effective 1 May — three days before OPEC's scheduled Vienna meeting, and with no consultation of Saudi Arabia. This is the first major member exit in OPEC's sixty-year history, from a country that has been a member through Abu Dhabi since 1967, four years before the federation itself was formed.
• The cascade runs through three layers with measurable thresholds at each step. Physical: UAE production fell from 3.4M to 1.9M bpd in March, a 44% drop, while OPEC aggregate fell 7.88M bpd — down 27% to 20.79M — the largest single-month OPEC supply destruction on record, exceeding 2020, the 1991 Gulf War and the 1970s. Cohesion: with spare capacity at historical lows the quota constraint becomes binding, Saudi Arabia is forced to absorb the shock alone, and the UAE was producing roughly 30% below its 4.85M bpd capacity at filing. Institutional: sovereign exit from the coordination body itself.
• The candidate concept is defined with an explicit threshold, and the entry is careful about what it does not yet claim. Cartel Defection Under Chokepoint Stress: when physical-layer disruption crosses a threshold — here roughly 44% production loss for the defecting member — the cost of cartel discipline exceeds the price-coordination benefit, and exit becomes rational. It is explicitly distinguished from settlement-layer defection over payment rails and from paper-power degradation, since this mechanic operates at the producer-cartel layer and is triggered by stress on the defector specifically rather than the cartel as a whole. Status is candidate pending a second instance, with Iraq the obvious next test and Kazakhstan and Nigeria on the watch list. The line that generalizes it: the mechanic is not UAE-specific, the threshold is.

2026-04-28 · open the original ↗

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

2026-04-28 · open the original ↗

• Investors should never be "passive" about an existing long position — selling out-of-the-money calls against it is encouraged even near highs. The host argues that customers who panic over being slightly down on the short calls while deeply up on the underlying stock are misreading the position: selling calls now would have capped some recent upside, but would preserve a chunk of the profit if the stock sells off instead.

2026-04-28 · open the original ↗

• Iran's toll regime is being institutionalized through two parallel intermediation tiers, and both bypass Washington entirely. Tier 1 is sovereign: non-aligned states route vessel access through Beijing as broker, anchored on Thailand approaching Wang Yi, who transmits the transit request to Tehran for clearance of specific vessels. Tier 2 is commercial: Western shippers route access through Marsh war-risk coverage clauses that require shipowners to coordinate directly with Iranian authorities as a condition of cover. Different counterparties, identical underlying flow.
• Both tiers settle through the same non-Western rails, which is what makes them one system rather than two responses. Yuan via Kunlun Bank and CIPS, with USDT on Tron as the alternative. The toll architecture runs up to $2M per VLCC and roughly $1 per barrel for oil tankers, with ad-hoc rates for non-tankers. Stripped of the difference in intermediary, the sequence is the same every time: vessel requires transit, intermediary obtains Iranian clearance, payment moves through Kunlun or Tron, vessel transits the strait.
• The structural claim is that the political and military layer has become disconnected from the architecture being built beneath it. This is presented as empirical proof that the toll-normalized path is the active trajectory rather than one scenario among several — and the sharpest evidence is that Marsh writes the coordination clause despite the announced blockade. On present evidence the US response is structurally absent from the flow it is trying to prevent.
• The escalation worth watching is contractual rather than military, and the tripwire is specific. Whether the wider Lloyd's market follows Marsh, and whether Munich Re or Swiss Re write the coordination clause into reinsurance treaties — which the entry treats as full architectural lock-in, since the primary insurers lay off to reinsurers on trillions of dollars of exposure. And the single line to watch for: if Marsh's language ever explicitly names the yuan or the Kunlun rail for fee transmission, the dollar bypass becomes contractually embedded inside Western paper.

2026-04-28 · open the original ↗

• His organising principle is that treaties only legalise what has already happened on the ground. After 1945 the lines were drawn where the armies stopped, not where negotiators preferred. Korea simply stopped fighting and has still never signed a peace treaty. So the useful practice is watching facts accumulate rather than waiting for documents — and on Hormuz the facts are that ships are already paying to leave, through two channels that have nothing to do with Washington.
• The Western half of the arrangement is what he finds genuinely striking. Thailand asked China to ask Iran, and Beijing obtained clearance — that much is a familiar non-aligned workaround. But separately Marsh has told shippers they must produce Iranian sign-off before it will insure a Hormuz transit. So a Western insurer is now making the Iranian toll regime functionally legitimate, using rails the US sanctions others for using, while the US simultaneously declares a blockade. His summary: the West is being inserted into the new financial system rather than resisting it.
• The tripwires he will watch are all commercial. Whether Vietnam follows Thailand through Beijing — which he flags as the big one. Whether the other major shipping insurers adopt the same clause. And whether the reinsurers write it into treaties, which he calls game, set, match. His conclusion is uncomfortable and clearly stated: the United States is no longer in the flow, and its own companies are the ones legalising the payment system it refuses to recognise.

2026-04-28 · open the original ↗

• A drone swarm reportedly destroyed an AWS data center unopposed. He says a swarm of cheap Iranian drones took out the facility last month with nobody stopping it, and warns that a coordinated swarm of a thousand drones — far larger than the single hobby drones most counter-drone systems are built to handle — would be even harder to stop.
• The cost math today overwhelmingly favors attackers, not defenders. He points to a $3 million Patriot missile intercepting a $500 FPV drone as the current exchange rate, and argues restoring that cost advantage to the defense side is the real problem to solve.
• YC wants to fund the counter-swarm stack as software, not weapons. He lists interceptors that neutralize dozens of drones at once, sensor-fusion software giving a single real-time picture, and non-kinetic tools like rotor-fouling aerosols and entangling streamers — arguing that with radio jamming becoming obsolete, the winning companies will resemble Cloudflare more than Raytheon.

2026-04-28 · open the original ↗

• The Hormuz blockade is porous but still targets Iran specifically. Eight vessels, including six Iranian-linked ships and one US-sanctioned tanker, departed the strait despite the blockade, though the US turned away two Chinese vessels — evidence, he argues, that the policy is aimed at Iran's shipping rather than the strait broadly.
• He links three chokepoints as evidence of growing US leverage. Majority influence over the Panama Canal (2.3 million barrels of oil a day) secured last year, the Hormuz blockade, and a new US-Indonesia defense pact granting American overflight access near the Strait of Malacca together give the US outsized reach over global shipping.
• The Strait of Hormuz carries 20 million barrels of oil daily, Malacca 23 million. He notes 40% of China's oil transits Hormuz and over 80% transits Malacca, meaning the US now holds disproportionate influence over two of the chokepoints supplying both China and Southeast Asia.

2026-04-28 · open the original ↗

• AI automation of companies is now blocked by scattered internal knowledge, not model quality. The speaker argues every business needs a "company brain" — a system that extracts tacit know-how from emails, Slack, support tickets and databases into a living, current "skills file" so AI agents can execute tasks like refunds or pricing exceptions safely and consistently.

2026-04-28 · open the original ↗

• The US blockade targets Iran specifically, not the strait itself. Non-Iranian ships can pass freely, but the policy aims to cut off Iran's imports, an estimated 23-29% of which is food — mainly soybeans, soybean meal, and corn shipped from China, Turkey, and Brazil.
• Iran's currency crash to 1.6 million rial per dollar triggered deadly January 2026 protests. He reports the IRGC killed a reported 30,000 to 50,000 protesters in response, while food prices had already risen 42% by September 2025 amid ongoing drought.
• He predicts the blockade is meant to provoke unrest that could topple the government. China, Iran's largest oil buyer, stands to lose the most; China's defense minister has said Beijing intends to keep honoring its Iran contracts via the strait, but he predicts China will instead simply buy more oil from Russia.

2026-04-28 · open the original ↗

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

2026-04-28 · open the original ↗

• Leaner work makes this the time to go independent. Haviv, who had just left her staff developer advocate role at Anyscale at the time of recording, guesses that maybe 50% of jobs will be at small companies and that two- to three-person, perhaps one-person, billion-dollar companies are coming; she flags this as her own impression, not a measured figure. She treats side work as insurance against being "a human resource", noting her Amazon job came from the company finding her cloud-computing videos on TikTok.
• AI-infrastructure problems are now everyone's problems. Haviv says non-technical builders now complain about memory, context and lag, which are infrastructure problems in different words, and that past toy examples the work becomes systems thinking. She argues engineers keep an edge because they know where things fail, since even model training and mixture-of-experts are orchestration problems; the host frames the premise that the whole AI pipeline has become doable in natural language within "a couple months".
• A Python-native framework spares ML engineers distributed-systems expertise. Haviv describes Ray as a Python-native open-source distributed computing framework, created at Ion Stoica's Berkeley lab by Robert Nishihara and others for reinforcement learning, now spanning Ray Data, Ray Train and Ray Serve with native vLLM and Hugging Face integrations. She says it suits multimodal, GPU-bound AI workloads better than Spark because those workloads do not fit on one GPU.
• AI infrastructure is infrastructure shaped by compute-heavy AI workloads. She defines it as infrastructure specific to AI/ML workloads, which are compute-heavy where older workloads leaned on data input and output, and says inference is where users first hit the difference. Her popcorn analogy for batch inference: popping one kernel is latency-optimised, a microwave full is throughput-optimised but costs money while on, so CPU preprocessing must keep the GPUs fed. She casts neoclouds (Nebius, CoreWeave, Crusoe) as specialists beside hyperscalers that "do everything".
• Open governance of the AI stack matters more than open weights. Haviv says it is crucial that Ray, vLLM, Kubernetes and MCP sit under the Linux Foundation rather than being owned by one company, and expects open models such as the newly released Gemma 4 and DeepSeek to catch up with proprietary ones. She runs Gemma 4 on her phone for offline use on flights.

2026-04-28 · open the original ↗

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

2026-04-28 · open the original ↗

• Multiple recession indicators are already flashing, the guest argues, even as GDP looks fine. PhD economist Katherine Edwards, a former RAND researcher and Bloomberg columnist, says hiring has been frozen for roughly a year and a half and 2025 saw the weakest job growth outside a recession this century — possibly marking a Q4 recession start by that measure alone — while GDP still looks healthy partly because it was inflated by import stockpiling ahead of tariffs.
• She likens the economy to cars braking toward a red light, not yet stopped. Growth, jobs, and unemployment data show slowing rather than collapse, and she argues the US "has everything it needs" to avoid a recession, but says the administration lacks the leadership or willingness to act on the warning signs.
• Her proposed "third rail" fix: delink healthcare from employment entirely. She would end the tax exemption on employer-provided health benefits, redirect that revenue into a universal Medicaid-style system covering primary care, and let private insurance shrink to catastrophic-only coverage.
• She frames the moment as needing an "FDR," not a visionary. Drawing on a memoir of Frances Perkins, she argues Roosevelt succeeded not through grand design but by simply acting on the fact that people needed help — a model she says fits an economy where wage stagnation since the Reagan era and rising healthcare costs have built decades of pressure now reaching a breaking point.

2026-04-28 · open the original ↗

• Viktor Orbán conceded defeat after 16 continuous years in power. Hungary's PM since 2010 (and earlier from 1998-2002) lost to Péter Magyar's EU-friendly, Ukraine-friendly center-right Tisza party.
• Hungary's economy badly lagged similarly-sized peers under Orbán, he argues. Comparing 2010-sized economies, Vietnam's GDP grew 223% versus Hungary's 68.63%, leaving Hungary the EU's poorest member state despite Fidesz's long rule.
• He predicts Orbán's exit strips Russia of a key EU/NATO ally. Orbán's government blocked Hungarian military aid to Ukraine and stands accused of leaking sensitive information to Russia; with him gone, only Slovakia's Robert Fico remains as a pro-Moscow holdout pending Slovakia's 2027 election, which he expects will mean more EU and Ukraine aid.

2026-04-27 · open the original ↗

• Falling diagnostic and gene-therapy costs, paired with agent harnesses, will enable personalized medicine at scale. The speaker says genome-sequencing costs have fallen faster than Moore's Law and that N-of-1 mRNA-based therapies plus greater FDA openness are emerging, arguing agents like Claude Code can combine genomic, EHR and wearable data into individualized treatment suggestions.

2026-04-27 · open the original ↗

• Episode 900 is a montage of guests' closing lines, not an interview. It strings together dozens of past guests riffing on the show's "technology and friends" sign-off, with no host questions and no single topic; the only speaker who names himself is Jeff Fritz.
• The common thread is that tech careers are built on friendships. Several guests credit conferences, speaking and user groups with friendships worldwide, one dating it to publishing and speaking from the late '90s and saying their company was founded on working with friends, and an entrepreneur advises starting a company with friends who share your vision and principles.
• A minority of clips warn about AI and technology's risks. These include "Friends don't let friends use GenAI technology without some safeguards", "Friends don't let friends build irresponsible technology", that technology can isolate as well as connect, and one tongue-in-cheek claim that AI has eaten the speaker's friends. Others argue developers need to understand AI to set guardrails.
• One guest argues the GUI made computing antisocial. The speaker says clicking gave nobody anything to show colleagues, while PowerShell scripts got shared, admired and repaid in gratitude ("I owe Bruce a beer"), and that this restored the social joy of computing.

2026-04-27 · open the original ↗

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

2026-04-27 · open the original ↗

• Staying small and diversified makes it statistically impossible to be "unlucky." The argument is that the law of large numbers removes bad-luck outcomes over enough trades — you can still lose, but the odds shift in your favor.
• Sticking to one strategy or product, or copying other traders, breeds unlucky outcomes. The hosts argue that experimenting across many strategies and products, rather than being a "one-trick pony," lets the law of large numbers work in an investor's favor.
• Learning to take profits and build a personal "culture of winning" avoids the all-or-nothing trap. One host criticizes the "Fooled by Randomness" mindset of holding out to "hit it big," and argues even hedge, private equity, and venture funds ultimately succeed by respecting randomness and large numbers rather than superior knowledge.

2026-04-27 · open the original ↗

• The central asymmetry is stated before any forecast: resolving the war does not unwind what has already crystallized. Crisis-premium unwind and structural degradation run on different clocks and point in opposite directions, so any sequencing framework has to separate them or it will read one as the other. The Hormuz extraction architecture persists regardless of political settlement.
• Four phases, each with a different mechanism rather than the same trend extended. Phase 1, pre-resolution: the dollar takes the reflex safe-haven bid, but the diagnostic is the quality of the win rather than the direction — rising 10-year yields during an intensifying crisis signals the safe-haven function has degraded. Phase 2, T+0 to roughly six months: the yen posts the largest percentage move, driven by three compounding forces — a mechanical carry unwind, direct energy-import relief, and a washout in the most one-sidedly short major currency. The euro rallies but stays capped by breakup risk; the yuan appreciates modestly on management. Phase 3, T+6 to T+18: AI capex arithmetic reaches earnings, with Big Five capex of $602B in 2026 against roughly $400B a year of depreciation unsupported by profits, and negative Amazon free cash flow as the crack — producing a second dollar leg down that is tech-and-debt led rather than geopolitical. Phase 4, beyond T+18: sovereign fiscal moats become the dominant evaluation lens, measuring which governments actually forward-converted commodity revenue into foreign-asset stock that insulates solvency from coercion.
• Gold is the structural winner because the framework can find no clean sell phase for it. Its behaviour is increasingly governed by physical demand — Asian buyers, central banks, post-London Indian pricing — rather than by real interest rates, which breaks the old rule that gold falls when rates rise. Silver is judged a phase three or four instrument rather than a phase one or two: the crisis bid is real, but the market is small enough that institutional flows distort it more than gold, and the post-resolution pullback can be sharper. The Swiss franc is called the most overrated of the alternatives — a two-to-three-month tactical vehicle around acute windows, not a portfolio anchor. Norway's krone is named the cleanest sovereign fiscal moat play among fiat currencies.
• The disconfirming signals are named explicitly, which is what makes the sequence usable rather than decorative. Formal recognition of the Hormuz management plan would accelerate dollar weakness in Phase 2 by handing the bypass infrastructure institutional legitimacy. A Chinese property or shadow-banking event drops the yuan out of Phase 4 entirely and leaves the sequence without a clear structural winner. And a JGB crisis arriving before normalization completes would truncate the yen rally and reverse it violently.

2026-04-27 · open the original ↗

• AI-native companies that sell the finished service, not a software tool, are the next wave. The speaker contrasts 2023–2025 "AI co-pilot" startups with a new category that performs the work itself — citing insurance brokerage, accounting, tax, audit and healthcare administration — arguing total services spend dwarfs software spend and is easier to disrupt because much of it is already outsourced.

2026-04-27 · open the original ↗

• Farmers are trapped in a worsening chemical treadmill. The narrator argues pesticide resistance keeps evolving, the pipeline for new chemicals has gotten slower and more expensive, and farmers must keep spraying more for diminishing results even as costs rise and margins fall, because failing to spray risks crop loss and food-price shocks.
• Three technologies converging at once make the problem newly solvable. He points to cheap sensors and cameras giving AI real-time weed/pest identification, robotics precise enough to treat individual plants instead of whole fields, and biological alternatives (microbes, peptides, RNA-based tools, engineered self-defending plants) capable of replacing whole classes of synthetic chemicals.
• This is framed as a generational investment opportunity for YC. The narrator predicts adoption will be "explosive" rather than slow because a company that cuts pesticide use by 90% while increasing yields would be attractive in one of the world's largest markets, and closes with an open call for founders building in this space.

2026-04-27 · open the original ↗

• His framing is rotation rather than a single move: four destinations for trillion-dollar money, taken in sequence. The dollar, yen, euro and renminbi are the only places big money can go, and big money moves slowly — so the question is not which currency wins but in what order. He states his own view ahead of the model's: the first move is into the yen.
• The inversion is the diagnostic he keeps returning to. During the recent crisis the dollar did rise, but so did yields — which breaks the pattern that has held since 1945 and indicates the safe-haven function has degraded rather than performed. He is careful that this is a judgement call requiring human reading: the dollar moved up, but not nearly enough for the severity of the crisis.
• On the alternatives he relays the model's ranking, including where it disagrees with him. Gold is the structural winner across every phase, the one asset with no clean sell point, because physical demand from Asia, central banks and post-London Indian pricing now governs it more than real rates. Silver — which he names as one of his own favourites — is rated a later-phase instrument, because the market is small enough for institutional flows to distort it and the post-resolution pullback can be sharper than gold's. The Swiss franc is a two-to-three-month trade, not an anchor. And Norway's krone is the cleanest fiscal-moat currency. He flags the risk that a carry unwind forces leveraged gold longs to liquidate, since CTA positioning rests on the same funding trade.

2026-04-27 · open the original ↗

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

2026-04-27 · open the original ↗

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

2026-04-27 · open the original ↗

• Iran tops his list for "choking" world oil supply amid the ongoing war. He says Iran is restricting Strait of Hormuz access and disrupting Persian Gulf oil and gas output, driving Brent and WTI above $120 a barrel.
• The US's Middle East ally that struck Iran first ranks second, though he avoids naming it. He describes it as a small, battle-hardened country whose actions disproportionately shape US Middle East policy.
• Russia and Ukraine round out his list as opposing forces in an ongoing war. He calls Russia the world's second-largest oil supplier, third-largest arms supplier, and Iran's key arms source, while crediting Ukraine — against pre-2022 expectations that NATO-level support couldn't match Russia — with pushing Russian forces back and striking Russian energy infrastructure with drones and missiles.

2026-04-27 · open the original ↗

• He argues investors control only sizing, timing, strategy and underlying selection — not outcomes. Sosnoff says defaulting to advisors or gurus forfeits both control and learning, and that the "opening" of a trade — the initial decision — is where all the strength lies.
• Volatility, not dollar amount, is the real measure of risk, he contends. He calls volatility "an opportunity index, not a fear index" and says his firm has used volatility-based math to size opportunity for nearly 40 years, a contrarian view of what most investors treat as danger.
• Selling against a long position lifts win probability sharply, he claims. A simple long position has roughly 53-47 odds of a positive outcome, he says, but capping upside by selling calls against it can raise the probability of profit to 65-68%.
• He recommends a small allocation to high-risk, non-correlated assets for asymmetric upside. Examples include crypto, pre-IPO companies, futures, or a side business — positions he says can return three to ten times what an index fund delivers, while the rest of the portfolio stays in liquid holdings.