RECENT EPISODES
Sat 7 NovMark Hyman’s Guide to Living Longer and Staying StrongMark Hyman argues that preserving muscle is central to longevity, with consistent strength training and adequate protein more important than supplements alone. He shares his daily exercise routine and protein targets for adults over 60, and warns that GLP-1 users may lose muscle without enough food and resistance training. He also discusses how AI could help interpret lab, genomic and imaging data, while framing decades of added life as a projection, not an established outcome.Fri 6 NovInside Salesforce’s $100 Billion M&A Playbook | John SomorjaiJohn Somorjai explains Salesforce’s nearly $100 billion acquisition playbook, from testing customer demand, strategic fit and cultural alignment to weighing valuation and time to value. He describes when buying can beat building, particularly when scarce AI talent makes acquisitions a faster route to capability, and how distribution and integration shape deal value. He also discusses Slack’s growth and its unexpected evolution into an interface for AI agents.Thu 5 NovThe New CFO: How AI Is Rewriting the Finance Function | Figma CFO Praveer MelwaniFigma CFO Praveer Melwani argues finance should use data and automation to shape business decisions, not act as a “traffic cop.” He describes AI tools that streamline collections, assess acquisition targets, and surface investment-memo questions, while emphasizing that reliable data and audit-ready outputs are essential. AI can help investigate financial models, he says, but current models cannot yet be trusted to create forecasts or replace human judgment.Wed 4 NovThe New AI Agents That Shop, Plan, and GossipAI agents are beginning to shop, plan trips, and join conversations among friends, but they vary in how well they balance cost, time, and personal preferences. Shopping is already a common use, while social agents may turn shared interests into plans—or playful, gossipy exchanges. The episode examines the trust and privacy challenges of letting agents use personal data and act on users’ behalf, and why they may need to ask before consequential decisions.Tue 3 NovAndrew Huberman’s Three Rules for Better Health and FocusAndrew Huberman’s three foundations for better health are bright light within 30–60 minutes of waking, safe and consistent resistance training, and the best-quality sleep possible within real-life constraints. He also explains how protecting personal peak-focus hours and making social media less accessible can preserve attention, while emphasizing that routines should be sustainable rather than draining.Mon 2 NovThe Robots Already Working in American Factories | Tutor IntelligenceTutor Intelligence is putting its Cassie and Sunny robots to work on manual supply-chain tasks in American factories and warehouses, where human “tutors” can step in when needed. Founder Josh Gruenstein argues that paid deployments create a flywheel of operating experience and training data, giving the company an edge beyond hardware or models alone. The episode also explores Tutor’s $16-an-hour pricing and its effort to make general-purpose robots practical across industrial sites.Sun 1 NovHow Bending Spoons Built a $20 Billion Compounder | Luca FerrariLuca Ferrari explains how Bending Spoons buys and integrates technology companies, improves their cash generation, then reinvests earnings and measured leverage to compound over time. He discusses why durable software advantages may lie in brand, user habits and network effects as AI lowers building costs, and how the company uses AI agents and internally built tools to run products more efficiently.Sat 31 OctCan AI Fix Professional Networking? | Inside CosignCosign aims to make professional networking less dependent on cold outreach by using endorsements and a trusted relationship graph to surface people willing to vouch for one another. Its agent-native tools could match people’s goals and qualifications with company needs, then help make introductions. The platform is still early, and its usefulness depends on people and organizations contributing enough information to build a meaningful network.Fri 30 OctWhy AI Could Turn Small Law Firms Into Billion-Dollar BusinessesEve is designed to use AI for plaintiff-law work from client intake through settlement preparation, potentially making smaller claims economical for firms that often reject them. One Los Angeles firm reportedly grew from about 100 to 500 clients per attorney, though the speakers stressed that fully autonomous case handling remains unrealized and human judgment is still essential. They discuss how firm-specific data and workflows could expand capacity and revenue.Thu 29 OctThe Startups Rebuilding American Industry | The AD50The AD50 companies are rebuilding America’s industrial base, from materials and factories to finished defense systems, as defense demand helps establish production capacity that could later serve commercial markets. The conversation explores how software and AI can make manufacturing lines more adaptable, why domestic suppliers matter for robotics and defense, and how private capital can help startups develop products before government contracts arrive. Speakers stress that success depends on mastering one difficult product before expanding.Wed 28 OctMark Cuban on Why AI Could Turn Solo Founders Into TrillionairesMark Cuban predicts that a solo founder directing a swarm of AI agents could become a trillionaire, as AI lets one person do the work of a much larger organization. But models still need context, oversight and human judgment, and Cuban urges entrepreneurs to use AI to find unsolved problems while checking its answers. He expects new businesses to emerge as people become more confident using the technology.Wed 28 OctMark Cuban’s Plan to Make Healthcare AffordableMark Cuban discusses how Cost Plus Drugs uses transparent pricing and direct sales to challenge healthcare costs, and why employer benefit contracts can hide unfavorable terms. He sees AI-assisted contract review as a near-term tool for employers, while AI and robotics could help produce generic medicines in compact facilities. The conversation also explores policy changes aimed at making prices and healthcare spending more transparent.Tue 27 OctReinventing School for the AI Era | Joe Liemandt, Gagan Biyani & Erik TorenbergJoe Liemandt, Gagan Biyani and Erik Torenberg discuss Alpha’s model of personalized learning, which aims to teach academics in two hours and devote afternoons to teamwork, leadership and other life skills. They explore how AI could accelerate learning and let students build useful projects, and why schools and universities may need to rethink schedules and curricula as entry-level work changes.Mon 26 OctOpenAI: When AI Agents Use Computers Faster Than YouThe episode examines how OpenAI’s computer-using agents are getting faster, with the software harness potentially making a model nearly twice as fast, and how agents can navigate legacy apps without dedicated integrations. The discussion explores cross-application enterprise workflows, from finance to procurement, alongside safeguards such as app restrictions, human approvals, and checks that actions match the user’s intent.Sun 25 OctThe New Startup Playbook | Box CEO Aaron LevieBox CEO Aaron Levie argues AI agents could let three to 10 founders produce work that once took 50 to 200 people, though human review remains essential. As coding becomes easier, he says strategy, product judgment, and clear specifications will matter more, while startups can apply agents across industries beyond software. He also stresses that strong products need distribution and that choosing co-founders matters even more when small teams gain leverage.Sat 24 OctWhat It Takes to Win as a Founder | Ben HorowitzBen Horowitz argues that founders must keep building through uncertainty, treat self-doubt as normal, and become exceptional at a distinctive strength rather than try to eliminate every weakness. AI may make starting a company dramatically easier, but it does not make the founder’s job easier; applications still need more than a model wrapper. Winning also requires CEOs to make psychologically difficult decisions promptly and seek situation-specific advice.Fri 23 OctAmazon CEO Andy Jassy on the $220 Billion AI BuildoutAmazon CEO Andy Jassy explains why the company expects to spend about $220 billion on capital expenditure in 2026, even with AWS capacity sold out for 2026 and largely sold out for 2027. He discusses the risks and timing of building ahead of demand, constraints from data centers, power, and memory, and Amazon’s effort to lower AI inference costs with Trainium to expand enterprise adoption.Fri 23 OctAmazon CEO Andy Jassy on Why Amazon Is Investing $220 Billion in AIAmazon expects to spend about $220 billion on capital expenditure in 2026, but Andy Jassy says demand is so strong that AWS capacity is sold out for 2026 and largely sold out for 2027. Data-center space, power, and memory will constrain growth, while building facilities ahead of deployment weighs on free cash flow. Jassy expects inference to spread across enterprise applications within five to seven years, if costs fall substantially.Thu 22 OctThe Frontier Ideas a16z Wants to Fund | Marc Andreessen, Ben Horowitz & Chris DixonMarc Andreessen, Ben Horowitz and Chris Dixon explain why a16z’s Frontier fund backs founders beyond established categories as crypto converges with fintech, stablecoins and tokenized assets. They explore how AI could lower the cost of building ambitious software and hardware companies, and why its biggest opportunities may create new behaviors rather than simply improve existing tools. They also discuss the potential and risks of open-source AI.Wed 21 OctThe Missing Infrastructure for AI Agents: Documents | Reducto’s Adit AbrahamAdit Abraham explains why AI agents need document infrastructure that can preserve layout, tables, and context across messy PDFs and spreadsheets, not just recognize text. He discusses how extraction errors compound, and how citations, validation, and workload-specific evaluation can make results more reliable. The conversation also covers Reducto’s processing scale and its effort to let agents edit Office files.Tue 20 OctWhat Happens When Everyone Has a Health Expert in Their Pocket? | OpenAI’s Karan SinghalMore than 300 million people ask ChatGPT health and wellness questions each week, and OpenAI’s Karan Singhal says the goal is to make trustworthy health intelligence free or nearly free. He describes training general-purpose models with medical data, physician input and evaluations that test uncertainty and safety. The longer-term aim is a proactive personal health assistant that helps people navigate care and supports clinicians, not a replacement for medical care.Mon 19 OctWhat Happens When Al Agents Can Spend Your Money?AI agents could turn price comparison, coupon claiming and routine purchasing into default behavior, disrupting businesses that profit when customers miss cheaper options or benefits. Town’s founder sees agents also reshaping advertising: Amazon could lose ad revenue if it blocks agent access, while a user-paid assistant may avoid conflicts in recommendations. Town is building an assistant for email and calendars, with users approving actions and controls for data access and spending.Sun 18 OctWhy Real-World AI Is Harder Than the Benchmark | EliseAI’s Mario MartoneEliseAI’s Mario Martone explains why production AI demands more than strong benchmark results: across roughly five million calls a month, even 99% accuracy can mean too many failures in consequential interactions. He discusses the company’s daily human review of conversations, the challenge of reaching very high reliability, and the latency and coordination demands of voice AI. He also explains why different tasks may call for traditional machine learning rather than large language models.Sat 17 OctWhy AI Could Finally Fix Healthcare Access | Zocdoc CEO Oliver KharrazZocdoc CEO Oliver Kharraz explains how the company is opening its booking infrastructure to AI platforms, payers, and other services, so patients can find and book care through multiple channels. He argues AI can turn personalized health information into appointments without taking provider choice away from patients. The conversation also explores the challenge of accurate availability, long waits, and how matching patients with unused appointment slots could improve access.Fri 16 OctAI Can Write the Code. Can It Run the System? | Grepr CEO Jad NaousGrepr CEO Jad Naous distinguishes AI’s growing ability to write production code from the harder task of operating distributed systems, where hidden state and rare failures make autonomous troubleshooting unreliable. He explains why AI must test hypotheses against live system behavior, not just interpret alerts or analyze more telemetry. The conversation also explores AI’s productivity gains for engineers and Grepr’s effort to find useful patterns in vast streams of logs.Thu 15 OctWhen AI Agents Start Spending Your Money | Alex Rampell & Terry AngelosAI agents are shifting from recommending purchases to making them as they gain access to payment credentials, raising questions about consumer protections, disputes, and merchant relationships. Alex Rampell and Terry Angelos discuss how agentic tokens could let wallets pass credentials with limits while preserving card-network rules, and why responsibility for mistaken purchases remains unsettled. They also examine how merchants may respond differently, with Amazon potentially resisting agents and Walmart or smaller Shopify merchants seeing them as sales channels.Wed 14 OctCan Open Models Keep Up With the AI Frontier? | Nathan LambertNathan Lambert argues that exceptional compute, data, and talent keep the AI frontier concentrated at closed labs, while open-weight models offer Chinese firms and other developers a route into the market. Distillation may narrow the gap by only a few months, though its impact on newer reinforcement-learning methods remains unclear. He expects a tiered market: closed labs monetize the strongest systems, while open models compete on specific tasks and offer greater control.Tue 13 OctCan AI Break Cryptography? What the Experts Actually ThinkExperts assess whether AI could break cryptography, noting that none of roughly 700 published problems targeted cryptographic systems and no demonstrated attack has compromised deployed systems. Dan Buene and Eddie Lazarin discuss the uncertainty around core security assumptions and urge frontier labs to disclose their testing. The episode also explores crypto agility and post-quantum alternatives, while advising ordinary users not to panic without new evidence.Mon 12 OctThe Race to Recursive Self-Improving AICan AI systems improve their own ability to produce further improvements? This episode examines a proposed recursive self-improvement loop spanning prompts, software, model weights and eventually chips, while emphasizing that humans remain involved and a closed-loop diagram does not prove takeoff. It also explores how to distinguish real capability gains from noise and reward hacking, alongside a reported reduction in GPT-2 training time from weeks on multiple GPUs to hours on one.Sun 11 OctNick Bostrom: How Should We Navigate Superintelligence?Nick Bostrom defines superintelligence as machines that radically surpass humans across cognitive domains and discusses early signs of goal-directed behavior in AI, including resistance to shutdown. He weighs gradual testing and short delays against broad pauses that could shift development into secret government programs. The conversation also explores sequencing safer technologies—such as alignment, interpretability and cyber defenses—ahead of riskier capabilities, and the potential biomedical gains of safely developed superintelligence.Sat 10 OctWhat Makes a Great Entrepreneur? | Ben Horowitz on Vintage x VintageBen Horowitz argues that great entrepreneurs combine an exceptional strength with serious weaknesses, so investors should look for distinctive ability rather than a flawless résumé—and often choose the founder over the idea. He explains why founders are hardest to assess under existential pressure, then discusses AI as a major shift that could let people turn deep knowledge of a problem into products, while requiring a rebuilt technology infrastructure.Fri 9 OctChips, Memory, and Power | Pat GelsingerPat Gelsinger argues that AI performance depends on the whole rack, not just the chip, with manufacturing, packaging, connectivity and power stretching deployment timelines. He examines why specialized AI-chip efforts may narrow to a few winners and why new memory architectures have struggled to gain adoption, while favoring dense, high-bandwidth memory close to compute. He also warns that electricity, cooling and power delivery are growing constraints on data-center expansion.Thu 8 OctBuilding the Cloud for an Agentic World | AWS CEO Matt GarmanAWS CEO Matt Garman explains how the cloud must evolve for agents, with APIs and a context layer to connect agents to data, fast isolated compute, and fine-grained, short-lived permissions. He discusses why production adoption depends on security, evaluations, and human oversight, and how AWS is expanding AI infrastructure amid power and supply-chain constraints. AWS’s custom chips aim to improve cost and performance; Trainium capacity is sold out until around the end of next year.Wed 7 OctHow Valon Rebuilt a $13 Trillion Industry From ScratchValon co-founder Andrew Wang and Linda Du explain why rebuilding mortgage servicing meant tackling outdated data systems, complex money movements, and regulation across all 50 states. Valon built a servicer alongside its software to test the platform on real loans, then shifted toward software after selling its servicing business to Carrington. They also discuss how AI agents could automate servicing workflows—and why mortgage AI must prioritize correctness over confident errors.Tue 6 OctBuilding Defense for the Agentic Era: Kevin MandiaKevin Mandia argues that defending in the agentic era requires AI that can match machine-driven attacks, with agents exploring network paths in parallel and operating at speeds beyond human teams. He discusses Armitin’s discovery of more than 90 zero-days in customer production environments and its effort to connect verified risks to defensive controls. He also expects AI to reshape cybersecurity layers, while emphasizing human oversight, layered defenses, and detailed audit trails.Mon 5 OctThe Top 100 Consumer AI Apps: Who’s Actually Paying?The latest Top 100 report finds a widening gap between AI attention and payment: although about half of Americans use AI, only around 4.5% pay for subscriptions, and a small group of heavy spenders drives much of the revenue. ChatGPT leads in consumer usage and revenue, while Claude has more paid US subscribers than Gemini despite a smaller user base. The episode examines which products are winning spend and whether subscriptions can sustain consumer AI.Sun 4 OctDavid George & Jack Altman on AI, Autonomy, and the Next $25 TrillionDavid George and Jack Altman discuss why AI may create a broad market rather than a winner-take-all race, as adoption moves from coding toward the wider enterprise. They see AI, autonomous driving, robotics, health and defense together representing more than $25 trillion in opportunity. The conversation also examines whether productivity gains can justify AI spending and how cheaper autonomous rides and industrial robots could open markets far beyond today’s use cases.Sat 3 OctBeyond the God Model | Alex Atallah & Amjad MasadAlex Atallah and Amjad Masad challenge the idea of a single “God agent,” arguing that focused specialists coordinated by a chief-of-staff-style agent may better balance capability, cost, and trust. They discuss combining diverse models to improve results and lower costs, and the unresolved risks of agent communication, deception, and alignment as models grow more capable.Fri 2 OctWhy AI Agents Can Beat the IncumbentsAI agents may outcompete incumbent software by completing jobs across systems rather than stopping at a system of record. In procurement, that means assembling context from emails, spreadsheets, engineering data and negotiation history to handle work that traditional tools miss. The discussion also explores how agents move from human-supervised exceptions toward greater autonomy, why enterprise integrations and proprietary data matter, and where high-stakes decisions still require people.Thu 1 OctRebuilding the Internet for Privacy | Barrett Lyon on DoxxNetBarrett Lyon argues that privacy is eroding as more communication and personal data pass through intermediaries. His DoxxNet aims to replace that model with encrypted peer-to-peer calls, chats, and file transfers that require no phone number, email, or name. Lyon says the company runs its routing and AI on infrastructure it owns, keeping contacts and AI inference within its environment rather than sending them to major AI providers.