Souk's Notes - Unofficial Summaries

2026-02-05 · open the original ↗

• The crypto selloff is called orderly selling, not a panic crash. One host says crypto is down almost half from its highs but shows no real panic, framing it as orderly selling down; he is not buying at current levels and says he'd like to see a flash crash over the weekend so he can place bids in the 40s and 50s.
• Crypto's decline is framed as a new experience for under-40 holders. An analyst's statistic cited on air holds that people under 40 are about twice as likely to own crypto as stocks, and one host suggests this may be the first time that generation has seen the asset actually fall.

2026-02-05 · open the original ↗

• The pattern he is tracking is governments discovering that pension funds are a pool of money they control. Four countries have already directed that 10% of pension assets be invested domestically, in the industries the framework identifies. The objection — that fiduciaries are meant to seek the best return wherever it is — he treats as beside the point once a government needs its own economy to move. He extends the same logic to wealth taxes, which he expects everywhere without exception, and predicts the tax havens go with them: the Caymans, the British territories, all of it, because governments need the money and there will be nowhere to hide. He also concedes a viewer's point on air, that water belongs among the factors and he is working out how.
• Japan is about to consolidate roughly $3 trillion of separately mandated public money under a single purpose. Takaichi wants to use the foreign exchange fund special account — $1.37 trillion in reserves at the end of 2025, throwing off around $34.5 billion in gains and some $60 billion a year into the general account — for policy priorities. The mechanism is effectively a state-run version of the carry trade: issue yen bills, sell yen for dollars, buy Treasuries, then sell the dollar assets to buy back yen and retire the bills. The opposition is not resisting but counter-bidding, proposing a full sovereign wealth fund to pay for the food consumption tax cut. Alongside it sit the GPIF and the BOJ's ETF holdings, which amount to over half of Japan's stock ETFs — assets with no obvious exit. Sovereign wealth funds are normally built on commodities; Japan would be building one on a public sector balance sheet.
• Europe's localisation push has a precedent that worked, which is why he takes it seriously. The FT reports a controversial proposal to raise European content in manufacturing from around 50% to 70%, opposed by business. China set the same target in its 2010 fifteen-year plan — from roughly 52-53% sourced domestically to over 70% — and came in near 74%. The quoted line is that European independence has a cost, but a much smaller one than the cost of dependency, and that after Made in China, Buy America and Buy Canada it is Europe's turn. What this means for asset prices is his closing point: the world was built on the assumption that capital flows to the best return, and it is being rebuilt on the assumption that capital stays home. Corporations, individuals and governments are all at record debt levels, and reshoring costs billions — so the pools get conscripted. He points to Bessent drawing on the Exchange Stabilization Fund to send $30 to $40 billion to Argentina without oversight as the American instance, and warns that when flows change, valuations change, and not upward.

2026-02-05 · open the original ↗

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

2026-02-05 · open the original ↗

• Agency should grow as trust is earned and verified. Barth traces her own AI-assisted coding from tab completion about one and a half years ago, to agents co-developing features she reviewed line by line, to 2026, when she acts as a supervisor who cannot read every line. What made that shift safe, she says, is verification: solid unit tests, and a move from vibe coding to spec-driven development.
• UI automation will follow the same path, with humans gating key steps for now. She expects a human in the loop today to approve things like a checkout workflow or other important decisions within a task, with agency sliding toward the agent as verifications build trust.
• Side projects are where developers learn what to automate. The host's research says Barth built Summarize Me with Amazon Q in VS Code, turning meetings into animated avatar video summaries. She says she built it for fun, out of frustration with too many meetings and too many AI summaries, and that most developers have similar projects for tedious work such as expense reports or tracking AI news.

2026-02-05 · open the original ↗

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

2026-02-05 · open the original ↗

• Younger investors see private markets as less risky than public ones. The guest, who is not named in the clip, reports that millennials and younger generations are more likely to invest in private markets. They attribute this to wanting access to the SpaceX- and OpenAI-style growth they missed, and to many having been paid in equity, so they understand holding a private asset.
• Private-market managers will diverge, so diligence must improve. The guest expects mistakes as the industry grows and says there will be winners and losers, as among public fund managers. Their firm aims to make research and portfolio transparency more rigorous.
• Private markets include infrastructure, not just private equity and credit. The guest cites roads, bridges and airports at a time when, in their view, governments can no longer finance them. They say their firm is making these more investable.

2026-02-05 · open the original ↗

• AI projects stall on tool choice and on reaching production. Christensen, Rocket Ride's chief architect, says teams commit to a framework or model and find 30 days in that it falls short. Even in the best case, he says, a week of development is followed by three weeks to two months fighting deployment, scaling and threading problems that only appear in production.
• A drag-and-drop canvas replaces hand-written glue code. Inside VS Code, Cursor or Windsurf, developers connect boxes from about 120 connectors (about 20 of them LLM modules) and call the result through TypeScript, JavaScript or Python SDKs; Christensen says an audio-transcription service from a webhook node takes about five minutes. Weigert, in developer relations, says his own vibe-coded RAG chatbots grew into glue code he had to throw away after one to five days, where the same pipeline on the canvas is about 6 hours' work.
• The engine handles concurrency and version matching for deployment. Christensen says matching module versions is hard (Whisper alone comes in 10 configurations) and calls LangChain and LangFlow single-threaded, while the engine handles 256 concurrent inference streams. Pipelines deploy either as a generated Docker image or to the company's cloud service, which manages modules and scaling under SLAs.
• The engine is being open-sourced under a separate foundation. Christensen says the engine was split out of the parent's roughly seven-year-old data-management product: a foundation owns the IP and patents, and Rocket Ride Inc. hosts the software for a token-based subscription in builder, pro and enterprise tiers, with enterprise support and integrations kept out of the open source. Recorded at the end of January, they said the product and landing page launch on March 1, with Discord as the hub for events until then.
• Shared GPUs beat renting your own unless you keep it busy. Christensen's case for the hosted service is that a rented GPU runs 24/7 for perhaps 2 minutes a day of use, while the service's model server shares a large GPU pool and can even be faster. He says self-hosting only wins if you peg the GPU constantly; Giard frames it as the classic rent-versus-buy.

2026-02-05 · open the original ↗

• He announces a new podcast with two initial episodes: one on his concerns for the US economy, the other on the Russia-Ukraine conflict traced from the 1970s to today.

2026-02-04 · open the original ↗

• Full AI integration into software is called inevitable, with holdouts framed as failures. The host acknowledges AI will bring negative, unintended consequences, calling that "collateral business damage" and a normal part of growth, but argues companies that can't adapt to it will fail while those that embrace it will innovate, predicting software and AI will become fully integrated with no separation between them.

2026-02-04 · open the original ↗

• Real-time AI guidance, not automation, is the near-term opportunity in physical work. For fields like field services, manufacturing, and healthcare where AI can't yet act in the world, the pitch is a worker wearing a camera while an AI sees what they see and talks them through the job step by step — "turn off that valve," "use the half-inch wrench," "that part looks worn, replace it."
• The model is instant competence rather than months of training. With AI coaching in real time, YC argues a worker can be effective immediately and pull in new skills on demand instead of needing years of apprenticeship.
• Three converging factors make this timely now: multimodal models that can reliably see and reason about real-world situations, hardware already in place (phones, AirPods, smart glasses), and skilled-labor shortages that make the problem economically urgent. YC suggests three possible approaches — selling to companies with existing workforces, building a full-stack vertical (e.g. HVAC repair or nursing), or a platform letting anyone become a skilled worker.

2026-02-04 · open the original ↗

• Recalling at the edge of forgetting stretches how long memory lasts. One speaker describes a "mnemonic medium" in which text is woven into a system built for durable internalization; the other explains spaced repetition to teenagers as a Flappy Bird that must keep being bounced. The first adds the mechanism: recalling something when you can only just do it repeatedly extends the interval, so "the bird falls slower and slower each time".

2026-02-04 · open the original ↗

• He answers the circuit breaker question by reasoning from the regulator's seat rather than assuming malice. The market was crashing on record volume, margin had been raised to 15%, and the Fed and Treasury were on the phone telling institutions to liquidate or post capital — with everyone levered, mostly liquidate, and CTAs were selling too. Halting for ten minutes would have stopped exactly the institutions that had just been ordered to sell from selling, with the market seeing a close and no bids. His question is what reopens after that: if breakers sat at $90 or $75, does it gap to $55 and manufacture more margin calls and more distressed firms? He thinks the authorities knew what had to be sold and chose to match it off instead. He is explicit that he is constructing this, and that nobody — regulators included — has their finger on the pulse in the middle of a day like that.
• The silver problem belongs to the same class as rare earths, which is the framework point. When a global system breaks, the repair takes months or years, not days. Rare earths run fifteen to twenty years. Silver cannot be fixed at all in the ordinary sense, because COMEX operated for eighty years on its own admission at well under 1% physical, and the day everyone wants delivery the contract's worth becomes an open question. His marker is 760 million ounces contracted for March against 104 million held — not all of which will stand for delivery, but last year 16% of contracts went physical against almost nothing historically, and some contract months ran at 100%. He expects paper to fall further with each day closer to the date, unless COMEX produces metal it has not produced.
• In Japan the insurers are the pressure point, and the accounting language gives it away. Insurers are the second largest holders of JGBs after the BOJ at 170 trillion yen, about $1 trillion, and can no longer buy the 30- and 40-year issues because of the asset-liability mismatch. Forty-year bonds that were at par six months ago are at 83 and probably nearer 81 now. What matters is the collateral: it is the JGBs the banks and insurers have borrowed against, so the yen is largely a symptom and the bonds are the mechanism. An insurer told Nikkei Asia there is a risk of impairment losses if market values fall significantly below book, while also saying it has already implemented measures such as replacing holdings — which he reads as an admission the values are already below book. The other tell is a manager saying yen rates have risen to levels that make them attractive, so it is time to reassess short yen bond positions. That is the carry trade holding, and visibly starting to reconsider itself. Daiichi, Sumitomo and Japan Post Insurance are the equities he names as the read-through, with the current offset being that global tech gains are papering over the JGB losses.

2026-02-04 · open the original ↗

• Repeated, reliable workflows still need apps, not improvising chatbots. Mark rejects the claim that apps are dead, defining an app as durable code with a user or API interface (revised after Hanselman cites David Fowler's vibe-coded money tool Tally, which has only a CLI). Repetitive work needing security and reliability, like filing expenses, should be a deterministic app with AI "sprinkled in" for OCR, not a chatbot improvising every time.
• Anything done twice should become deterministic code. Hanselman agrees that using an LLM as a "poor man's" scripting language wastes it, and that the non-deterministic model should generate a script; AI is for ambiguity and reasoning. His counterpoint is that tools such as Claude's ask-user-question, which generates dropdowns and yes/no prompts on the fly, will blur the line between chat and UI.
• Throwing AI at bad interfaces dodges fixable business problems. Hanselman says needing "the world-burning computer" to drive an expense site means the UI should be fixed, or the receipts should come straight from the card issuer. Mark calls AI travel booking "dumb"; Hanselman says we keep reinventing remote calls (DCOM, CORBA, RMI, REST, SOAP, MCP) and wrapping them in an ambiguity loop.
• Vibe-coding a replacement for enterprise SaaS ignores what it carries. Mark argues services such as Dynamics or SAP embed regional deployment, data sovereignty, latency, security, compliance and decades of real customer workflows, most of it deterministic, and calls the "SaaS is dead" extrapolation senseless.
• Expertise cannot be extracted into a markdown file. Hanselman calls "knowledge extraction" anti-human and wants AI to unlock people's knowledge, not replace them. He proposes revisiting the episode in 3, 5 or 10 years to see whether Mark was wrong.

2026-02-04 · open the original ↗

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

2026-02-04 · open the original ↗

• Users want plain dashboards, not the data world's vocabulary. Javier, a founder of Index, says that after talking to close to 500 people, there is a gap between conference and investor talk (semantic layers, metric definitions) and what people use: bar charts, sheets, dashboards. Customers keep saying they do not want notebooks and have no mental model for them, so form factor sells. Javier adds that polish matters but, as another founder put it, is "necessary, but not sufficient".
• The next version is chat on the warehouse, with proactive questions. In Javier's account, Index 1.0 is a dashboard plus a SQL data studio with plain-English query writing. The previewed 2.0 suggests questions by scanning a connected source before the user asks, and shows the query, lets the user edit it and previews the underlying table. A demo thread took under two minutes. Javier positions it to sit beside Looker or Metabase rather than replace them, because asking a firm to migrate 300 dashboards is "a huge ask".
• Bad data defeats the model, so human judgment stays. On a cleaned Kaggle NBA dataset the AI reported 219 championships because it confused playoff games with titles, even with reasoning on. It also mishandles dates (UTC offsets, year-to-date returning December) and flips bar charts. Asked what a company needs first, Javier invokes the Pareto principle: data quality, with everything else "a variation" of it. Javier likens the product to the year or two before Waymo was fully self-driving.
• SQL will matter more than ever, but be more abstracted. Javier argues SQL is the lingua franca everything resolves to — Index 2.0 connections to Stripe, Salesforce and HubSpot are all queried as SQL — while users stop caring about it. The durable skills are judgment, understanding goals and metrics, and knowing which questions to ask, without which Javier warns of "analytics slop".
• The next step is acting on data, not just reading it. Beyond proactive insight, which a host also predicts, Javier forecasts tighter interconnection: emailing your top 10 leads straight from the data tool, which Javier says is neither reverse ETL nor the current paradigm. The argument is that this collapses the days-to-months lag between an insight in a deck and someone acting on it, and turns data people into more embedded, action-oriented business roles.

2026-02-04 · open the original ↗

• An agent that only sometimes works is useless in production. Barth says customer feedback put typical demo agents at working maybe 60% of the time, and "an agent that is reliable 60% is 0% useful"; that Nova Act's purpose is to complete workflows reliably 90% of the time or more. It went generally available at AWS re:Invent this past December after a research preview that began in March of the previous year.
• The workflow runs from a free playground to Python code to AWS. With no AWS account, a user gives a website and natural-language steps (her example: find an AI meetup on Luma, click RSVP and fill the form), watches the agent in an embedded browser with its reasoning traces, then downloads the Python script it wrote. IDE extensions with an embedded live preview, added after developers complained about pop-up browser windows, let them customise the script and deploy it to their AWS account.
• Action agents need reinforcement learning, not imitation. Because the model predicts the next action rather than the next token, Amazon trains Nova Act by trial and error in what it calls web gyms: hundreds of replicas of typical UIs such as forms and shopping flows, with thousands of tasks to self-play, much as AI learned chess and Go. Barth says this lets the agent generalise when buttons move or are labelled "draft", "new", "create" or just a pen icon.

2026-02-04 · open the original ↗

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

2026-02-04 · open the original ↗

• AI is earlier in its cycle than the internet analogy suggests. Haas puts it at the internet's 1992 stage, not the 1995 or 1996 that Dan Ives suggested, because he sees AI as far larger than the internet, touching nearly every human interaction with devices and software. Picking the winners in flights, groceries or doctor bookings in the mid-1990s would have been impossible, he says.
• This time the incumbents' scale may keep them on top. Unlike the early internet, Amazon, Microsoft, Meta, Google and Apple are already huge in capital, markets and consumer access. That makes them hard to displace, though Haas notes OpenAI's ambition to try. Frost raised whether on-device AI makes Apple a hardware story again; Haas allowed it "certainly could."

2026-02-04 · open the original ↗

• A tech company can have no sacred cows. Haas cites Andy Grove's Only the Paranoid Survive and his time in startups and working for Jensen Huang. He says he constantly reassesses where the market is going and where the company should go. He admits to "FOMO" about new technology and will take meetings with startups working on language models, AI for chip design or photonics so as not to miss shifts.
• Great tech leaders pair clear strategy with engineering fluency. With a 10,000-person company he can't be involved in everything. So he stresses communicating the vision and strategy clearly enough to cascade through teams, while still being able to talk technical detail with engineers. "If you can do both, that's magic."

2026-02-03 · open the original ↗

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

2026-02-03 · open the original ↗

• AI lets agencies flip from selling services to selling finished products at software margins. The speaker argues agencies can now use AI themselves to produce finished work upfront — a design firm pitching finished creative before a contract is signed, an ad agency generating video without a physical shoot, a law firm drafting documents in minutes — letting them charge far more and scale past the labor-bound limits of today's fragmented agency markets.

2026-02-03 · open the original ↗

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

2026-02-03 · open the original ↗

• Hard courses often assume automaticity students have not built. One speaker says working memory holds about seven items, plus or minus two, and we handle complex material by chunking or by internalizing steps until they take no slot at all. A physics course that seems "just really hard" is often assuming fluency students lack, leaving them like an old computer short of RAM, constantly paging to disk; the other calls it "thrashing to disk".

2026-02-03 · open the original ↗

• His reconstruction starts with a trade almost nobody noticed in the first week of January. Between the 1st and the 7th, participants he calls smarter than the average bear rolled March contracts backwards into January and took immediate delivery, removing around 24% of COMEX vault silver. Everything after follows from that. The institutions left holding paper had been, in his word, bamboozled — and from the 8th onward they went out and bought physical hand over fist, which is why the physical price ratchets up daily while the paper contract goes the other way. December had already produced the worst silver day since the Hunts in 1982; Monday beat it outright.
• The scale of the down day is what makes the standard villain story implausible to him. Volume hit 1.8 billion ounces, an all-time record and more than double the previous high. The move was 30% down and took $15 trillion out of the market — against a total US investment universe he puts at $137 to $140 trillion. He also thinks the record volume is itself the answer to the circuit breaker question: algorithmic trading at all-time volume on an all-time move overwhelmed a system calibrated to smaller ones.
• He then argues the opposite of the consensus, and says plainly he expects to be attacked for it. With the market in free fall, firms facing bankruptcy, and no way to fix the underlying problem because the physical silver does not exist — the clean solution, government selling roughly 650 million ounces to COMEX, was never available — the practical option was to call the largest bank with the largest metals desk in the world. His claim is that JP Morgan was funded to go in, make a large bid, set the bottom and clear the forced sellers, because the CTAs and margin-called accounts had to be cleared that day even though the real problem would take months. The widely repeated version — that JP Morgan shorted the market down and covered at the lows for profit — he calls BS, on the grounds that the thing needed stopping and he cannot construct another mechanism that stops it. He invites anyone with a better one to supply it.

2026-02-03 · open the original ↗

• Agents chattering on their own forum show emergent agency, not sentience. Hanselman says the Moltbook claim that "30,000 AI agents made their own Reddit" confuses behaviour that looks goal-directed with having goals. Wire a planner, a memory store and tools into a loop with retry logic and you get behaviour that breaks down tasks and corrects itself; like ant foraging, market price discovery or a thermostat, that is an external description, not an inner state.
• The goals belong to the humans who built the loop. He argues the agents lack self-originating goals, a continuous self across runs and any stakes: one can be deleted and no one is harmed. "If you unplug the loop, none of these agents want anything."
• Three stacked illusions make agents feel sentient. He names a language illusion (fluent talk, like a parrot), a continuity illusion (memory markdown files that greet you again) and a competence illusion (high capability mistaken for consciousness, as with Deep Blue beating Kasparov). His analogy: emergent agency is to sentience as a flight simulator is to a bird.

2026-02-03 · open the original ↗

• BitChat messages nearby phones over Bluetooth with no network at all. The free iOS and Android app, created by Twitter co-founder Jack Dorsey, needs no account or phone number, is anonymous by default and end-to-end encrypted; the host demonstrated iPhone-to-Android text with Wi-Fi and cellular off, and it also sends pictures and voice notes.
• Mesh relaying extends range beyond one Bluetooth hop. Messages hop, still encrypted, through other BitChat users' phones, so coverage grows with local users; the host likens it to Meshtastic without the electronics. Location channels offer neighborhood, city and region scopes.
• The host sees it as a fallback where cell service fails. He cites crowded festivals and stadiums, CES in Las Vegas where he could barely send a text, flights, power or internet outages, and conversations that should avoid a server. Triple-tapping the logo ("panic mode") clears all conversations.

2026-02-03 · open the original ↗

• Dark patterns are design choices that steer users against their interests. Appel, a JetBrains developer advocate in Germany, defines them as UI and UX choices that coerce, trick or deceive people into choices they would not make with full information. Some are mere bad design copied from big sites without thought, and she warns of a growing body of legislation, especially in the consumer-protective EU, plus US lawsuits.
• The online playbook: blocked cancellations, sneaked subscriptions, disguised permissions. Her Audible cancellation was refused by an email saying her reason was unacceptable; Giard's stamp purchase silently became a $5-a-month subscription he caught three months later. Other examples she gives are warranties added to carts, notification prompts saying "press allow to prove you're not a robot", false scarcity, cookie pop-ups, and guilt-trip opt-outs like "No, I don't want 15% off". She says elderly relatives are hit hardest.
• Physical stores use them too, on a sliding scale. She says grocery stores place children's items at children's eye level (she once wrote shelf-scanner code for a food manufacturer), casinos trap people, and IKEA's stores route and tire shoppers while its website has no dark patterns. Giard and Appel agree good versus bad is not binary.
• They work by hijacking the brain's autopilot. Appel cites Robert Cialdini's principles of persuasion and his "click, whirr" effect: because people cannot consciously weigh every decision, the right trigger sets them off. Her examples are 99-cent pricing, Amazon A/B-testing how far apart a struck-through and a sale price sit, the mere exposure effect, and decision fatigue.
• Dark patterns cost more later, and developers can refuse them. She cites estimates that they can cost up to three times their gain in complaints, ill will and refunds, from a talk by Harry Brignull and a collaborator. Her remedies are for developers to push back on marketing's requests, QA to add them to quality metrics, a part-time ethics committee to review products, and an attorney to check them; deceptive.design keeps a "wall of shame" of examples.

2026-02-03 · open the original ↗

• A $500 million investment four days before inauguration, followed months later by a major export policy reversal, is the core fact pattern. Abu Dhabi royal Sheikh Tahnoon invested $500 million for a 49% stake in Trump's World Liberty Financial (crypto venture) just before the January 20, 2025 inauguration; in March 2025 the administration approved exporting 500,000 advanced Nvidia AI chips to the UAE that had previously been blocked on national-security grounds.
• This doesn't fit the standard bribery model, which is precisely what makes it legally novel. The FCPA bars Americans from bribing foreign officials, but here the money flowed from a foreign official to the Trump and Witkoff families — so prosecutors would need to argue the reverse: that the AI chip export approval was the "thing of value" used to induce the $500 million personal investment, effectively treating chip access as currency for a private windfall.
• Federal bribery law (18 U.S.C. § 201) may be the stronger theory than the FCPA. It criminalizes a public official receiving anything of value in exchange for being influenced in an official act; the author argues the timing (payment front-loaded before scrutiny could begin) and the subsequent chip approval fit a "quid pro quo" pattern more cleanly than FCPA's foreign-bribery framework, especially since Trump's own February 2025 executive order paused FCPA enforcement.
• Neither Trump nor Witkoff has structurally separated from the business benefiting. Unlike Jimmy Carter's blind trust for his peanut farm, neither placed his holdings at arm's length; Trump can claim broad presidential immunity for official acts, but Witkoff, as a Special Envoy, is a federal employee bound by conflict-of-interest statutes and lacks that protection — and the author argues the case's scale exceeds anything alleged against the Biden family.

2026-02-03 · open the original ↗

• Agents must be near-certain to be worth deploying. Barth, a member of technical staff and developer relations lead at Amazon AGI Labs, says customers report demo agents working maybe 60% of the time, and "an agent that is reliable 60% is 0% useful". Nova Act, generally available since AWS re:Invent in December, targets 90% or more; she says it exceeds 90% on early enterprise customer workflows, evaluated on those customers' own tasks rather than just public benchmarks such as WorkArena and REAL Bench.
• Reinforcement learning in "web gyms" teaches actions that generalise. Imitation learning suits next-token chat models, she argues, but agents must predict the next action, so Nova Act self-plays thousands of tasks across hundreds of replica UIs, with variations as small as light versus dark mode. To stop agents learning shortcuts, success is judged on a verifiable end state, such as the meeting actually appearing in the calendar or a database entry existing, while the path is left free.
• Training model and actuator together avoids "Frankenstein" agents. Barth contrasts stitching a model, orchestrator and browser actuator with Amazon's approach of training "the brain and the body in one", while noting AWS also offers the Strands Agents SDK for teams that want to build their own. Runs execute in the customer's AWS account with step-by-step reasoning traces, next-action predictions, scheduled parallel runs and CloudWatch logs; for passwords the agent hands control back so the secret is entered locally and never sent over the network, an area she says 1Password built on, and CAPTCHAs are handed back the same way.
• "Normcore" agents automate the boring work, starting with QA. PGA Tour checks its site before tournaments, Hertz raised shipping velocity 5x by automating rental-workflow tests, and Amazon Leo generates test cases from wireframes, by her account; the host cites her count of over a thousand generative AI applications at Amazon. She expects QA engineers to stop maintaining brittle Selenium-style or RPA scripts and shift to natural-language tests and reviewing traces.
• Agency should expand as verification builds trust. Barth's coding moved from tab completion to line-by-line review to supervising agents in 2026, made safe by unit tests and spec-driven development, and she expects UI automation to follow, with humans approving steps like checkout for now. Her forecast is agents as digital co-workers that learn from interaction like humans, with every interaction an agent call and a mix of small edge models and large reasoning models.

2026-02-03 · open the original ↗

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

2026-02-03 · open the original ↗

• A consumer expectations index reading below 80 is a possible recession signal, and the index has been oscillating around that line since 2022. He says he doesn't know how much longer it can hover there without tipping into an actual recession, even as he notes offsetting GDP data: October GDP growth of 4.4%, with January figures suggesting year-over-year growth could come in over 5%.

2026-02-03 · open the original ↗

• The current delinquency rate, about 2.6%, is now higher than it was during COVID. He attributes the earlier, lower COVID-era rate to how fast governments worldwide flooded the economy with cash to avert collapse, suggesting today's higher rate reflects the absence of that scale of support across mortgages, credit cards, and other debt.

2026-02-02 · open the original ↗

• Federal interest payments as a share of GDP are approaching their worst level in US history. Citing FRED data, he says the prior peak was just over 3% in 1990, versus a 1.7% peak during World War II, and the US is now nearing that 3% level again — close to $1 trillion a year in interest with no offsetting return.

2026-02-02 · open the original ↗

• Doubt about US debt is driving gold and silver higher, with gold hitting $5,600 an ounce and silver breaking $120. He attributes silver's rise partly to China's export controls and the US designating it a critical mineral, given about 70% of silver demand is industrial, and describes gold as the standard safe haven against the dollar.
• He suggests major foreign holders like China and India could use maturing US debt, rather than new sales, to exit dollar exposure. His proposed mechanism: let existing Treasury holdings mature without rolling them over, then redirect the proceeds into gold.

2026-02-02 · open the original ↗

• AI coding tools solve building software, not deciding what to build. YC argues Cursor and Claude Code are good once it's clear what needs building, but the harder, more important work — talking to users, understanding markets, synthesizing feedback — still lacks an AI-native system covering the full product-discovery loop.
• The pitch is a tool that turns customer data directly into a proposed feature. Upload customer interviews and usage data, ask what to build next, and get back a feature outline with a rationale grounded in that feedback, plus proposed changes to the UI, data model, and workflows.
• The tool would also hand off directly to coding agents. It would break the proposed changes into development tasks sized for a coding agent to implement, reflecting YC's view that as agents take the first pass at implementation, how teams define and communicate what to build needs to change.

2026-02-02 · open the original ↗

• Every dollar spent on home construction generates roughly $3 or more in GDP growth, a multiplier he attributes to the Bureau of Economic Analysis. On that basis, he projects that $1 trillion spent on home construction could translate into about $3 trillion in GDP growth.

2026-02-02 · open the original ↗

• He names America's most pressing problem as a potential loss of faith in the government's ability to manage its own debt. He frames this as distinct from, but tied to, broader doubts over US trade policy stability and the dollar's standing in the world economic order.

2026-02-02 · open the original ↗

• At an $80,000 salary, buying a home now costs five years of saved annual income, which he says is freezing up the real estate market. Rising rates discourage buyers who can't afford payments, while owners locked into older 3–4% mortgages are reluctant to sell and trade up into a pricier home with a much higher rate.

2026-02-02 · open the original ↗

• A silver trade taking "a ridiculous amount of risk" is called not worth it despite breaking even. The host compares it to his GameStop trade, where he says he lost twice as much money on the way up as he did in silver before making it back, leaving him up small on GameStop lifetime-to-date; he says the episode shows commodities can trade like meme stocks and that an asset can always move further than assumed possible.
• The losses are described as resetting the host's trading discipline. He attributes the damage to adding to his position on the way up and shorting at multiple levels along the way, calling it the kind of setback that periodically "slaps" a trader back to reality.

2026-02-02 · open the original ↗

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

2026-02-02 · open the original ↗

• Looking things up fails when you don't know what's missing. One speaker argues that in technical work it may never occur to you that a problem is essentially logarithmic, and that creative insight comes from noticing a coincidence, contradiction or connection, which requires both pieces already in your head. The other agrees you can't Google what you don't know to ask, and says AI will not solve that.

2026-02-02 · open the original ↗

• He asks how much longer the US can keep servicing its debt if meeting interest payments is already this difficult despite an apparently good return on investment.

2026-02-02 · open the original ↗

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

2026-02-02 · open the original ↗

• He argues the headline $38 trillion debt figure is less meaningful than debt as a share of GDP, which he says is back near its World War II-era peak. He traces debt-to-GDP from about 125% in 2020, down to roughly 118% under Biden, before it climbed again in the second and third quarters of 2025 as total debt rose from about $36 trillion to just over $38 trillion; against an estimated ~$30 trillion GDP, he puts the current ratio at about 123–124%.

2026-02-02 · open the original ↗

• One statistic from the crash tells you nothing was resolved. Futures volume hit 369,000 contracts, or 1.8 billion ounces, more than double normal daily levels — and open interest fell only 5%. All that trading changed hands without discharging the obligation: silver is still owed on 95% of the contracts. On the missing circuit breakers, CME has said nothing, and he offers the mechanical possibilities rather than a theory — algorithmic trading overwhelming the system, or breakers calibrated to smaller prior moves simply not fitting a day with no precedent, the Hunt brothers included.
• The drain has been running for a year, and the January acceleration is what broke it. COMEX gold inflows rose 153% in Q1 2025, but every contract month set delivery records, and silver saw 16% of registered inventory delivered in some months against the historic well-under-1%. December alone saw a record 64 million ounces delivered, roughly 25 to 30% of registered and eligible inventory, with 47 million claimed in the first four trading days of January. London settlement stretched to T plus eight weeks. Then in the first week of January, warehouses drained by 33.45 million ounces — about 26% of registered stock — in an ordinarily quiet non-delivery month, driven by industrial users rolling futures backwards for immediate metal. China's export curbs on refined silver, of which it supplies 60 to 70%, took effect on 1 January.
• The arithmetic he keeps returning to is a coverage ratio, and it reframes the entire question. Inventory fell to 107 million ounces on 27 January against March open interest of 760 million — 14% coverage. So the question is not whether silver goes up or down but what a futures contract is worth: 100 cents on the dollar, or 14? Analysts he cites warn that a 20 to 40% delivery stand exhausts supply and forces cash settlement, eroding exchange credibility, with the CME default waterfall standing behind it and the defaulter paying first. The detail that makes it worse is buried in the January delivery data: 1,430 contracts open at the start, 1,624 discharged, and 2,155 open at the end — more silver owed in March than before, not less. Anyone who did not roll back in the first week is now looking at those numbers, which is why he thinks the downward move in paper may only be beginning, and says outright he thinks this is the end of COMEX.

2026-02-02 · open the original ↗

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

2026-02-02 · open the original ↗

• Consumer confidence has fallen past its 2020–2021 lows to its weakest level since 2014, which he ties to tariff effects. He links the decline to falling oil prices and other weakening economic metrics, citing the Conference Board's survey of 5,000 households as showing a broadly negative economic outlook.

2026-02-02 · open the original ↗

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

2026-02-02 · open the original ↗

• Agentic drug discovery is a cross-vendor, multi-model pipeline. Answering the host's recap of his episode 941 vision, Pandey lays out the stages: an LLM iterates with the user to define the goal and break it into tasks, an AlphaFold-like scientific foundation model explores protein structures, compliance, effectiveness and manufacturability agents filter the candidates over many iterations to perhaps 10 options, and robotic labs run wet-lab experiments before any animal or human trials.
• Passing messages is where the field is today; shared intent is missing. Pandey says AGNTCY, MCP and current work only connect the agents. They also need a common goal (a drug for a specific purpose, price point, population and risk profile) reached through semantic protocols for discovering information, resolving conflicts, negotiating and coordinating.
• Shared knowledge and guardrails complete the stack. Agents must pool, add to, redact and modify knowledge of what has and has not worked; his example is a compliance agent drawing on the past to warn that a promising candidate would be stopped at FDA approval. Above that sit cognitive engines that both accelerate experiments and keep them within bounds, which he calls shared innovation.

2026-02-02 · open the original ↗

• AI workloads will move from data centers to local devices. Haas calls AI just a compute workload. He says workloads historically migrate to local devices such as phones, PCs, wearables, glasses, watches, "something that's not invented yet." This won't replace the data center entirely. He is "100%" sure Arm, already present in these low-power devices, will be at the heart of it, though he doesn't know what future models will look like.
• There will be AI overinvestment at some point, but not yet. Haas separates a valuation bubble, which he won't comment on, from an overinvestment bubble. He doesn't think there is overinvestment in large language models now. But he expects them to hit limits on efficiency relative to the capital they need, and more efficient or different models to replace them.

2026-02-02 · open the original ↗

• Housing starts collapsed from 2 million to about 400,000 after the Great Recession and still haven't caught up to population growth. Starts peaked at only 1.6 million in 2022 despite the population growing by 30 million since 2005, when about 2.1–2.2 million starts were typical — a shortfall of roughly 600,000 units he reads as a drag on construction-linked employment across materials and trades.

2026-02-02 · open the original ↗

• Data quality is the precondition for agentic analytics. The speaker argues that teams told to "use more AI" will find it is not a switch: with bad data agentic analytics will not work, just as ordinary analytics will not. Invoking the Pareto principle, they suggest nearly every other requirement they could name is a variation of data quality.

2026-02-02 · open the original ↗

• AI analytics should move from reactive to proactive, like an analyst. One speaker describes a future where the system surfaces what you need, knows its own limits and asks humans what to do next, leaving people to judge whether it makes sense. Another likens this to an analyst's career: reactive and told what to do, then connected with the team, then proactive — anticipating questions the team would never have thought of.

2026-02-02 · open the original ↗

• Consumer confidence drops no longer reliably predict recessions. Confidence fell sharply before 2008 and before the 2020 shutdown (that time dropping from about 123 to 82), both followed by recession, but since then comparably large drops and partial rebounds have not led to one.

2026-02-02 · open the original ↗

• Agents are next-token prediction plus context, tools and a loop. In this explainer for non-technical viewers, Hanselman says LLM calls are stateless, so context (a CLAUDE.md, Copilot instructions, your email) must be passed in each time; 25 years of marriage is why "let's go to the…" means the bagel shop for his wife. Tool calling lets the model read files and call code, and MCP servers are "USB for AI".
• An agent that writes and runs its own tools is powerful and dangerous. Asked his blood sugar, it can search for the Nightscout API, ask for his URL and token, make the REST call and report "your blood sugar is 100". The building blocks are "effectively unlimited" because all open-source software is available, which is how people will "get pwned".
• "By any means necessary" invites uncanny-valley behaviour, not consciousness. Asked to book a table at a restaurant with no API, he says an agent might download a text-to-speech model, clone your voice and phone via a service like Twilio. His own OpenClaw texted him repeatedly during a low blood sugar until he told it to chill; he calls these emergent behaviours that make it seem smarter than it is, and says this is not AGI.
• Don't run these agents in YOLO mode with your accounts. He advises against giving them full access to email or bank accounts, and mocks a Twitter user handing a bot his account with instructions to "buy anything".

2026-02-02 · open the original ↗

• Debt mismanagement is eroding global faith in the dollar, and he expects that to show up as fewer imports rather than more exports. He cites the dollar index (DXY) weakening and gold outperforming the dollar as signs of lost faith; a weaker dollar should normally lift exports, but trade barriers mean the data hasn't shown that yet, so he predicts import volumes will fall instead.

2026-02-02 · open the original ↗

• Credit card delinquencies nearly doubled between 2022 and 2024 and remain elevated. The rate rose from about 1.8% in early 2022 to roughly 3.2% by early 2024, held there, then eased only slightly to 3.0% in late 2024 — he reads the still-high level as a sign of ongoing and future financial strain on borrowers.

2026-02-01 · open the original ↗

• Price increases from the past five years haven't gone away for most Americans. He says the average American pays 30% more for goods than five years ago, and argues every income bracket except the top 15% is still feeling that burden regardless of what headline economic numbers suggest.

2026-02-01 · open the original ↗

• Home construction spending has an outsized multiplier effect on GDP. He cites Bureau of Economic Analysis data showing every dollar invested in home construction generates about $3 or more in GDP growth.

2026-02-01 · open the original ↗

• He flags federal interest payments, as a share of GDP, as the real economic threat. Citing FRED's "federal outlays: interest as a percent of GDP" series, he notes the previous worst point for this ratio was just over 3% of GDP, reached in 1990.

2026-02-01 · open the original ↗

• He works the JP Morgan accusation as an arithmetic problem, which is the whole episode. The bank's balance sheet is around $3 trillion, call it $4 trillion, of which roughly 96% is borrowed against 4% capital, levered fifteen to twenty times. Friday's silver move was 30% and removed what he reckons at $15 trillion, implying a market several times that size. His question is simply how a bank of that size sells enough silver it does not own to move a market that large by a third — and then covers the whole position at the bottom. It does not compute, which is why he locates the cause in the plumbing rather than in a single actor: the metal owed against those contracts does not exist.
• The situation he wants people to picture is the industrial buyer's, not the trader's. Holders who rolled February and March contracts backwards into January took immediate delivery and removed 26%, by some accounts up to 60%, of COMEX silver. Everyone still holding February, March and April paper is now a manufacturer or a technology firm with procurement teams searching the world for metal — which any CEO would do and most now must — and finding none. So they look at a contract for ten million ounces held by a venue that does not have ten million ounces, which is expected to source it in a market where they themselves cannot buy. At that point the paper contract trades at a deep discount because it cannot perform, and he expects it to last months. His endgame is that COMEX may disappear, possibly with CME forced to combine it with the Shanghai exchange, because a paper exchange no longer works.
• On Europe he is collecting proposals, and the striking feature is what none of them contain. The E6 grouping, the Coalition of the Willing, the European Security Council, a tiered EU, a Nordic-Baltic vanguard, a Union of Unions, with the illiberal bloc — Hungary, Slovakia — already being talked about as excluded. Not one of these proposals describes the EU as it exists today. He treats that as the answer to a question people are not asking: the euro is a currency attached to a political union, second in transactions at 18 to 20% against the dollar's forty-some. If the political union is being reorganised, what stands behind the currency — whose taxation, whose military, whose ability to feed and power itself? He thinks Europe's future is bright and that the transition is the caterpillar-to-butterfly passage the title refers to: no signing ceremony, no applause, just something new and contested arriving this year. His stated conclusion is narrow and falsifiable — the euro goes down this year, and no, it does not disappear.

2026-02-01 · open the original ↗

• AI is far older and broader than LLMs. Appel, a JetBrains developer advocate, objects to "AI" being used to mean an LLM, especially ChatGPT, and to the claim it came out of nowhere. Recommendation engines and car-factory robotics run on classic computer-science algorithms, and she argues that once people get used to an AI they stop calling it AI, which she expects will happen to LLMs too.
• Pattern matching underlies all of it, LLMs included. She says transformer architectures still match patterns at bottom, so "fancy autocorrect" under-hypes LLMs while the marketing over-hypes them.
• A Commodore 64 book already ran core AI algorithms. Her talk draws on AI Projects for the Commodore 64, published in 1985 for a machine with 64K of RAM and no hard drive; her husband retyped its samples into her GitHub repo, which runs in an emulator, and similar books are on archive.org. Its programs include Towers of Hanoi (recursion, state representation), the Knight's Tour via Warnsdorff's rule or backtracking (which she says could guide a Roomba), tic-tac-toe that reapplies winning moves as basic reinforcement learning, an Eliza-style chatbot, and a logic program that infers a cat is not a chihuahua.
• The book already warned against treating chatbots as people. She quotes it saying people took Eliza seriously and shared personal feelings with it, and telling readers to beware that attitude. She ties this to humans anthropomorphizing machines, and says an LLM reflects us back by applying patterns to our own internet data.
• What is new is data and compute, which deserve thrift. Appel, and Giard with her, credit cheap compute and storage for today's LLMs, while she notes data centers draining water and power grids in some communities. She points to DeepSeek as working within limited resources and argues for revisiting classic algorithms and the C64-era habit of counting bits, rather than throwing hardware at problems, to make AI cheaper.

2026-02-01 · open the original ↗

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

2026-02-01 · open the original ↗

• Natural language stays the common denominator while humans remain involved. Pandey expects all future work to happen in "multi-agent human societies" with agents and humans on the same level rather than in hierarchies. Agents can work at machine speed through API calls in ways humans cannot, but with humans in the mix "for a long long long time", agents must communicate in natural language.
• Even with no human in the loop, language persists because vendors differ. Under his definition of superintelligence, agents doing discovery or economically viable work 100% of the time without human intervention, components still come from different vendors and frameworks, so language remains the shared medium.
• Vector-space communication is more efficient but needs universal agreement. Pandey says converting vectors into tokens and language loses a great deal of information, so if everything were standardised and everyone agreed, vector space would be the way to communicate; he does not expect that in practice. His caveat is world models: citing recent work from FAIR and elsewhere, he says tokenization is a big detriment there and vector space the only way, a problem to be solved eventually, though for now his distributed-superintelligence work targets tokens and language.

2026-02-01 · open the original ↗

• Trump nominated Kevin Warsh as the next Fed chair on Friday, right after a January FOMC hold that saw two dissents — Governors Waller and Miran — favor a 25-basis-point cut. Warsh's confirmation timeline remains uncertain, clouded by Senate resistance tied to the unresolved DOJ investigation into current Chair Powell.
• Global trade realignment advanced on two fronts, with the EU striking what it called "the mother of all trade deals" with India while the UK pursued closer China ties Trump called "a very bad idea." This came as Brent oil topped $70 a barrel midweek on rising US-Iran tensions and the dollar swung on yen-intervention rumors before rebounding on the Warsh news.
• The week ahead centers on Friday's January jobs report, with consensus at just 40,000 new jobs and unemployment at 4.5%, amid signs of central bank policy divergence. The Reserve Bank of Australia is expected to buck holds from the ECB and Bank of England with a hike, underscoring what he calls an emerging theme after years of synchronized global policy moves.

2026-01-31 · open the original ↗

• Friday's 29% collapse in paper silver happened while physical went up two dollars, and the venue list makes the split unmistakable. London and COMEX printed $78 to $85, all paper. India was $120 to $125 physical, Shanghai above $150, Japan $125 to $135, Europe $90 to $110. This is the worst silver crash on record and the physical market did not participate in it. CME's response is another round of margin increases across gold, silver, platinum and palladium effective 2 February — the lever that reaches the contract and not the metal, which is why he wonders aloud whether the futures market is turning into a pure cash market.
• The mechanism he identifies is a backwards roll that emptied the vault in the first week of January. Between the 1st and 7th, warehouses drained 33.45 million ounces — about 26% of registered inventory — in a non-delivery month, driven by industrial and sovereign buyers selling March contracts and buying January and February for immediate metal, against China's new export licensing regime choking 67% of refined supply. On 7 January, 8.1 million ounces were delivered against 1,600 contracts, 99% issued by JP Morgan — which he notes reflexively gets read as conspiracy until you remember JP Morgan is the largest metals trader in the world by a distance. The number that matters most: open interest rose through all of it, from 1,431 contracts to 2,155. They delivered a quarter of the vault and finished owing more.
• His theory of the crash is that the discount stopped being a discount and became a repricing. For six weeks paper traded $10 or $11 below physical — the market's way of pricing uncertainty about whether a contract converts into metal, the mirror image of a premium on an option expected to pay. After the first week of January cleaned out the vault, the remaining institutions asked the obvious follow-up: if I cannot buy physical anywhere, what makes COMEX confident it can? Somewhere in the past week they concluded it cannot, and sold the contract for what it is worth rather than what it promises. On the widely circulated claim that JP Morgan dumped and covered at the bottom, he poses the mechanical objection he keeps returning to: if you are short, you can only buy — so what exactly did a short seller sell to drive the price to $74, and what did covering consist of?

2026-01-31 · open the original ↗

• Rising unemployment and loan delinquency point to weakness under the headline GDP numbers. Unemployment has trended up since 2023, credit-card delinquency nearly doubled from about 1.8% (2022) to a 3.2% peak (2024) and still sits near 3.0%, and all-loan delinquency at roughly 2.6–2.8% is higher than during COVID.
• A housing shortage is dragging on the whole economy, in his view. Housing starts peaked at 1.6 million in 2022 versus roughly 2.1–2.2 million in 2005 despite 30 million more people since then; the median home price is about $410,800, and he cites a Bureau of Economic Analysis multiplier of roughly $3 in GDP per $1 of construction spending as the case for loosening zoning laws.
• He argues the reported 4.4–5% GDP growth is an artifact of tariff-driven import timing, not real strength. Imports spiked 15% year-over-year in Q1 2025 as firms front-loaded before tariffs, then fell to -3% by Q3; because imports are subtracted in the GDP formula, that drop mechanically inflates growth, and he predicts growth will slow toward roughly 1% as the effect fades.
• Federal debt service, not the $38 trillion debt itself, is what worries him most. Debt-to-GDP is back near 123–124%, last seen in World War II, and interest payments are approaching 3% of GDP — above the 1.7% peak paid during WWII — which he links to gold near $5,600/oz and silver breaking $120/oz as signs of eroding faith in the dollar.

2026-01-31 · open the original ↗

• Data scientist pay at Facebook climbs steeply with each level. The speaker gives average annual pay of $157,000 at IC3, $250,000 at IC4, $370,000 at IC5 and $515K at IC6; the IC7 figure is cut off mid-clip ("around $750").