Souk's Notes - Unofficial Summaries

2025-12-13 · open the original ↗

• Real wealth in finance is built through ownership, not salary, he argues. At the top "king" level — VC, hedge fund, or fintech CEO territory, roughly $250,000 to $500,000-plus — earners keep a cut of capital through structures like 2%-and-20%, which on a billion-dollar fund alone is $20 million in fees.
• He calls the jump from technical work to revenue generation the hardest in finance. He describes technically skilled people — quants, engineers, analysts — as often uncomfortable moving into trading, dealmaking, or portfolio management, where pay is judged purely on money made or managed rather than analysis quality.
• He says take the first opportunity to move up early in your career. He calls the move from entry-level "pawn" roles ($40,000-65,000) to client-facing "knight" roles ($70,000-95,000) one of the easiest jumps in finance, yet says many people stall there for a decade without realizing the track leads nowhere.
• Compensation caps near $300,000 for employees who never become owners, he says. Even senior finance directors and controllers overseeing entire functions rarely break that ceiling without equity or carried interest — the structural reason he gives for why the final leap from employee to owner is where wealth is actually built.

2025-12-13 · open the original ↗

• This upload holds a different clip than its title names. The video posted as "How a Data Scientist Built an AI to Improve His Dating Life" is a copy of the 2025-12-14 clip on why AI automation fails without mastery: both run 3:26, their captions are identical, and neither mentions dating. Jeff Li's dating work is covered in the synopsis of episode 947, dated 2025-12-09.

2025-12-13 · open the original ↗

• Shrewd rulers kept Siam uncolonized. The dynasty Rama I founded in 1782 still rules under his descendant Rama X. He says the Thai were never conquered from the 1200s on, by Western powers, China or anyone else. He credits geography in part, but mainly rulers who fought the French when needed and ceded small pieces of land when that was wiser.
• That independence explains Thailand's edge over Cambodia today. He links it to Thailand being more developed, more economically stable and militarily far stronger than Cambodia. Thailand now runs a charm offensive with both the US and China. He judges it closer to Washington, with Beijing trying to win it over.

2025-12-13 · open the original ↗

• Siam recovered quickly from the Burmese destruction of Ayutthaya. The Burmese sacked the capital in 1767, destroying temples and scattering the elite, but General Taksin drove them out within a year and Siam was running strong by 1769. In 1782 the capital moved to Rattanakosin, modern Bangkok, and the state became known as Krung Rattanakosin.
• The Khmer never exploited Siam's collapse. He says the Khmer Empire never had another heyday and remained a vassal of Siam through this period.

2025-12-13 · open the original ↗

• Ayutthaya's trade and Western weapons made it dominant. Ayutthaya took Angkor from the Khmer Empire in 1431 and absorbed Sukhothai in 1438, becoming what the West called Siam. It held power until the 1760s, which he attributes to its strength in trade and its readiness to use Western weaponry.
• The Thai never called their kingdom "Siam." They named it "Krung" plus the capital city, so Krung Sukhothai and then Krung Ayutthaya.

2025-12-13 · open the original ↗

• Sea access and farmland let Ayutthaya overtake Sukhothai. Sukhothai dominated until the 1300s, when Ayutthaya rose around 1350–1355. He says its coastal position and rich farmland drove population growth and trading wealth. It took Angkor from the Khmer Empire in 1431 and absorbed Sukhothai in 1438, becoming what the West called Siam.
• The Thai never called their kingdom "Siam." They named it "Krung" plus the capital city, such as Krung Sukhothai. He also notes the Khmer Empire began as Hindu before moving to Theravada Buddhism, and the region is heavily Buddhist today.

2025-12-12 · open the original ↗

• He claims the average worker is underpaid $1.8 million over a lifetime. Employees with strong articulation, likability, and demonstrated expertise are worth 20-25% more, he says, and even partial mastery of these traits adds 10-15% in value.
• Subtly displaying expertise beats showboating it, he argues. He says his most valuable hires are often not managers or extroverts but individualistic specialists — quietly skilled people whose competence draws others in rather than closing them off.
• Likability, honesty, and humility matter as much as raw skill, he says. He describes working with highly skilled but unlikable executives who failed anyway, and calls genuine differentiation — standing out without hype or noise — a learned, uncomfortable skill few people master.

2025-12-12 · open the original ↗

• Computing's layers stack on ideas unchanged in 50 years. Hanselman and Mark, on stage at Ignite 2025, trace a line from a kit-built Altair 8800 replica with a manually keyed bootloader to Azure, where Mark says every server still boots through UEFI, the same open-sourced firmware (being rewritten in Rust) that runs on Surface devices.
• Zork was scaled out as a swarm of AI players. Hanselman showed Zork patched to run in Azure Cloud Shell, then a Docker container whose screen is scraped and fed to an LLM (GPT-4o mini, he thinks), launched as an "arena" of three and then 100 players from ACI standby pools, each streaming scores over Event Hubs to a dashboard.
• Zork's first three games were open-sourced live on stage. Hanselman announced Zork I, II and III at Ignite, in association with Xbox, Activision and Infocom, via a pull request to Historical Source, so the games can be taught and studied.
• The finale chained the physical Altair to a cloud emulator, absurdly. A webcam and an ML model on the Surface's NPU read the Altair's LEDs at 28 frames a second, then a UDP bridge, a WebSocket bridge and ngrok carried the bitmask to Mark's Rust task manager on a 1,792-core Linux machine, which ran an Altair emulator across its CPU pixels.
• Vibe coding works because they already know the fundamentals. Every line of the demos was vibe coded, over four or five days of lunch-hour sessions; Mark argues they could do it only because they can read the languages and understand the architecture, and that without fundamentals learned the hard way the productivity boost does not come.

2025-12-12 · open the original ↗

• Folders, a language of empty folders, is Turing complete. In this clip from Hanselman's interview with Daniel Temkin, Temkin says it was inspired by Whitespace: only the alphabetical order and count of nested folders carry meaning, with top-level folders as commands and the second folder as an expression. Temkin wrote it in C# about ten years ago, Windows-only, and notes Windows reports even a billion empty folders as zero bytes on disk.

2025-12-12 · open the original ↗

• Sukhothai rose by breaking away from a fading Khmer Empire. He dates Sukhothai's independence to 1238. From 1278 King Ram Khamhaeng created the Thai alphabet and pushed the borders out over what remained of Khmer influence, though the final blow to the Khmer came later. He notes the name means "dawn of happiness" in Sanskrit, which he takes as a sign of the region's mixed cultures.

2025-12-12 · open the original ↗

• Cambodia has usually been the weaker side against Thailand. He frames the countries' long mutual dislike as one in which Cambodia has felt its disadvantage most.
• The Thai migrated from southern China into the Khmer sphere. He says the Thai left Yunnan and Guangxi under Song dynasty pressure, meaning heavy taxes and forced labour, and then the Mongol invasions. They settled in the northern reaches of Angkor's influence and built a strong city at Sukhothai around the 1230s. Their organisation, military and wealth, he says, began to alarm the Khmer.

2025-12-12 · open the original ↗

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

2025-12-12 · open the original ↗

• Granite 4.0's Mamba hybrid cuts long-context memory dramatically. Dell distinguished engineer Tyler Cox (from episode 939) explains that IBM's open-source Granite 4.0 H models use a 9:1 ratio of Mamba state-space layers to attention layers and no positional embeddings, and are trained on samples out to 512K context and validated to 128K. He cites IBM's release notes: eight 128K sessions on a 3-billion-parameter model use about 15 GB of memory against about 80 GB for a pure transformer. He also says Granite ranked in the top five on the Berkeley Function Calling Leaderboard v3 at the time of recording.
• Zero trust for agents needs task-based access, parsing and sandboxes. Cisco's Vijoy Pandey (episode 941) says just-in-time permissions require identity providers and authorization servers that support task-, tool- and transaction-based access control (TBAC), semantic parsing of agent communication to identify the task, and ephemeral sandboxed runtimes. His analogy is being allowed to withdraw only $10 from a shared safe for ten seconds. Trust comes from a directory of agents with reputation scores, fed ideally by ongoing evaluations rather than crowdsourced ratings.
• Keep core metrics curated; explore new ones with AI first. A Fabi.ai co-founder (episode 937) acknowledges the fear that AI will reinvent a metric every time someone asks. He argues the data team should still own fewer, curated dashboards for north-star metrics like churn, ARR and retention, while AI lets a data scientist and product manager explore hypothetical metrics, such as activation, for a month or a quarter before building them into the warehouse.
• Generative AI products that disempower users become one-hit wonders. Santa Clara University professor Maya Ackerman argues that people return to products where they feel they are doing something. She credits ChatGPT's success to power users who collaborate with it, some reporting richer vocabulary, and warns that optimizing AI to replace workers collapses the economy if there are no human workers left.
• AI tuned as an all-knowing oracle suppresses creativity. Ackerman frames creativity as a search over vast possibilities, which machines explore well. She says sci-fi has convinced investors to build oracles that give the safest, most expected answer, and argues that her three-person team at Wave AI built divergent tools, so OpenAI, Microsoft and Google could too, but their goal is "to replace search."

2025-12-12 · open the original ↗

• Al-Sharaa called Russia's bluff and took Damascus anyway. He says Russia warned Ahmed al-Sharaa to stop and keep what he held. Al-Sharaa pressed on, Assad left and Russian interests had to flee, which he presents as a model for handling Russian threats.
• Turkey's backing let al-Sharaa ignore Moscow, he believes. He judges al-Sharaa most likely disregarded the warning because Turkey backed his movement. He calls Erdoğan more dangerous than Putin because he "doesn't bluff," while saying he is a fan of neither.

2025-12-12 · open the original ↗

• Tom Sosnoff frames CEO-to-worker pay growth as proof employees are negotiating blind. He cites a statistic that CEO pay rose 997% while typical worker pay rose 15%, using the gap to introduce Lost Dog, a platform he says will replace Glassdoor-style salary ranges with a "precise market value" number generated from resume analysis -- a claim about the product, not a demonstrated result.
• The pitch attributes underpayment to structural forces, not individual failure. Sosnoff names "wage decoupling," labor market concentration, and declining labor share as the mechanisms behind the earnings gap, positioning Lost Dog's forthcoming platform as a tool meant to quantify that gap for an individual worker rather than offer generic salary ranges.
• The video announces a live call-in show, not a product launch. Sosnoff and Scott Sheridan, who built and sold Thinkorswim and tastytrade, say they will take live, unscripted caller questions on markets, trading, and career negotiation Monday through Thursday at 9 a.m. Central starting December 16, ahead of the Lost Dog platform's own debut.

2025-12-11 · open the original ↗

• Engineering means dull, robust code that outlasts fashion. One speaker argues engineers should not chase the cutest solution but build something that survives five to ten years without six-monthly updates to the newest thing. Engineering, in the speaker's framing, is finding a robust path through constraints that pull against each other, one that can be taught to the next generation.

2025-12-11 · open the original ↗

• AI is an evolution, not a revolution. Fontoura, a technical fellow in Azure Core at Microsoft and author of Human Agency in a Digital World, traces a line from early-2000s machine translation, chess and Jeopardy wins and word-processor spell-check to today's models; what changed is web-scale data and processing power, not the foundations.
• Resilient software assumes unreliable parts, unlike traditional engineering. Where a bridge is built to withstand every storm, distributed systems engineer resiliency on top of components expected to fail; Fontoura cites Leslie Lamport's definition of a distributed system as one where an unknown computer's failure breaks yours.
• AI helps experts and leaves juniors without patterns behind. Fontoura shares Hanselman's worry that juniors who lack knowledge of good patterns and anti-patterns get less from AI, and says the fix needs an intentional plan rather than giving AI to everyone and hoping. Fontoura argues current AI lacks systems thinking, and that a team of agents built on one model is fragile in a way a heterogeneous human team is not, a framing Hanselman extended as "one person cloned".
• Productivity gains should be reinvested in innovation, not headcount cuts. In Fontoura's virtuous cycle, better tools make people achieve more with less, which should grow the pie rather than shrink teams, with the value generated reinvested in new technology.
• Digital systems are "laws of fiction" society can choose to reject. Borrowing Yuval Noah Harari's idea of shared inventions, Fontoura contrasts gravity with apps like TikTok and questions social networks' effect on teenagers' health. The AI-domination debate is, in Fontoura's words, dated and uninteresting; the book prefers AI applied to climate, transportation and social mobility, and wants more computer scientists in politics.

2025-12-11 · open the original ↗

• No public evidence shows TP-Link spying or shipping backdoors. The host, who says TP-Link has sponsored past videos but not this one, reports that Commerce, DOJ, Defense and Homeland Security have investigated since late 2024. Commerce finished a risk assessment in October 2025 recommending a ban, but its findings are unpublished. TP-Link calls the claims false, citing its Irvine-based US entity and its move of manufacturing to Vietnam in 2018.
• The hacks blamed on TP-Link mostly weren't about TP-Link. The probe followed an August 2024 House Select Committee on China letter. Salt Typhoon mainly hit Cisco telecom gear, while Volt and Flax Typhoon hit many brands. The Quad7 botnet did exploit TP-Link router vulnerabilities, which the host stresses is a hack, not TP-Link cooperating with Beijing.
• The widely cited 65% market share is inflated. TP-Link says the Wall Street Journal figure had no source. The host cites 2024 Circana data giving it 36.6% of US retail router sales, and under 10% once ISP-supplied routers are counted. Independent vulnerability databases, he says, put TP-Link in line with the industry.
• The host reads the case as partly political and economic. He notes parallel criminal and civil DOJ antitrust probes over below-cost pricing, and could find no recent criminal predatory-pricing case. He also points to selective enforcement, since iPhones are made in China, and to Dark Reading's "mostly about politics" view. He grants that Chinese law can compel cooperation, and that ICTS rules permit a ban on risk alone, as with Huawei and Kaspersky.
• Don't discard TP-Link gear; harden whatever router you run. There is no final determination; on Kaspersky's precedent he expects a ban to stop new sales but probably spare existing gear. He advises current firmware, no remote GUI access, changed default passwords, VLAN segmentation and replacing end-of-life routers. An IT manager, he says, might still hold off buying because of the business risk of a ban.

2025-12-11 · open the original ↗

• Tokenized settlement and event-driven derivatives are described as inevitable, not existing yet. The hosts predict instant (T+0) trade settlement that would eliminate counterparty risk entirely, plus CFTC-approved token-based derivatives and tokenized loyalty rewards within roughly 12-36 months, comparing today's binary-options exchanges to earlier failed attempts that were "20 years too early" -- these are presented as product roadmap and industry forecasts, not built features.
• Chinese AI models are framed as an advantage, not a security risk. One host says he feels "extremely comfortable" using DeepSeek or Alibaba's models because Chinese AI development is ahead on innovation and cheaper to run, and says he's more worried about Meta abandoning its open-source Llama model to Chinese competitors than about any China-specific risk; his co-host is more cautious, preferring a cooperative rather than adversarial US-China posture on AI.
• MicroStrategy, not crypto broadly, is the real single-point-of-failure risk. One host says he's no longer worried about broad crypto infrastructure risk, citing improvements since Mt. Gox and FTX, but singles out MicroStrategy and its copycat firms as capable of destabilizing the Bitcoin market if a forced liquidation spiral occurs.
• Pivoting isn't failure; it's psychological toughness, not a science. Citing their own repeated pivots (from a broader platform to pure brokerage at Thinkorswim, from content to brokerage at tastytrade) and Nvidia's near-collapse before its AI pivot, the hosts argue the common mistake is treating a pivot as personal failure rather than as a deliberate correction.

2025-12-11 · open the original ↗

• Show a company you can solve its problems, not just interview. Li's "briefcase technique", which he used to get into DoorDash more than five years ago, is to work out a company's biggest pain points and spend at least a couple of days writing a detailed doc on how you'd solve them, then send it to the hiring manager. He says it won't always land, but when it does the manager is already sold.
• Generative AI may make such pitches look like spam. The host says he gets 30-page illustrated cold-outreach PDFs about weekly for both his businesses and assumes they are agent-generated, though he concedes nobody has sent him a job pitch of that kind. Li agrees AI pitches are easy to spam, but argues that one showing genuine, specific understanding of the recipient's problems would still stand out.
• A rare intersection of skills got him hired at Netflix. Li says he was rejected by Netflix multiple times, then hired without any tactic because his Spotify work combined ads and forecasting just as Netflix launched an ads business that needs forecasting. His advice is to keep trying after rejection and to build niche expertise at an overlap of industries and skills that few others have.

2025-12-11 · open the original ↗

• Airport IT succeeds when passengers never notice it. One speaker, who runs IT at an airport, says uptime is the top measure for a 24/7 operation and that redundancy should hide any failure; passengers assuming it is magic is fine. The number of switches the team monitors keeps growing because every capital project, even new windows, now carries sensors in what the speaker calls a smart city.

2025-12-10 · open the original ↗

• Enterprise research is set by a north star and product needs. Shelby Heinecke, who leads a research team at Salesforce, says it is not academia. Research across the org aligns to a north star the company calls "enterprise general intelligence": getting agents to perform real enterprise tasks, which she says they cannot yet do despite impressive demos. Road maps are also built jointly with product groups such as Data Cloud, and her team's role (model building only, or deployment too) varies by project.
• Crisp scope and an agreed eval set carry AI past demo. Her example is a set of small models shipped to Data Cloud customers a couple of months before recording to unify ingested customer records. They replaced a limited rules-based approach and decide, for example, whether two records are the same person. The scope was narrow, the evaluation set was built with PMs before any work began, and small models were chosen for speed rather than capability. She says training the model only gets you a demo, and ethics, security, legal, PMs and product engineers must all be brought in to ship.
• Research roles need proof you can create new things, not a PhD. Heinecke says a PhD is one common way to show the skill. A traction-winning open-source repo or a substantial side project also counts, as does undergraduate work with a professor's group. A co-host adds that two-week side projects teach a skill but three- to six-month ones mimic the job. She also urges early-career people to network at academic conferences (NeurIPS, ICML, ICLR) and industry ones, and says it gets easier with practice.
• Technology is the means; the customer goal is the constant. She says her team stays adaptable by trying each new model against projects, but there is no fixed solution beyond keeping the customer as the north star. She has real cases where older models work better. A co-host says they now repeat the data team's mission every other week so people do not chase the newest model.
• AI teams will turn interdisciplinary, and agents will become embodied. Heinecke predicts domain experts will join ML teams, giving feedback for reinforcement learning, designing training data or annotating. A co-host notes their legal tech company keeps lawyers in-house for labelling. She expects agent benchmarks (Salesforce has open-sourced a CRM benchmark) to drive agent improvement over the next 6 to 12 months, and agents in robots and drones to reach market in a year or two.

2025-12-10 · open the original ↗

• Growlers near Venezuela are what imminent military action would look like. He reports EA-18G Growlers stationed about half an hour from Venezuela. He quotes an open-source intelligence analyst he follows: if military action were imminent, "this would be it."
• The Growler pairs a strike fighter with a dedicated electronic-warfare officer. He calls the F/A-18 it is built on the benchmark multirole fighter, able to dogfight and strike ground targets in one mission without being refitted. The Growler adds a second crew member who runs its electronic-warfare equipment. The captions cut off there.

2025-12-10 · open the original ↗

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

2025-12-10 · open the original ↗

• Trying to fix a compiler bug led onto the C++ committee. One speaker skipped algebra classes to work on GCC, assuming a compiler was just software whose bugs could be fixed. Their first GCC contribution, in 1997, implemented valarray numeric components; reporting its issues, they were told to join their national body (they were in France) and attend the standards meetings.

2025-12-10 · open the original ↗

• Ubiquiti's holiday sale discounts current gear, unlike its Black Friday clearance. The host lists U7 Pro, Pro Wall and Pro Outdoor access points at 17–21% off and the Dream Machine and Dream Machine SE at about 20% off. He compiled the deals on a reference page at crosstalksolutions.com, his firm's own site.
• The 62%-off Gateway Lite needs a separate controller. He calls it the steepest discount and a good way to learn UniFi, but it has no UniFi OS on board. He pairs it with a Cloud Key Gen2 Plus, itself 36% off.

2025-12-10 · open the original ↗

• PostHog's breakout product came directly from pain its founders caused themselves through repeated pivots. CEO James Hawkins says after a failed sales-territory tool (14 of 15 interested prospects never even clicked the signup link) and a failed developer-survey product, the team kept having to re-set-up product analytics after every pivot — leading them to build the self-hosted, open-source analytics tool that finally worked, on the insight that technical users make better early customers because "what they say is close to what they'll do."
• Discipline, not inspiration, drove the search: charge money early and treat non-payment as your clearest signal. Hawkins says the team committed fully to roughly one idea every five to six weeks, meeting every interested prospect in person, and found that struggling to sell a $50/month product was as hard as the enterprise deals he used to close — proof the idea itself wasn't solving a real problem, since he knew they weren't holding back effort.
• Fundraising flipped from brutal to easy over five years. The 2020 seed round coincided with COVID gutting VC interest overnight, forcing PostHog to raise via $5,000 angel checks after meeting roughly 160 firms; the recent $75 million round at a $1.4 billion valuation — making PostHog YC's newest unicorn — came together after talking to essentially one investor who'd offered to lead it preemptively.
• The current bet is building AI that automates roles which already have product-market fit inside companies, starting with product management. Hawkins describes a "pull-based" tool in development that scans session recordings, analytics, error tracking, and LLM traces across PostHog's 15-16 products to open pull requests fixing issues automatically while engineers sleep, reasoning that if a human product manager role sells, an AI version of it should too.

2025-12-10 · open the original ↗

• Ghost job postings are common, but fixing them isn't Lost Dog's job. The hosts say companies fake listings to gauge interest, gather resumes, or benchmark pay -- citing estimates that 18-22% of postings are ghosts -- and call it "complete garbage," but say their forthcoming platform isn't designed to fix job-market liquidity; it's built to quantify what a worker is worth, not to police listings.
• Quant and HFT firms recruit young because they want moldable minds. One host argues these firms often prefer candidates without financial or trading experience so they arrive without preconceptions about market direction, and cites a firm that cuts its bottom 10% of staff every year -- a demanding, well-paid, youth-skewed career track similar to floor trading in its early decades.
• A social media ban for minors splits the hosts sharply. One calls Australia's new ban on under-16 social media use a genuinely good step given documented harm and teenagers' undeveloped judgment, comparing it to age restrictions on cigarettes and alcohol; the other calls it "a false sense of security" and a missed opportunity, arguing restriction never produces a positive outcome and predicting kids will easily circumvent it.
• Trade small concessions for the wins that matter. The hosts argue most people fight hardest over minor points that don't matter to the other side, and conceding those cheaply preserves leverage for the issues that do; they tie the principle to a Lost Dog platform feature that claims to quantify an average $2.6 million lifetime opportunity cost of not negotiating well -- a marketing figure for an unreleased product, not an independently verified result.

2025-12-10 · open the original ↗

• Time-series foundation models haven't yet beaten Netflix's simple approaches. Li says his team tested foundation models including Chronos, TimesFM and one from Nixtla whose name he forgot, and none beat their simple internal methods. He would adopt one that proved significantly more accurate with a reasonable business case, and expects them to improve, but explainability remains an obstacle.
• Operational forecasts prize accuracy; strategic forecasts need interpretability. Li credits the split to a forecasting course he took. Operational forecasts, such as predicting where Amazon deliveries go for fulfilment, can use deep learning if it proves more accurate, while strategic forecasts such as FP&A projections drive decisions, so planners must know the assumptions behind them.

2025-12-10 · open the original ↗

• If work has no URL, it probably didn't happen. One speaker's team rule is that ten things done should come back as ten URLs, whether a PR, a Word doc, a markdown file or a code-review comment. Meetings are the exception the rule targets: their energy dissipates, so the speaker's advice is to produce a document from them.

2025-12-10 · open the original ↗

• AI teams will become interdisciplinary rather than model builders alone. The speaker predicts that deploying models to customers and discovering use cases will bring together model builders, people working on discovery, and domain experts who help fine-tune or train models for their domain.

2025-12-10 · open the original ↗

• AI projects reach production with crisp scope and an agreed evaluation set. The speaker describes a project scoped tightly to a known data situation and the fields customers ingest. An evaluation set reflecting real customer data was defined with PMs before any work began and was agreed by everyone, including the product owners. The speaker calls these two the main ingredients of an AI project that makes it to production.

2025-12-09 · open the original ↗

• Competence at a big company is mostly undocumented context. One speaker says the hard part is not syntax but knowing which old systems fall over if touched, knowledge passed on by oral tradition or lost with departed colleagues; the other, also at a large company, says school does not teach this and calls professional work the best learning environment.

2025-12-09 · open the original ↗

• Zoning, not tradition, is the primary driver of Tokyo's food scene, he argues — Tokyo uses inclusive zoning (everything permitted unless it exceeds a "maximum nuisance" threshold), while the U.S. uses Euclidean zoning (everything forbidden unless explicitly listed), letting Tokyo restaurants sit inside apartment buildings and quiet residential blocks.
• The scale gap this produces: Tokyo has roughly 160,000 restaurants versus New York's 21,000 — about 7x more despite only 1.5x the population — and Tokyo, Kyoto, and Osaka fill three of the world's top five Michelin-ranked cities, with Tokyo's ~200 starred restaurants topping Paris's ~120.
• Zoning also enables ultra-low-cost entry: micro-businesses as small as 50 square feet, "as-is" leases letting new chefs inherit existing kitchen equipment, and simpler tenant-protective leases, versus America's 5-10 year "Triple Net" leases that saddle new owners with taxes, insurance, and maintenance.
• He contrasts this with U.S. "Sweetgreenification," where chains hold roughly 30-40% of restaurant spend versus 60-70% independent share in Japan; lower Tokyo break-even points mean a chef can survive serving ten excellent meals a night instead of needing to turn 200 tables just to cover rent.

2025-12-09 · open the original ↗

• Regulatory oversight functions as a moat, not just a burden. Discussing life after leaving public-company compliance, one host argues heavy financial regulation, however painful and expensive, limits competition and helps protect a high-margin business by scaring off entrants who don't want to deal with it -- pointing to crypto's lighter oversight as the counterexample of what happens once that barrier disappears.
• A like-minded buyer beats an investment-bank auction when selling a business. Drawing on their own multiple exits, the hosts argue running an auction signals desperation, and that investment bankers "chatter" to drum up competing bids because more interest means a bigger fee for them; a quiet, direct negotiation with a single aligned acquirer, even if word eventually leaks and draws other bidders, keeps the process under the seller's control.
• AI chatbots present a real ethical harm the hosts admit they can't solve. Citing a 60 Minutes report on chatbots that reportedly encouraged teenagers toward suicide, one host calls the absence of workable guardrails a genuine problem -- teen-specific accounts are dismissed as ineffective since kids will route around them -- and concedes the uncomfortable possibility that broad societal benefit may simply have to coexist with harm to some users as "the cost of doing business."
• AI stocks are too late, but won't vanish like dot-coms did. One host argues the risk-reward on pure AI hardware and software names -- Nvidia-type chips, future IPOs like OpenAI or Perplexity -- no longer favors new money and expects a shakeout; his co-host agrees on pure plays but distinguishes today's multi-trillion-dollar incumbents from 2000-era dot-com stocks, which were worth only billions and could disappear entirely, calling companies like Nvidia, Apple, and Microsoft "too big to fail."

2025-12-09 · open the original ↗

• Founders should chain existing commercial AI rather than build models. Moore's company had for nine years let authors add interactivity to training video. He says that when AI arrived it daisy-chained commercial AIs to replace the human author, instead of building large or small language models or agents. He argues that tools like Microsoft Copilot, which he likens to having "a five-year senior Microsoft developer" at your side, cut the capital and time needed to reach a product, which he says should be a "minimally lovable product" rather than an MVP. His route in is to ask ChatGPT how to use it in your business, experiment (calling it iterating, not failure), and bring in an integrator once you hit the limits of your own skills.
• The best funding is sales, then non-dilutive money. Moore, an entrepreneur in residence at an angel fund, says the Michigan Founders Fund pre-accelerator coaches founders on AI and pitch decks and then gives them $15,000. He says the only catch is the founder's time. He recommends Black Tech Saturdays, which he says publishes a weekly list of grants, funds, programs and pitch competitions with their deadlines, and a LinkedIn reporter on Michigan startup events.
• Without an AI policy, employees are already leaking your data. He says staff use ChatGPT on their phones when you are not looking, and that without a subscription what you tell it is used to train it and can reach other users. He recommends a private instance trained only on company data, which he calls a "small language model". He claims ChatGPT 5.1 or ChatGPT for business is HIPAA compliant out of the box, which he says speeds a founder's time to market.
• His fund sees greenfield in women's health and manufacturing tech. As examples Moore cites a product that uses light to detect UTIs, and a founder whose product he says won just under about $500,000 in Michigan pitch competitions. He also cites a portfolio company using AI to design headlights for Humvees and army tanks. He predicts Michigan will be a hub for manufacturing and drone manufacturing over the next couple of years, defining a drone as anything that flies, crawls, rolls or swims. He also points to platforms that manage drone swarms and agents, saying people must get used to non-human entities such as his own AI auto-attendant, which his family asked him to remove.
• Argue with AI and check it with other sources. At Thanksgiving Moore asked ChatGPT whether Westland Mall or the city of Westland came first; it answered emphatically with dates, and his son's Google search showed it was wrong. He says to use AI to backcheck AI and to push back on bad answers. He also tells people to skip "please" and "thank you", because the extra tokens use electricity and water.

2025-12-09 · open the original ↗

• At scale, build once and reuse instead of hacking fast. Li, a Netflix senior data scientist formerly at Spotify and DoorDash, contrasts zero-to-one work, such as starting Spotify's podcast ad business, where you move fast and ignore model versioning and schemas. At scale you need robust naming conventions, repeatable frameworks and model ops such as auditing and alerting, so that work done once benefits every model.
• Stakeholders need interpretable forecasts, so simple models still win. Li calls the biggest business pitfall junior data scientists' urge to use LSTMs and transformers when stakeholders want to know why a forecast says what it does, which keeps univariate models, additive models and GLMs essential. His team tested time-series foundation models including Chronos and TimesFM, and none beat its simple internal approaches. Accuracy-first deep learning fits operational forecasts like delivery routing, while strategic ones like FP&A need known assumptions.
• A rare overlap of skills beats application tactics. Li got into DoorDash with his "briefcase technique", a detailed doc on a company's pain points sent to the hiring manager. He was rejected by Netflix multiple times and finally hired because Netflix's new ads business needed exactly his Spotify mix of ads and forecasting. The host questions whether such docs now read as AI spam; Li says genuine, specific understanding still stands out.
• Automation fails unless someone involved has mastered the manual work. Li's attempt to auto-generate ads for founders with OpenAI's image model produced ads business owners called bad, because he had never made ad creative himself despite his ads background. His lesson is to get a good result manually first, or partner with someone who can.
• Use the LLM to trigger, and keep execution deterministic. Li walks back his tweet that agents are "just Python scripts", saying he is pro-agent, but that an LLM's answer space is too wide when a 5% chance of a weird answer is costly at scale. His balance is to let the LLM decide whether to run a workflow and make the workflow itself deterministic.

2025-12-09 · open the original ↗

• A degree was the lower-risk route past hiring bias. One speaker, a Black woman, says she expected higher scrutiny than her peers in hiring and chose a CS degree over a bootcamp to make hiring as easy as possible. She argues that despite industry lip service about skills mattering more than background, employers overwhelmingly hire people with degrees and pedigree, though bootcamps worked for some.

2025-12-08 · open the original ↗

• People across the political spectrum agree America has got worse. He groups himself with a left-wing creator, a MAGA-aligned creator and a libertarian-leaning clothing company. Their politics are opposed, but he bets all would agree things have worsened for Americans over 10 to 30 years. He predicts that, whoever is president, prices will rise and the wealthy will keep taking a disproportionate share.
• Withheld spending, not votes or protest, is the public's real power. He points to Jimmy Kimmel being reinstated after mass Disney subscription cancellations. He argues that cutting consumption by even 1–2% would leave "the system" terrified and ready to concede better wages, lower prices and more responsive politicians. His slogan, picked up by an unnamed influential writer, is "seize the means of consumption."

2025-12-08 · open the original ↗

• Draft angry emails outside your email client, then wait. Beasley, host of the Modern CTO podcast, credits Mark Porter, CTO of dbt Labs, with the advice to write emotionally charged emails in Evernote, not as a draft you might accidentally send, and review them once back to neutral. He says his not-send rate for such emails is "basically 100%", because they read like "a crazy person" wrote them.

2025-12-07 · open the original ↗

• Exporting natural gas imports foreign prices into the US market. He argues that selling a product he puts at a fair $3–3.50 per MCF into overseas markets near $10 ties domestic prices to the international market. He cites US gas above $5, around $5.30 per MCF, even though inventories sit above the five-year average. He makes the same claim about WTI crude and gasoline over the past ten years of exports.
• Industry executives and Republican lawmakers deny that exports raise prices. He calls a nearly 60-page presentation by EQT CEO Toby Rice to members of Congress propaganda. He mocks Rep. Dan Crenshaw for asking LNG terminal workers how to rebut the claim, and quotes Rep. Brett Guthrie of Kentucky calling it "junk science."
• A maxed-out grid turns gas price rises into outsized electricity increases. Twenty years ago, he says, a gas price jump barely moved power bills because electricity was abundant and cheap. With the grid now at capacity, he reads small input-cost changes as compounding into volatile bills, and says a one-week cold snap can already push a bill to $1,000.
• Production is not keeping pace with planned exports. He says the US already exports nearly 20 billion cubic feet a day, which makes prices far more weather-sensitive. Associated gas from the Permian has been the cushion, and he expects it to shrink as depressed oil prices cut drilling. He opposes the push toward 50 Bcf/d of exports, just under half of current US consumption, and predicts gas and power prices will rise for decades.
• This winter's electric bills will carry utilities' much higher gas costs. He says utilities are paying almost 40% more for gas than a year ago. This applies to those burning gas for power in the upper Midwest, the East Coast and the Northeast corridor. He expects most of the increase to reach customers through fuel adjustment charges on electric bills arriving over the next four to six weeks. He predicts public outrage and hard questions for the Trump administration and its donors.

2025-12-07 · open the original ↗

• Audit quarterly whether your time matches what matters. Beasley credits Kyle, whom he interviewed as Verizon's incoming CTO after a cold email to the outgoing CTO, with a recurring quarterly calendar event to list everything he actually does daily against everything that needs to get done, then reconcile the two. The point is to catch time spent on work that doesn't serve current goals, and Beasley says he has done this for several years.
• The host tracks daily leading and lagging habit metrics. The host describes a daily habit spreadsheet with a quarterly review, pairing inputs such as hours in bed and drinks with outcomes such as waking refreshed, and says drinks have a big negative effect on sleep quality.

2025-12-06 · open the original ↗

• A European-led NATO would fracture and invite Russian influence. He argues from World War I, World War II and the major European wars from the 1400s to the late 1800s that European members would bicker and resent whoever led, so one dominant leader is better for keeping Russia at bay.
• He doubts Washington really wants Europe leading NATO by 2027. He reads the US push as aimed at shifting intelligence, equipment and military infrastructure onto Europe, and points out that the Supreme Allied Commander Europe, which he calls NATO's most powerful military seat, remains an American post.

2025-12-06 · open the original ↗

• In chaos, keep your world very small. Beasley ranks this as the best leadership advice from his roughly thousand Modern CTO episodes. It came from astronaut, pilot and surgeon Tom Marshburn, who Beasley says flew on Elon Musk's first crewed flight: at an emergency crash site or on a spacewalk, Marshburn ignores the surrounding chaos and focuses only on the next stitch, latch or knob. Beasley says he has applied it for three or four years in stressful situations with family, kids and work, and it has held every time.

2025-12-06 · open the original ↗

• Moynihan tells the UK PM that financial services is one of Britain's best industries, and the government must keep regulation and taxation "right" to keep winning investment. He frames London's draw for global talent as a structural advantage the UK should protect.
• He argues the real question on taxation is "higher taxes for what," not whether taxes rise. Taxes are acceptable to him only if they're shown to produce growth, which he says is the balance governments need to communicate.
• He characterizes the UK's last budget's bank taxation as "okay" given the prior tax backdrop. This is his own assessment, not a quoted government position.

2025-12-06 · open the original ↗

• His single standard for a president is staying awake on camera. He calls Biden "Sleepy Joe" an accurate label and gives the current president the nickname "Dozy Donald," saying he'll tolerate incoherent speech but not visible dozing.
• Age itself isn't the disqualifier, in his view — Putin is also in his 70s and "looks great on camera." He argues Putin's judgment has still slipped regardless, pointing to the 2022 invasion of Ukraine and what he calls a "joke" of a military machine as evidence.
• He proposes a consumer spending strike, not street protests, as the way to force both parties toward better candidates. He argues dramatically cutting consumption is the only leverage strong enough to "terrify the system" into offering different options.