Joe Lonsdale: American Optimist

Anthropic is one of the fastest-growing companies in history and sits at the forefront of the AI race. This week, we’re joined by two key leaders on Anthropic’s technical team — Sholto Douglas and Nicholas Marwell — for an inside look at the AI frontier. Are we on the cusp of AGI — and what does that mean? What are they most excited about and most concerned about? And where does Anthropic stand on open source, as well as criticisms of its regulatory approach?
Sholto Douglas grew up in Sydney, Australia, where he studied robotics and taught himself Machine Learning research; he went on to work at DeepMind before becoming a tech lead in Reinforcement Learning (RL) at Anthropic. Nicholas Marwell, a San Francisco native, studied at Yale and joined Anthropic early in its journey to lead RL Science and now the Horizons team. We begin with the state of AI progress, and why Sholto and Nicholas believe AGI is likely within the next couple of years. We unpack their case for AI optimism, why biology is the most exciting next frontier, and whether a post-scarcity world is truly achievable. Then, they explain why the next decade “belongs to the generalists” and offer advice for young people entering the workforce in the coming years. Then we dig into the hard questions: offense vs. defense in cyber and bio, regulatory capture, open source, distillation, and competition with China. We also discuss the importance of virtue, faith, and institutions, and whether the leading AI researchers in Silicon Valley share the values of the heartland. Finally, they offer a few predictions for the years ahead and whether AI will soon win a Fields Medal or even a Nobel Prize.
NOTE: This conversation was recorded about a month ago, and delayed beyond our control, but it’s still timely.
00:00 Episode intro
02:00 Self-taught ML; Unlikely paths to Anthropic
04:13 What are the most exciting things you’re working on?
07:48 Have we reached AGI — and what comes next?
10:10 How to measure AI progress
14:05 Is biology the next great frontier?
16:30 Career advice & the decade of the generalists
19:10 What could go wrong? What are the biggest AI risks?
22:50 Offense vs defense; how technology evolves
24:50 Answering the regulatory capture critique
30:15 Anthropic’s stance on open source
31:30 Would you slow down and let China go ahead?
32:50 Anthropic’s view on distillation
39:50 Is a post-scarcity world achievable?
48:30 How do we preserve values, tradition & community?
54:45 Predictions for 2028


This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit blog.joelonsdale.com

John Jovanovic is the President and Chairman of the Export-Import Bank of the United States. After years of mismanagement, Ex-Im is now profitable and has become a key driver of American reindustrialization. How did he turn the bank around? What's behind a new $12 billion critical mineral reserve to secure supply chains? And how is he seeing AI impact factory floors across America?The grandson of a Chicago steelworker, Jovanovic spent his career in energy, commodities, and supply chains — most recently as Chief Strategy Officer at Asplundh, the largest provider of critical infrastructure services across North America. He went on to serve in the first Trump administration at the Development Finance Corporation, where he opened the DFC's first overseas office to counter Chinese and Russian influence in the Balkans. In the fall of 2025, he was confirmed by the U.S. Senate to lead the Export-Import Bank.We begin with Jovanovic's turnaround of Ex-Im: he lays out his key reforms and explains why he spent his first hundred days in the building instead of traveling the country. Next, we dive into Project Vault, the $12 billion public-private critical mineral reserve he announced alongside President Trump to shore up American supply chains. Then, we explore his thesis behind the "revenge of the physical world," why AI is a shot in the arm for reindustrialization, and what Jovanovic is seeing on factory floors across the country. Finally, we discuss the backlash against data centers and lay out the case for AI optimism and how to ensure all Americans benefit from this massive opportunity.
(00:00) Episode intro
(02:00) If you can’t manufacture, you can't innovate
(09:00) Grandson of a steelworker; John’s background
(11:18) Countering Chinese & Russian influence in the Balkans
(15:55) Senate confirmation & how to build bipartisan support
(21:45) How John made Ex-Im profitable again
(28:10) Securing supply chains: $12B critical mineral reserve
(35:40) Superhuman strength: AI & reindustrialization
(38:08) Addressing data center backlash & AI safety
(41:57) Optimism for DC & America’s future


This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit blog.joelonsdale.com

Brian Chau is an investigative journalist using AI to uncover widespread corruption and billions in fraud. He recently revealed how the Effective Altruism movement bought and planted AI doomer articles, and he exposed the firm behind the frontier lab hack that's fueling AI hysteria. Now he's breaking his newest story: a $21 billion racial wealth transfer in healthcare payments across 35 states. How is he using AI to uncover stories that no other journalist could find? Why does he believe that AI is ultimately on the side of liberty and transparency? And can AI-powered accountability rebuild trust in media?
We discuss these timely issues and more with Brian, founder of Effort News. At 15 years old, Brian was the youngest person in North America to win an IOI gold medal in competitive programming. He went on to study at Waterloo University before leading Alliance for the Future, an AI policy organization in DC. There he encountered the broken state of media firsthand and decided to build Effort News.
We begin with Brian’s path from competitive programming to the “dirty news business,” and how he saw a generational opportunity to save journalism by applying AI to uncover massive fraud and corruption. Next, we break his latest news: an investigation into Health Equity Incentives — a $21 billion discriminatory transfer of Medicaid and Medicare dollars away from White and Asian people to Black and Hispanic. Then, we discuss his recent investigation into the UK counterterrorism program that flagged the writings of Douglas Murray and George Orwell as terrorist threats, and how Effective Altruism NGOs are funding AI doomer articles. Despite the splintered, tribal nature of journalism today, Brian believes that provable, AI-verified reporting will become the standard and rebuild confidence in our key institutions.
00:00 Episode intro
01:40 Youngest American to win IOI gold
04:30 AI journalism; uncovering what legacy media can't
09:30 Is Effort News nonpartisan? Answering the critiques
11:25 Breaking news: $21 Billion Racial Wealth Transfer
17:30 Exposing the UK's counterterror incompetence
22:00 How Effective Altruism & AI doomers bought the media
24:00 Anthropic "whistleblower" connection to left-wing NGOs
28:50 How do you account for bias in your reporting?
34:30 How to build a media business in 2026
42:30 AI for liberty and accountability; optimistic vision for the future



This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit blog.joelonsdale.com

Until recently, robots were constrained to narrow, specific tasks. Think assembly lines, medical devices, or household vacuum cleaners. But now all that is changing. One robot trained by Generalist's foundation model can do everything from folding clothes and washing dishes to sorting objects and repairing other robots. How are Pete Florence and the Generalist team bringing AI to the physical world? Why is intelligence, not hardware, the key unlock for robotics? And how could general-purpose robots unlock 10 times more scientific progress in the years ahead?Pete grew up in the Bay Area, earned his PhD at MIT, and became a star scientist at Google DeepMind. In 2023, he left to co-found Generalist, which builds robotics foundation models to help bring general intelligence to the physical world. Its latest model, GEN-1.5, can now train robots in a matter of seconds on simple tasks such as folding laundry and sorting objects. For this conversation, we're also joined by 8VC Partner Vivek Gopalan. We begin with Pete's entrepreneurial journey and his epiphany in grad school — walk into any robotics lab and the robots were standing still most hours of the day. Pete unpacks why robots, like humans, learn through data and experience, and why intelligence, not hardware, is the next frontier for robotics. At DeepMind he specialized in bringing multimodal learning (language models + vision models) into robotics, and he's now bringing that to the next level with robotic world models at Generalist. We dive into robotics' GPT-3 moment, the emergent capabilities Generalist is seeing — including self-taught ambidexterity — and the data moats that will shape the industry. Finally, we explore what a world of abundant robots means for science, reindustrialization, and American prosperity.
00:00 Episode intro
01:30 Bay Area to MIT and DeepMind
03:55 Pete's epiphany: robots sitting still
08:30 Why intelligence, not hardware, is the next frontier
13:20 The GPT-3 era of robotics
14:55 Why leave DeepMind?
16:50 How to train robot brains
18:30 Emergent ambidexterity
20:40 LLMs vs robot world models
23:50 GEN-0, GEN-1 & scaling laws in robotics
27:28 10x more science per year
32:50 Skilled trades, reindustrialization & future of robots


This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit blog.joelonsdale.com

Aaron Levie has spent two decades building Box into a cloud storage leader — and he's never been more excited for the future. What's next in the AI wave? Where is he already seeing evidence for new job creation? What do the doomers and accelerationists both get wrong? And why is Silicon Valley's biggest AI challenge self-inflicted?
We discuss these topics and more in this week's episode with Aaron. Originally a college project to access his files from anywhere, Aaron dropped out of USC in 2005 to build Box. He raised his first funding from Mark Cuban and pivoted to the enterprise. Today, Box is a multi-billion dollar company serving 120,000 global customers, and Aaron has become one of tech's leading voices on AI in the enterprise.
We begin our conversation with the state of model progress and the unprecedented speed of the AI era — companies going from zero to $500M in revenue in a year or two. We also discuss the Kimi K3 debate and China closing the gap on the U.S. in the AI race. Next, Aaron explains why diffusion, not intelligence, is the real rate limiter for AI adoption, and what the challenges will be at the enterprise level. Then we dive into the jobs debate: Aaron explains the 13 new types of jobs AI has created at Box, why law firms now need teams of engineers for their own custom AI models, and why mass job destruction means betting against markets themselves. Finally, we tackle Silicon Valley's self-inflicted AI crisis and why it may become the number one issue of the next presidential election.
(00:00) Episode intro
(01:15) AI wave / every two months is a year now
(06:30) China, Kimi K3 & open-source debate
(10:45) Diffusion is the real rate limiter
(15:50) Agent-first startups vs incumbents
(19:00) Nobody reads those contracts — AI’s new work
(23:35) Silicon Valley’s AI Problem / Larry David making AI models
(29:20) 13 new types of AI jobs at Box
(32:20) The origins of Box / dropping out of college
(37:00) Favorite angel investments right now
(40:20) Optimism for America’s future


This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit blog.joelonsdale.com