Souk's Notes - Unofficial Summaries

2026-01-10 · open the original ↗

• The index rebalance is a mechanical, pre-announced seller of exactly what performed, and it lands this week. Everyone can see the algorithms and everyone can see how the weights drifted, so the flows are calculable in advance: about $6.1 billion of silver and $5.6 billion of gold to be sold. A JP Morgan estimate puts silver's net selling at 10% of the value of all open derivative contracts, and the biggest of any commodity. COMEX silver notional is put at $23 to $25 billion in the source, computed off $50 to $55 prices — at current levels he reckons closer to $65 or $70 billion. Portfolio managers who know this is coming start early, which is why gold is down 0.6% and silver about 4% into it. His framing is that this is a genuine test: whether the metals can absorb a forced seller of that size and keep going.
• He spends most of the episode dismantling the viral bank-collapse claims, using filing rules rather than opinion. JP Morgan going under has quietly been replaced by Goldman Sachs, allegedly $376 billion in the hole on short silver — while trading at an all-time high. His disproof is procedural: anything material to the balance sheet requires an 8-K filed within four business days, filed electronically, hitting public screens the moment it reaches the SEC, downloadable by anyone. There is no such filing. The story also requires a board room of fifty people to have learned of an existential loss between 9:03 and 10:17 on a specified morning without the street finding out, which he does not believe for a moment. And the claimed short entry around $41 or $42 would date to September, meaning a firm that trades this every day sat still for months.
• His broader instruction is to learn the plumbing so the clickbait cannot land. The repo facility that supposedly proved a silver bailout was year-end balance sheet dressing, and it disappeared the moment the books were closed — which is precisely what the timing showed. He restates his own position without hedging: positive on the metals for three years, still positive, and yes they can correct. He passes along TDS reiterating a short call targeting $40 within three months, and says without irony that differing opinions are what make a market. The distinction he wants is between recognising stress, which is correct, and inventing a mechanism for it, which is not.

2026-01-10 · open the original ↗

• Enterprise agents were the year's biggest disappointment. For 2025, St. Lawrence's best-performing Substack post of the year, "Agents of Disappointment", argued the gap between hype and practical implementation was too great. She says vendors like Salesforce with Agentforce and SAP promoted agents inside existing enterprise tools, but they "just don't work" or companies are not set up and do not know how to structure them. The host agrees there is too much talk relative to impact, blames data silos and security setups, and expects deployments to come.
• Phone-based AI still adds little a year after launch, the host says. His disappointment pick is Apple: he says Apple Intelligence, announced in autumn of the prior year, still adds little he uses beyond once generating an emoji, and Siri should understand speech as well as OpenAI's Whisper. St. Lawrence contrasts the seamless voice chat she gets from Grok in her Tesla. Both see a big opportunity for Apple given how ubiquitous its devices are.

2026-01-10 · open the original ↗

• Reactions to the Venezuela operation split along who holds the presidency. He argues Republicans would be denouncing a war on Venezuela under a Biden presidency, and that Democrats now echo the same objection — though he says they'd voice stronger dissent under Biden without fear of political reprisal.
• Russia is unlikely to be emboldened toward further Ukrainian gains, he argues. He notes Russian forces are already only advancing a square mile or kilometer per month, and that China has been preparing for a possible Russian collapse.
• Japan says it would intervene if China moves on Taiwan. He expects the US to back Japan in that event, and notes China is already contending with natural disasters and economic turmoil, leaving it vulnerable to a sanctions escalation most of the world would likely observe.
• He opposed the Venezuela invasion but says decisive action became unavoidable. Once US personnel were deployed there, he argues, backing off at that point would have made the country look weak.

2026-01-10 · open the original ↗

• Venezuela's Russian- and Chinese-supplied air defenses were destroyed by US forces. He notes this is the third country relying heavily on Russian air defense systems to see them wiped out by American aircraft.
• Capturing Maduro sent a deterrence signal to Xi, Putin, Khamenei and Kim. He disputes claims that the operation emboldened China or Russia to seize Taiwan or more of Ukraine, arguing instead that it frightened them by showing Washington's willingness to act.

2026-01-09 · open the original ↗

• Raising capital is called an ask, not a pitch that sells itself. One host says the hardest part of fundraising is simply asking directly for a number, illustrating with two firms that lowballed early buyout offers for Thinkorswim and were later shut out — TD Ameritrade returned years later and paid nearly $1 billion for what it once dismissed.
• Hosts split on matching a competing job offer to retain staff. One says he'd do whatever it takes to keep a valued employee who brings an unsolicited outside offer; the other says he's never fought to retain anyone, arguing that once someone is actively interviewing elsewhere they've already checked out — congratulating them and letting them go instead.
• Short-premium capital allocation should scale with the VIX, not stay fixed. With opportunity and volatility highly correlated, one host says he runs roughly 40-60% allocation when the VIX is elevated but drops to the mid-20s when it's under 15, as it was this week — and says a caller's 40% allocation is fine for a smaller account but too high for a larger one.
• In-person meetings are called irreplaceable for closing deals. Hosts say every significant deal they've made in decades of business was closed in person, never remotely, describing their office as a deliberate "home court advantage" meant to control the environment and make it harder for a counterparty to say no.

2026-01-09 · open the original ↗

• Maduro's capture took under an hour, far faster than his first estimate. He reports the US suffered no loss of life or equipment hitting a target 1,800 miles away despite telegraphing its intentions beforehand, and notes Trump later released a video mocking Maduro over the operation.
• He contrasts this speed with Russia's years of threats to invade Ukraine. Russia, he notes, massed hundreds of thousands of troops on the border while repeatedly denying any intent to invade.

2026-01-09 · open the original ↗

• The US captured President Maduro in an operation using about 150 aircraft, completed in under 24 hours. Vice President Delcy Rodriguez fled to Russia but is reportedly (per unnamed sources) willing to cooperate with the US.
• The speaker frames the operation as a deliberate humiliation of Russia and a warning to China. He contrasts the swift Venezuela outcome with Russia's three-year, unresolved war in Ukraine, and argues it signals to China that a Taiwan invasion is comparatively more daunting than what the US just executed.
• His inference: alliances with Russia and China offer "zero protection" from the US. Both countries only issued condemnations without military assistance for Venezuela, which he says mirrors their response when Iran was attacked earlier in 2025.

2026-01-09 · open the original ↗

• NATO's $2.4 trillion in US debt holdings function as leverage against a forced US land grab, the speaker argues. He says a seizure of NATO-ally territory would fracture the alliance's cohesion and crash world trade, hurting even wealthy US interests that could otherwise hedge against economic damage.
• The speaker doubts Trump personally grasps this dynamic, suggesting he is being used as an "autopen" by people who hold real power and are getting the outcomes they want regardless of the execution's sloppiness.
• His broader view: the post-WWII order has been shifting since the late 1990s, only now becoming visible, with China as the focus. He argues the US still needs a strong, independent NATO to manage that shift.

2026-01-09 · open the original ↗

• The US is pressuring Greenland with unreasonable demands, in his view. He argues the goal is twofold: building US bases there and preventing Greenland — which wants autonomy or full independence from Denmark — from aligning with Chinese or Russian interests.
• NATO allies hold financial leverage that could deter US overreach, he argues. He notes NATO members invest hundreds of billions of dollars annually in the US and hold roughly $2.4 trillion in US debt, which he says gives them a "nuclear option" to damage the US economy if truly threatened.

2026-01-09 · open the original ↗

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

2026-01-09 · open the original ↗

• Bun began as one stranger's question in a Discord. One speaker recounts that a single person turned up asking about C interop because he wanted to build his own JavaScript engine; he later ran a Discord channel where typing "I want bun" got you an invite to the private GitHub repository, before publishing it. The other speaker takes this as a vote of confidence in Zig as a language and a valid C replacement.

2026-01-09 · open the original ↗

• Arctic traffic near Greenland is pushing the US and NATO toward fortification. He says adversarial trading and military vessels are increasingly moving along the North American Arctic border, with Greenland sitting in the middle of that route.
• He again frames the demands on Greenland as a pretext for fortification. The stated aim, he says, is to install US bases and ensure Greenland — which seeks autonomy from Denmark — doesn't align with Chinese or Russian interests.

2026-01-09 · open the original ↗

• The US wants to annex Greenland mainly to block Arctic shipping routes. He says the main driver is the emerging trans-Arctic shipping lanes that Russia and China are seeking to claim, since the route cuts shipping time compared to the traditional Suez route.
• Greenland's rare earth minerals are a secondary driver for US military tech. He describes a map showing Greenland's long Arctic border alongside Russia, Siberia and Scandinavian countries to illustrate the region's strategic position.

2026-01-09 · open the original ↗

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

2026-01-09 · open the original ↗

• The hard part of agents is the last mile, not the prototype. Carlson says LLMs and vibe coding have compressed the path from idea to v1, but rolling out to thousands or hundreds of thousands of enterprise users is "where the rubber hits the road". The episode is a partnership with Salesforce announcing that its Agentforce 360 platform is open for building on. Carlson says it bundles a multi-LLM planner, a scripting language for deterministic control, Salesforce's data and metadata model as business context, the security and permissions layer, and observability and monitoring.
• Natural-language rules fail too often, so agents need deterministic rails. Carlson says customers wrote agent logic as "always do this, never do this", which LLMs handle poorly, and an enterprise cannot accept a task done 90% of the time and skipped the other 10%. Agentforce Script lets developers mark where the agent may reason freely and where it must follow pre-set paths, much of which already exists in Sales, Service and Marketing apps.
• Small teams ship faster on a platform than on hyperscalers. Carlson says AppExchange, launched 20 years ago as the first enterprise SaaS app store, reaches 150,000-plus Salesforce customers, with low-code builders and pro-code options in Apex and React. Teams that want to run infrastructure down to the server should use AWS or Google, he says, but a founder can ship faster with the app, workflow and agent layers provided. His example is a family ranching business that built a cattle-management app for its own herd and then sold it to others.
• Founders over-focus on the best LLM instead of the customer's job. Carlson calls LLM performance a distraction: commercializing AI is like building any product, with requirements and jobs to be done. Not every step needs an LLM when code, database structure and standard permissions already solve it. He sees many niche problems ripe for someone to build something.
• Agents will not replace applications; they are applications. Carlson argues any sophisticated agent at scale is a multi-LLM orchestrated application, with logic choosing which LLM or deterministic step to use, where data lives and what the permission model is. He expects applications to look very different but rejects "agents versus applications", predicting the two merge.

2026-01-09 · open the original ↗

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

2026-01-09 · open the original ↗

• Mellon expects the yen, which he calls severely undervalued, to rise as much as 20% this year. He acknowledges he has made this call before without it yet playing out.
• Japan's new fiscal stimulus push is driving inflation and pushing long bond yields above 3%. He likens the prime minister's "pump priming" to "blowing air into a dead horse," but notes it has still lifted long-dated JGB yields well above levels of five years ago.
• He isn't worried about a Japanese debt crisis because the Bank of Japan is tightening and most bonds are held domestically. He adds that the BOJ is already running quantitative tightening to offset the stimulus, with little foreign ownership of JGBs.
• A yen turn would trigger fast appreciation given the scale of Japanese savings invested overseas, especially in the US. He notes that if Japan's yield curve flattens and short-term yen deposits start paying 1–3%, holding yen becomes more attractive than holding pounds, euros or dollars.

2026-01-08 · open the original ↗

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

2026-01-08 · open the original ↗

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

2026-01-08 · open the original ↗

• Once a year the commodity indices reset their weights, and this is the week it happens. The BCOM launched in 1998, holds 23 to 24 commodities weighted two-thirds by liquidity and one-third by world production, capped at 15% for any single name. The S&P GSCI, launched in 1991 and bought by S&P in 2007, weights purely by five-year average world production, which makes it 55 to 60% energy. Both reset targets in January and phase the roll over four to five business days beginning 8 January, with most trading on the 9th and 10th because everyone knows it is coming. His practical framing: if you believe in oil, the biggest trade in the world, you buy the GSCI; if you want balance, the BCOM.
• The arithmetic of rebalancing makes last year's winners this week's sellers, and silver is the extreme case. Silver rose about 170% and gold about 65% in 2025, so their realised weights have drifted far above target. In the BCOM, silver's realised weight of about 9% resets to just under 4% — implying over $5 billion of futures selling across five days. Corn, soybeans and soymeal fall about 21%, and crude is cut to 6.6%, its lowest ever BCOM weight, alongside trimmed natural gas. The GSCI, weighting by production, goes the other way on gold — miners produced more because the price rose, so gold gets one of the largest increases in the index — while buying copper, aluminium, corn, cattle, hogs and natural gas, and cutting oil, gas, Chicago wheat, soybeans and cocoa. The two indices disagree outright on cocoa and Brent, one buying what the other sells.
• What he takes from the buy list is a signal about next year's inflation. The commodities being bought are the ones that fell — grains, livestock, energy, cocoa — and he notes why several of them fell: American farmers are ploughing crops back in because China stopped buying and the prices no longer cover the cost of selling. Mechanical index buying now arrives underneath depressed food and energy prices, which he reads as unhelpful for 2026 inflation. He also flags the structural point running underneath the whole exercise: these funds do not own commodities, they own futures contracts, and their returns include roll yield as well as price. Given the paper-versus-physical divergence he has been documenting, the question he leaves open is whether the indices themselves eventually have to change.

2026-01-08 · open the original ↗

• A clip of Senator Lindsey Graham declaring "Cuba's days are numbered" frames the speaker's argument that Cuba is next. He calls it a "two birds with one stone" opportunity following the Venezuela operation, given the countries' long-standing ties.
• Cuba–Venezuela relations trace to 1902, but deepened after Castro turned to the USSR in 1960 following the failed 1961 Bay of Pigs invasion (roughly 1,000 US troops), and closed further when Hugo Chavez, who took power in Venezuela in 1999, treated Castro as a mentor.
• Venezuela came to trade oil for Cuban doctors, teachers, and military advisers, especially after professionals fled Venezuela's economic turmoil. The speaker says a 2002 failed US-orchestrated coup against Chavez pushed the two countries closer still.
• The speaker argues Cuba is now uniquely vulnerable to internal collapse. US pressure cut Mexico's oil shipments to Cuba from about 22,000 to 7,000 barrels a day, and he names Secretary of State Marco Rubio — whose parents were born in Cuba — as having long sought Cuban regime change.

2026-01-08 · open the original ↗

• Stoke Space is targeting full rocket reusability, including the second stage that every other reusable rocket still throws away. A co-founder explains that industry has already made first-stage reuse routine, scaling launches from roughly 10-20 to about 150 a year, but the upper stage still burns up structurally around 2,700°F re-entering at 17,000 mph; Stoke's Andromeda capsule survives this with a heat shield cooled by liquid hydrogen and 24 small thrusters for landing.
• The founders quit stable jobs at Blue Origin in 2019 to found the company on real personal risk. One co-founder describes it as "wildly irresponsible" and his hardest personal decision, having a 3-month-old at home; they set a 6-month deadline to find traction, got their first funding check right around that point, just as COVID hit, and built their first prototype engine in a shipping container in a backyard.
• YC's network was central to raising capital in a market skewed toward SaaS. The founders say they wildly underestimated how hard fundraising would be as first-time hardware founders with no investor connections, and credit YC with teaching them the fundraising process and adding momentum to the round; the company has raised about $990 million to date.
• Fast iteration is treated as the core competitive advantage, enabled by in-house manufacturing and custom software. By building nearly every part themselves rather than depending on outside suppliers, a co-founder says they cut a test-fail-rebuild cycle from about a month down to a day or two; the company also built its own operations software, called Bolt Line, to track part history and maintenance as it scales toward its first orbital launch, planned for later this year from a new site at Cape Canaveral.

2026-01-08 · open the original ↗

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

2026-01-08 · open the original ↗

• Demand for AI engineering skills has outgrown data science skills. St. Lawrence says her 2025 prediction to that effect held, citing a LinkedIn jobs report showing much higher growth in AI prompting and AI engineering skills than in data science. She adds that AI is being stamped onto job descriptions much as CEOs stamp GenAI onto products. The host points to his episode 856 naming AI engineer the fastest-growing job in most developed countries.
• Data science spawned a family of roles now being rebranded as AI. The host sees AI engineering as a subspecialization of data science. St. Lawrence says the 2014-era "unicorn" data scientist split into roles such as data engineering, ML engineering and MLOps, and is now rebranding from ML to AI. She notes the irony, from the book Thinking Machines, that "machine learning" was adopted in the 2010s to escape the unserious-sounding term AI.

2026-01-08 · open the original ↗

• Data-center buildout for AI will require roughly 20% more US electricity supply, yet energy stocks remain unloved. Mellon says this demand case has been overlooked despite talk of oversupply and the broader shift toward electrification.
• He is "loading up" on oil and gas as the next unloved sector, following his earlier bets on gold and silver. He cites dividend yields and the sector's low weighting in the S&P and other major indices as reasons it's underowned.
• He reads the market's reaction to the Venezuela situation as confirming oil and gas stocks are positioned to benefit from geopolitical instability. He notes any resulting increase in Venezuelan production would take years, but the episode highlights an unstable environment that could support the sector through the year.

2026-01-08 · open the original ↗

• Unusual US aircraft movements toward the Middle East echo the pattern seen before the 12-day war, the speaker says. He and a fellow creator he names "War Monitor" suggest a new intervention in Iran could be imminent, coinciding with the eighth day of intensifying Iranian protests and reports of police and military members turning against authorities.
• Unverified reports claim Iran's supreme leader plans to flee to Moscow with up to 20 aides and family if security forces called on to suppress the unrest begin deserting.
• The speaker predicts the US could strike Iran's military and suppression infrastructure again, potentially producing two regime changes — Venezuela and Iran — within the same week, which he frames as a hypothetical rather than a certainty.

2026-01-08 · open the original ↗

• A file explorer can ship as one tiny, runtime-free executable. The interviewer's own pre-interview analysis found a 2.3 MB single executable with 14 DLL dependencies and 312 imported functions; the developer says even 312 is high, needed only because a file explorer is tightly tied to the OS for files and thumbnails. The developer does not link the C runtime at all, having written the standard-library pieces from scratch. The clip matches Hanselman's interview with File Pilot developer Vjekoslav Krajačić.

2026-01-07 · open the original ↗

• The speaker rejects the idea that the US would let Russia take Ukraine or China take Taiwan in exchange for a free hand in the Western Hemisphere. He argues this is illogical given Taiwan's dominance in semiconductors the US military depends on.
• He cites Russian losses of over 1.2 million troops, 11,500 tanks, and 100,000+ drones in nearly four years without controlling Ukraine, and notes Europe fields about 1.8 million troops and is providing over $160 billion in 2026 military aid to Ukraine, arguing full Russian conquest is implausible.
• He argues China lacks proven amphibious and joint-operations capability against a heavily fortified Taiwan, predicting Japanese and US intervention would follow any attempt; he separately speculates China may instead absorb Russian territory through economic coercion over the next 10–15 years.

2026-01-07 · open the original ↗

• The delivery rate is the number he thinks people should be watching, and it has moved. COMEX historically settles 1 to 2% of gold contracts physically; silver runs slightly higher at around 2.5% in demand spikes and rarely exceeds 8% even in tight markets. In December, 12,800 contracts settled physically out of roughly 150,000 — 8.1% — and it has stayed near that level since. He also puts the exchange in historical context: COMEX only became important after 1971, launching gold futures in 1974 once Nixon ended convertibility, replacing a system where gold was fixed by statute and silver followed the 15-to-1 bimetallic ratio set in the 1792 Coinage Act. Its settlement prices are watched worldwide, but spot has always been shaped elsewhere — London, Shanghai, ETF flows.
• His analogy for what a divergence actually means is retail gasoline, and it reframes the question. Physical prices differ by location for perfectly ordinary reasons — minting, transport, margins, local scarcity — the way upstate New York differs from California. But you do not trade twenty-five gasoline prices; you trade one paper price with physical prices attached to places. What he wants to know is whether silver is now developing a permanent disconnect, with the physical market rather than the paper market setting the direction. The pressure is all one way: hedge funds buying, corporations buying, defence contracts, governments buying, governments banning export — the US among them. He explicitly declines the suppression theory while doing it: people try things, but nobody is that good, and it does not work that way.
• The gold reserve accounting contains a distortion most people never see. Foreign central banks hold roughly 24% of reserves in gold and 23% in Treasuries, and they carry the metal at historical cost. US law prices its 8,133 tons at $42.22 an ounce — about $11 billion on the books rather than the trillions it would fetch. The comparison he draws is China at $150 billion against Germany at $42 billion despite far smaller holdings, which tells you about purchase timing rather than tonnage. The open question is whether central banks return toward the gold share they held in the 1970s, and what the price looks like if they do. His larger frame is a hybrid world: nobody is happy holding dollars and there is nowhere else to go, so instead of a single successor you get zones — the hemisphere on dollars, China on renminbi, Europe on a euro perhaps backed by gold — with far more bilateral, single-commodity trade than people realise underneath.

2026-01-07 · open the original ↗

• The speaker concedes the Venezuela strike violated UN Charter Article 2, but argues this fits a century-long US pattern. He lists post-WWII precedents including the 1949 Syrian coup, the 1953 Iranian coup, the failed 1961 Bay of Pigs invasion, the Vietnam War, and the 1989 Panama invasion.
• Despite this history, he argues the US retains dominant "soft power" by Joseph Nye's framework. He cites Hollywood holding 55–66% of global box office, US music at roughly 40% global share, and a #1 ranking in Brand Finance's soft power index.
• His assessment: the Maduro operation is a minor addition to this pattern, trading limited temporary soft power for a hard-power boost. He notes a supposedly prepared Venezuelan government and military collapsed in about 30 minutes.

2026-01-07 · open the original ↗

• Anyone can give a talk, but it needs a credibility hook. Hanselman argues everyone has a valid perspective and could talk on, say, their Python journey without being an authority; Mark Russinovich, CTO of Azure, pushes back that audiences want perspectives from people they have reason to trust. They settle on needing a hook: an employer, a title, long experience, or being new. Mark's first big "soft talk", on Azure strategy around 2015, was five slides he winged for 75 minutes.
• Set expectations up front when a talk is deliberately silly. Their Ignite talk, a 50-year history from the Altair to Azure cut to 45 minutes from Hanselman's hour-plus Portugal version, worried Mark up to the night before as having almost no educational value; he now admits he was wrong. They opened by telling the audience to lower its expectations, so nobody judged it against a bar it wasn't trying to hit.
• Azure's largest VM became a Task Manager pixel display. Mark draws games (Pong, Tetris, Breakout, Space Invaders) by running a thread per processor and varying its idle time per half second to set each cell's shade, on a VM with 24 TB of RAM and 16 sockets normally sold for SAP. Its processor count is captioned as both 792 and 1,792; the roughly 42 by 42 grid they derive fits 1,792. On Linux they had to vibe code their own Task Manager.
• The finale was a deliberate Rube Goldberg chain of Azure parts. A vibe-coded Python app (40–50 prompts) read the Altair's front-panel lights by camera, calibrated through a JSON file and thrown off by stage lighting, sent them over UDP, relayed them as websockets through a tunnel, and drew them on the big VM's Task Manager at about 5 frames per second. Zork, newly open sourced by Microsoft, ran on the Altair and in Azure Cloud Shell, with 100 GPT-4o instances playing it as a swarm.
• Losing presenter view broke the talk's tempo. Ignite's machine switcher left them unable to see the next slide of about 30, so they couldn't tell where a section ended; they fell behind halfway, one passed the other a "hurry up" note, and the final segment ran faster than expected because it all worked. Hanselman's lesson: seeing just the next slide would have been enough.

2026-01-07 · open the original ↗

• The author describes spending the 2010s on "cultural cruise control" while focused entirely on a conventional legal and Wall Street career. He attended law school (2010–2013), worked Wall Street trading floors, and was promoted repeatedly, but says he consumed only surface-level culture (Breaking Bad, Kendrick Lamar) without engaging deeply with literature, film, or the rise of social platforms like YouTube and TikTok.
• Quitting his job in 2021 to care for his newborn son during the pandemic prompted the reassessment. The slower pace of caregiving — pushing a stroller through Manhattan while others worked in offices — created space to ask whether life had to remain defined by career advancement alone, a question he says he'd never made time to ask.
• He describes deliberately rebuilding cultural literacy as a discipline, not passive catch-up. This included watching roughly 800 Casey Neistat vlogs to study filmmaking craft, buying a dedicated camera, and reading more widely (The New Yorker, Werner Herzog), framing the effort as necessary groundwork for his later pursuits in writing and video.
• Looking into 2026, he commits to long-form, slower work over viral short-form content, despite proof the latter works. A New Year's Eve thread of his went viral (1.3 million views) after he shared an authentic moment rather than commentary on a news cycle; he argues short-form video rarely produces anything memorable years later, and states his intent to keep chasing depth over virality.

2026-01-07 · open the original ↗

• The speaker argues deep US intelligence penetration, not just military strength, makes WWIII unlikely. He cites CIA claims of tracking Maduro's daily routine and exact location, and a leaked 2024 memo claiming US and UK forces knew more about Russia's Ukraine offensive than Moscow itself.
• He claims Iran's government and military were so heavily infiltrated that 20–30% of its top leadership was eliminated on the first day of an attack, and speculates the US or CIA may be behind China's recent internal military corruption purges, though this is not confirmed.
• His inference: US satellite dominance and suspected disinformation feeds erode rival trust in their own information, discouraging war. He states the US has more satellites than China, Russia, and the EU combined, and believes Putin was fed bad intelligence that led him to underestimate the difficulty of invading Ukraine.

2026-01-07 · open the original ↗

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

2026-01-07 · open the original ↗

• Agentic AI dominated 2025 so thoroughly that audiences are tired of it. St. Lawrence counts her prediction that agentic AI would be 2025's dominant trend as correct. Her evidence is more vibes than data: conference audiences no longer want to hear about agentic AI and want something fresh in 2026.
• The host argues agents keep getting more useful on a doubling curve. He says the length of a human task an agentic system can do doubles every seven months, at about 50% reliability on software or ML tasks that take a human expert several hours. On that curve he expects a full eight-hour workday by the end of 2026, which the speakers note is when talk of replacing workers heats up.

2026-01-07 · open the original ↗

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

2026-01-07 · open the original ↗

• The speaker revises his own earlier estimate: the Maduro capture reportedly took under one hour, not under 24. He says it was done with no American casualties or equipment losses, despite Venezuela having roughly 1,800 miles of distance and ample time to prepare, and having Russian and Chinese-supplied defenses destroyed.
• He reads the operation as a deliberate signal to Xi Jinping, Putin, Iran's supreme leader, and Kim Jong-un that the US "could do the same" to them, arguing it is more likely to deter than embolden Russia's advance in Ukraine or a Chinese move on Taiwan — noting Japan has said it would intervene if China attacked Taiwan.
• The speaker states he personally opposed the Venezuela invasion, but argues once the US publicly committed forces, backing down was not politically viable without appearing weak.

2026-01-07 · open the original ↗

• The US captured Maduro and his wife in an operation using about 150 aircraft, completed in under 24 hours. Vice President Delcy Rodriguez fled to Russia but reportedly may be willing to cooperate with the US.
• The speaker frames Venezuela as the "third domino" against a Russia-China-Iran axis, following Syria (which he says served as Iran's supply corridor and hosted a Russian naval base) and Iran's 12-day war, arguing the axis is unraveling.
• He notes oil markets showed little price movement despite Venezuela's large reserves, suggesting the outcome was already priced in, and raises open questions about succession — Rodriguez, opposition figure Machado, or a US-backed transitional government — and the risk of pro-Maduro resistance.

2026-01-06 · open the original ↗

• Chasing a stock that already tripled to $70 isn't worth it. Asked whether he'd buy a chip stock whose price jumped from $20 to $70 in two days, he says he "wouldn't buy it in a hundred years" and would instead look to get short, since such moves typically reverse.
• He's more worried about short sellers than chasers, citing GameStop's $5-to-$500 run. Because a stock that jumps from 20 to 70 could keep going to 100, 200, or 300, he says the real danger sits with anyone short it without a plan, not with longs chasing the move.
• His concrete advice: use real or mental stop-loss levels. Without one, he warns, these stocks "can break really really quickly" and traders get caught not knowing what happened, whether trading the move long or short.

2026-01-06 · open the original ↗

• US pressure on Greenland is aimed at countering Russian and Chinese Arctic ambitions, the speaker argues. He points to the Northern Sea Route as a faster alternative to the Suez Canal that Russia and China are pursuing, and to Greenland's rare-earth minerals needed for US military technology.
• He reiterates that a forced land grab from NATO ally Denmark risks NATO weaponizing its roughly $2.4 trillion in US debt holdings, fracturing the alliance and crashing world trade.
• His inference: Trump may not fully grasp this dynamic but is being used by more strategically aware interests. He frames the pressure as part of a broader shift in the post-WWII order, underway since the late 1990s and now centered on countering China.

2026-01-06 · open the original ↗

• He says up front that he is unsurprised by Venezuela and that this is not approval. The claim that runs through everything is that the old world is dead and the new one is being built differently — the old order was political, allowing everyone to trade with everyone under Pax Americana, and the new one is economic, defined by two words: secure and control. Countries will make decisions on their ability to secure and control energy, food, technology, demographics, and whether they are easy to attack. The last factor drew years of pushback, and his answer now is to ask Venezuela, then Colombia, then Guyana.
• Trump is an accelerator rather than a cause, and the distinction carries the argument. His original framing was fifteen to twenty years, and he does not know who will be president or what any country will be doing at the end of it — only that existing powers will co-opt the resources they need. Trump is not waiting: Venezuela taken, Greenland drafted (a document offering money and local self-government while the US runs everything, delivered to Greenland directly rather than through Copenhagen), Panama likely next by taking the canal back from its Chinese owner, with Mexico and Canada already threatened. The underlying admission is in the published military strategy: out of Europe, out of Africa, leaving Asia. Since the US cannot control the world and is pulling back, the question becomes where to pull back to, and the answer is the hemisphere.
• The symmetrical case is eastern Russia, and it is the one people miss. Greenland offers the Arctic, rare earths, farmland, oil and gas; eastern Russia offers farmland, rare earths, water, oil and gas. Chinese maps still show the territory taken in 1858, 1900 and after the Second World War as Chinese, with the old Chinese city names. And the demographics have already turned — there are reportedly more Chinese living in eastern Russia than Russians, with nobody there to stop the crossings. He treats it as a matter of time. He also cautions against reading any single week's news as direction: South Korea, Japan and Australia just formed a defence pact while South Korea's president is in Beijing seeking fully normalised relations with Xi. His closing judgement is not triumphal — America cannot go it alone and will need allies, but the eighty-year arrangement with rights and obligations is over and everything is transactional now: be with us or we will not defend you, will not help you, and will hurt you if we can.

2026-01-06 · open the original ↗

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

2026-01-06 · open the original ↗

• Protests erupted against the government and IRGC across several Iranian towns. In Assadabad, protesters set fire to an IRGC-linked base; in Isfahan — heavily bombed during the 12-day war — crowds marched toward the central square chanting against Ayatollah Khamenei; in Fasa, protesters stormed a government building while being fired upon.
• He reads the unrest as an extension of the 12-day war. He says it is unclear whether the government or the protesters currently have the upper hand, and argues this is exactly what Iran's adversaries planned, expecting them to exploit the instability further.
• Iran's economy has deteriorated sharply since the 12-day war, he reports. He cites the rial dropping 40–50% in value to roughly 1.44 million per US dollar, youth unemployment reaching as high as 50%, and oil — Iran's key export — trading below $60 a barrel for an extended period.

2026-01-06 · open the original ↗

• Reports describe protesters attacking an IRGC-linked base in Asadabad and storming a government building in Fasa despite gunfire. In Isfahan, crowds reportedly chanted for the fall of the "butcher" regime; the speaker says the government's response so far has been comparatively restrained relative to its historical capacity for suppression.
• The speaker's inference: the 12-day war left Iran's regime weaker against unrest than in 2009, 2017, or 2019. He cites a 40–50% currency drop (roughly 1.44 million rial per dollar), youth unemployment near 50%, and destroyed suppression infrastructure and personnel as reasons this round of protest is different.
• His prediction: authorities may be gathering resources for a rapid crackdown once protests are dispersed, but he suggests any counterassault also risks being undermined by the same infiltration reportedly exploited during the 12-day war.

2026-01-06 · open the original ↗

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

2026-01-06 · open the original ↗

• Recall is browser history for everything on your screen. Rey describes the Windows 11 preview feature as periodic screenshots across all applications, roughly every 20–30 seconds by an algorithm he cannot explain, with no audio; it needs a Copilot+ PC with at least 40 TOPS. He first saw it at Ignite 2024 in Chicago while on the Surface team, got access in spring 2025 and has run it since summer 2025.
• Users set storage, retention and exclusions up front. He believes the default storage is 25 GB (he set 150 GB on a 300 GB drive) and the default retention 90 days. It skips content it deems sensitive, and an excluded app such as a confidential Excel workbook blocks capture of its whole monitor; websites can be excluded, capture paused until tomorrow, and single snapshots or all of an app's snapshots deleted.
• Microsoft answered privacy concerns with local, encrypted, opt-in storage. Rey says snapshots stay encrypted on the device behind Windows security sign-in, the feature is off unless you opt in, and his employer can disallow it by default or choose to pilot it.
• OCR makes the screenshots searchable, unlike a folder of your own. He searched his history for "David" and found matching snapshots, and used the timeline to recover code Jeffrey Snover, the inventor of PowerShell, briefly shared in a 31 December meeting he had not screenshotted; snapshots also link back to the saved file.
• Recall's future is uncertain until its value outweighs comfort concerns. Rey compares it to browser history, which people once distrusted and now accept; Giard adds his own view that society chose convenience over privacy long ago.

2026-01-06 · open the original ↗

• Last year's five predictions mostly held, some only weakly. Of St. Lawrence's 2025 calls, agentic AI dominated, to the point that conference audiences are tired of it. AI spread into devices, from AirPods' real-time translation to CES refrigerators and a Costco toothbrush, though she doubts much of it was wise. AI-driven science was mixed: a paper claiming large gains in materials discovery was retracted, while she credits DeepMind's GNoME project with compressing about 800 years of discoveries. Enterprise monetization stayed crucial, with many projects stuck at MVP, and a LinkedIn jobs report showed AI engineering skills outgrowing data science.
• Intelligence is getting radically cheaper while agents' task horizons double. The host says the same capability costs about 100 times less than a year earlier, and the task length agents can handle at about 50% reliability doubles every 7 months, implying a full 8-hour workday by end of 2026. He limits the curve to software and ML tasks, where success is verifiable and training data can be simulated. He adds that higher-reliability curves, at 90% and 99%, trail behind, and discloses his bias as an AI consultant. St. Lawrence says this forces businesses to rethink their models, pointing to Cursor's revenue growth.
• Their awards crowned an established search giant, a chipmaker and "vibes". For 2025, both name Google comeback of the year again: St. Lawrence for its seamless integration into Search and G Suite, the host because users already trust it with their Drive and Gmail data. Her wow moment was Nano Banana redoing a room with her tile sample; his was co-presenter Ed Donner's crew of four agents building a working simulated trading platform during a 90-minute lunch break at ODSC East. For disappointment she names enterprise agents and he names Apple; the overall winners are Nvidia and the theme of vibes, with GPT-5 a letdown.
• Next progress will come from specialization, continual learning and research. For 2026, St. Lawrence predicts more AlphaFold-style domain models evaluated by experts, because training data for general models has hit a limit. She also expects progress on continual and nested learning, citing a new Google paper. And she expects a return to research, since no frontier lab has a clear path and nobody talks about GPT-6. The host agrees an orthogonal breakthrough beyond scaling weights and inference time is needed, especially far more sample-efficient learning, but doubts it arrives in 2026.
• Spatial intelligence and AI ops will rise, but robots come later. St. Lawrence holds that 2027 is "the year of the robot" and expects physical AI and spatial intelligence, such as World Labs' simulated environments, to matter mainly as new, harder training data, even as she awaits her Neo robot in 2026. Her final prediction is AI ops as a hot new job managing GPUs, model orchestration and agent reliability, akin to DevOps around 2010.

2026-01-06 · open the original ↗

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

2026-01-06 · open the original ↗

• China's backdoor into the US market via Canada and Mexico is collapsing. Canada has become a "no-go zone," he says, tariffing Chinese EVs around 100% and Chinese steel and aluminum 25%, which pushed China to rely on Mexico instead.
• Mexico is now imposing tariffs of up to 35% on China. Citing a Raw Readers article, he says the tariffs extend to any country without a Mexican free-trade agreement and cover thousands of products — including automobiles, auto parts, textiles, clothing, plastics and steel — leaving China without a cheap route into the US market.

2026-01-06 · open the original ↗

• Mexico is imposing tariffs of up to 35% on China, covering autos, auto parts, textiles, plastics, and steel, closing off Mexico as China's last major backdoor into the US market after Canada tariffed Chinese EVs 100% and steel/aluminum 25%.
• The speaker attributes Mexican President Claudia Sheinbaum's compliance to Mexico's dependence on the US and the USMCA trade framework, noting 85% of Mexican exports go to the US even though China supplies about 21% of Mexican imports.
• His assessment: this shrinks the US trade deficit with China and the world, but at a domestic cost. He cites over 1 million 2025 layoffs and a reported 2.7% inflation figure he expects to be revised upward once shutdown-delayed data is incorporated.

2026-01-05 · open the original ↗

• Venezuela's small economic footprint explains the market's shrug. Outside of oil, one host says Venezuela is a minor player in global GDP, with no risk of other countries being drawn in, so a change at the top — its president is called "a crook and a criminal" — barely registers for markets.
• The real worry is precedent: this could greenlight China moving on Taiwan. Hosts argue that intervention against Venezuela's leadership sets a dangerous precedent that erodes prior boundaries on global intervention, and that a comparable move toward Taiwan would be "significantly different and much worse" for markets than anything happening in Venezuela.

2026-01-05 · open the original ↗

• Clarity over certainty is the whole frame: what backs a currency is a government whose purpose you can read. Of four currency blocs he sets the renminbi aside as roughly 30% undervalued with Beijing showing no interest in changing that, and the dollar as the one everyone is already loaded on and now trimming for risk — particularly central banks, given the Miran proposals to tax holdings and force conversion into hundred-year zero-coupon paper. That leaves the euro and the yen, and the difference between them is not economics but legibility.
• Japan's direction is stated plainly, and it is the five factors under another name. Takaichi wants responsible, proactive fiscal policy — leaving behind a strong competitive economy rather than merely sound government finances — with the supplementary budget as a down payment on energy security, rare earth minerals, healthcare, cybersecurity, disaster mitigation and defence. She rejects tax cuts for their own sake and wants a smaller government while spending on capability. Meanwhile the BOJ has raised to 75 basis points, the highest in thirty years, with Ueda signalling more on 5 January and willingness to go toward 1.75%, and the government saying openly that it expects the central bank to operate in close coordination and wants 2% inflation delivered through wages rather than cost-push. Spending more, raising rates, fighting inflation, deeply indebted — which he notes sounds exactly like the US, and calls the future.
• Europe's problem is that nobody can say what it intends, and he expects that to cost the euro. The EU has no taxing power and no legislative power beyond regulation, and it now faces economic, geopolitical, military and tariff decisions it cannot agree on. His position is stated without hedging: the EU will not make it — and that is the good news, because the old has to be destroyed for the new to be built, and the United States of Europe now being discussed is what eventually emerges. But it falls apart first and reconstitutes afterwards, and the euro gets hit hard in between. So where does money holding 20% of world currency go? Some to the dollar, probably not to the renminbi — Hotel California — and his call is the yen. Which loops back to the trade: when JGB yields rise far enough, Japanese institutions are forced to sell US securities, bring the money home, buy yen, and repay against the JGBs. He also opens by noting there were no repos over the weekend, as he predicted, and asks the audience to help him plan a first-ever livestream.

2026-01-05 · open the original ↗

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

2026-01-05 · open the original ↗

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

2026-01-05 · open the original ↗

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

2026-01-05 · open the original ↗

• Every country will soon have a sovereign AI strategy. Roese's fifth prediction, for 2026 going into 2027, builds on three models he described the previous year: government building data centers for government, government building for its indigenous industries, and government convening with industry. He calls the acceleration dramatic and sovereign AI now mainstream.
• Sovereign infrastructure will serve far more than government work. Roese's examples are resilient backups of a country's critical-infrastructure AI, and agents certified by a government, such as a "Swiss lawyer" agent, whose knowledge graphs and back-end data live physically under that sovereignty. He expects government's role to grow where AI meets national interests such as critical infrastructure, defense, intelligence and healthcare, and calls the inseparability of government and private sector in AI probably a good thing for pooling resources.

2026-01-04 · open the original ↗

• He opens with a story from his own career that explains the entire year-end repo pattern. When his retail firm merged with an institutional one, the retail side was forced to move its fiscal year-end from 31 December to 30 November — an enormously disruptive change to how advisers get paid. When he asked the CFO why, the answer was that this is how the world works: financial firms close their books on 30 November precisely so that they can expand their balance sheets afterwards to fund every other corporation closing on 31 December. Before the financial crisis, firms running 60 to 100 times leverage would blow the balance sheet up three, four, five hundred times across the turn — and not report for another ninety days, by which point it had come back down. The Fed stopped that practice afterwards, but the seasonal funding need never went away.
• The leverage arithmetic at JP Morgan is the reference point for what "conservative" now means. About $4.2 trillion of assets against roughly $4.55 trillion of total exposure, an SLR of 5.8%, a tier one leverage ratio of 7% — which works out to roughly 17 times: every dollar of capital supporting seventeen dollars of investment, down from twenty under Basel III. And the capital is fully deployed, because deployed capital is what generates revenue.
• Which is why the December repo surge is issuance plumbing rather than distress, and he gives the disqualifying detail. Dealers must fund their allocated share of every new Treasury auction before distributing the paper, and when SOFR runs above the rate the Fed charges, the standing facility is simply the cheaper source. The Fed spent 2025 actively asking banks to use the window to reduce the stigma, and the banks resisted precisely because visible borrowing gets read as distress — which is exactly how it is now being read. His disqualifying point is about duration: a repo is one day, maybe three, Friday to Monday. If you had a genuine problem in silver or anything else, an overnight facility is useless. You would need capital, and a repo is not capital — it is a stopgap.

2026-01-04 · open the original ↗

• An agent is a software system, not a feature of an LLM. Roese says one camp treats agentic behavior as a property of the large language model, which he calls not very useful. Dell's camp, which he says many others share, treats an agent as an autonomous software system of which the LLM is one part.
• The LLM supplies world knowledge, communication and much of the reasoning. Roese counts it as the first of four components, large or small, without which agents cannot be built.
• Knowledge graphs supply specialized knowledge and long-term memory. Roese says agentic RAG or, more often, knowledge graphs give agents domain information they can manipulate. Pathways through the graph become skills for repeating tasks and hold state over long periods outside the LLM.
• A tool protocol lets agents perceive external data and act. Roese calls MCP, or a similar interface, the protocol of choice for this third component.
• Agent-to-agent protocols matter most, because ensembles can reach consensus. Roese calls this fourth component, chiefly A2A, the most important. In Dell's studies, he says, agents sharing a common knowledge graph as "truth" (the LLM being "opinion") and told consensus matters ignored a hallucinating member, like a team overruling one person saying crazy things. He predicts this four-part definition will prove enduring, against those who treat the LLM as a "black box of magic".

2026-01-04 · open the original ↗

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

2026-01-03 · open the original ↗

• The margin increases explain most of the paper price action, and they are not subtle. COMEX raised silver margin requirements from around 5.25% to 6.5%, then two days later to 30%; China went from roughly 10% to 20%. A 30% requirement removes demand for paper silver by construction, and forces liquidation from anyone already positioned. Meanwhile Tokyo trades physical at $130 an ounce against $71 on the paper market — a dislocation that has persisted far longer than backwardation normally does, which is the part he treats as significant rather than the level itself.
• He takes the JP Morgan bankruptcy claim seriously enough to test it, and it fails on several fronts. The rumour traces to an AI-generated site with a history of fabricating sensational financial news, linked to an alleged $34 billion overnight liquidation via the Fed repo. But the stock is at an all-time high, and if the street believed the largest bank were in trouble it would be down hard and everyone would be piling on — because, as he puts it from experience, the universal strategy when a rival is known to be in trouble is to trade against them, which is why firms guard their positions. On scale: $895 billion market cap, 2.75 billion shares, $4.55 trillion of assets, and over $16 billion paid out annually in dividends alone. A $4.8 billion loss is inside the noise. And the desk in question is the largest metals operation in the world, staffed by people fluent in every spread, straddle and ratio structure there is — the last people likely to be trapped in a one-way position.
• His actual explanation is a third force entering a market that ran the same way for fifty years. Silver demand is now high-technology volume — EVs, solar, batteries, weapons — arriving at the same moment supply is being withdrawn, with China supplying 40 to 60% of refined silver and halting exports. Demand for paper was cut by margin; supply of physical was cut by policy; industrial demand is inelastic. He is careful to say that nobody has been here before and that anyone claiming to know how it resolves does not. What the indicators point to is direction rather than mechanism: backwardation resolves about 99% of the time by physical pulling paper up, and silver lease rates at 9 to 10% against a normal 0.1% say the same thing. The gap closes upward.

2026-01-03 · open the original ↗

• Monitoring tells you about failure before a customer's call does. Guadagno lists the reasons: spotting outages before a customer reports them, security, per-user performance (one user waiting 30 seconds where others wait 15), and proving SLAs such as 99.99% uptime; Giard adds that without data a "feels slow" complaint cannot be tested.
• Azure Monitor covers whole systems, including code outside Azure. Formerly App Insights and then Application Insights, the suite monitors VMs, databases and app services, and through SDKs can take telemetry from an Android app or a third-party DLL. Guadagno says Microsoft built it to monitor Azure internally before selling it; Giard says its ADX database ran internally for about five years before he heard of it.
• Turn telemetry on early, start high-level, and keep looking. He says CPU overhead is negligible and the real cost is data volume and time spent sifting, so begin at warning and info level and drill down as you learn the system. Review it weekly or monthly, since a query fine on 100 local records can slow on 100 million in production.
• Traces exposed a slow partner call behind a flood of error emails. On one team getting 100,000 "error" emails a day, the emails nobody could act on were split into a new monitoring filter and categorized, revealing 25,000 a day from one routine. A partner call taking over a second, at a couple of hundred thousand calls a minute, was tying up threads, so a reported database problem was not the database.
• The application map shows live latency per dependency. Besides portal charts, KQL queries, alerts (e.g. responses over 200 ms or more than three errors a minute), Visual Studio and Rider plugins, and snapshot debugging of a running Azure service, he favours the map: it showed a 10 ms database link against a 600 ms external API call and is his quickest way into an unfamiliar app.

2026-01-03 · open the original ↗

• Autonomous agents keep the knowledge layer permanently "hot or warm". The host's premise is a nightly batch job refreshing a knowledge layer on an edge AI PC. Roese says autonomous agents go further: in Dell's tests, ensembles tasked with learning to run a CNC machine or produce a research report started from a batch-loaded knowledge graph. As they worked, they added pathways recording how they got from point A to point B, so the graph never goes cold.
• AI's knowledge layer is new data infrastructure, not just another database. Roese warns against applying ordinary database approaches, because the layer lives in different places and behaves differently. He argues the gains come from deploying knowledge graphs close to the point of presence and making them highly dynamic.

2026-01-02 · open the original ↗

• FPGA hardware recreation beats software emulation for retro games. Hanselman explains that an FPGA core reconfigures the chip to match a specific console or computer, aiming at cycle-accurate recreation rather than software emulators like those on Android handhelds. The open-source MiSTer project, around since about 2016–2017, runs such cores but traditionally means buying and assembling a stack of boards.
• The MiSTer Multisystem 2 is the easiest way into MiSTer. The UK-made, 3D-printed console takes MiSTer cores, offers HDMI and analog video, and accepts add-on adapters for original controllers; he says anyone who can find a device's IP address and SSH into it will be fine. He uses an 8BitDo arcade stick's 2.4 GHz dongle, a small USB keyboard and a $16 wireless dongle, and never bothered with the controller adapters.
• Zaparoo's NFC cards make a digital library feel physical. Each card holds only an identifier, written from a phone app, that launches a game when tapped on the internal reader or from the phone over Wi-Fi. He prints label art from design.zaparoo.com onto vinyl stickers; one live write failed because he didn't hold the card long enough.
• He bought it himself, and credits the open-source community. He says the video is not paid for and he bought the unit with his own money; he shows community scripts such as update-all and a remote web UI, and pays for the Patreon of one FPGA core developer whose cores run on it. The footage is straight HDMI capture with no upscaler.